Inspired by the success of language models (LM), scaling up deep learning recommendation systems (DLRS) has become a recent trend in the community. All previous methods tend to scale up the model parameters during training time. However, how to efficiently utilize and scale up computational resources during test time remains underexplored, which can prove to be a scaling-efficient approach and bring orthogonal improvements in LM domains. The key point in applying test-time scaling to DLRS lies in effectively generating diverse yet meaningful outputs for the same instance. We propose two ways: One is to explore the heterogeneity of different model architectures. The other is to utilize the randomness of model initialization under a homogeneous architecture. The evaluation is conducted across eight models, including both classic and SOTA models, on three benchmarks. Sufficient evidence proves the effectiveness of both solutions. We further prove that under the same inference budget, test-time scaling can outperform parameter scaling. Our test-time scaling can also be seamlessly accelerated with the increase in parallel servers when deployed online, without affecting the inference time on the user side. Code is available here. https://github.com/aTitye/TTS4CTR.

Deep-research agents are capable of executing multi-step web exploration, targeted retrieval, and sophisticated question answering. Despite their powerful capabilities, deep-research agents face two critical bottlenecks: (1) the lack of large-scale, challenging datasets with real-world difficulty, and (2) the absence of accessible, open-source frameworks for data synthesis and agent training. To bridge these gaps, we first construct DeepResearch-9K, a large-scale challenging dataset specifically designed for deep-research scenarios built from open-source multi-hop question-answering (QA) datasets via a low-cost autonomous pipeline. Notably, it consists of (1) 9000 questions spanning three difficulty levels from L1 to L3 (2) high-quality search trajectories with reasoning chains from Tongyi-DeepResearch-30B-A3B, a state-of-the-art deep-research agent, and (3) verifiable answers. Furthermore, we develop an open-source training framework DeepResearch-R1 that supports (1) multi-turn web interactions, (2) different reinforcement learning (RL) approaches, and (3) different reward models such as rule-based outcome reward and LLM-as-Judge feedback. Finally, empirical results demonstrate that agents trained on DeepResearch-9K under our DeepResearch-R1 achieve state-of-the-art results on challenging deep-research benchmarks. We release the DeepResearch-9K dataset on https://huggingface.co/datasets/artillerywu/DeepResearch-9K and the code of DeepResearch-R1 on https://github.com/Applied-Machine-Learning-Lab/SIGIR2026_DeepResearch-R1.

Online platforms such as Amazon and Netflix serve users across multiple countries and regions, underscoring the importance of multi-market recommendation (MMR). Most MMR methods adopt a pre-training and fine-tuning paradigm, in which a unified model is first trained on centralized, global data and subsequently adapted to specific markets. However, this approach ignores the privacy of market data. While traditional federated learning preserves privacy, it typically aims to obtain a global model by aggregating model parameters and does not account for significant market heterogeneity. Additionally, because ID spaces are disjoint across markets, embedding-based aggregation strategies become ineffective. To overcome these challenges, we propose a federated collaborative signal quantization (FedMM) method for multi-market click-through rate (CTR) prediction. Our core idea leverages a discrete codebook mechanism to achieve privacy-preserving transmission and align disjoint ID spaces. We further employ a hierarchical codebook structure to capture cross-market shared patterns and market-specific characteristics. Specifically, we deploy a residual quantized variational autoencoder (RQ-VAE) with a dual-layer codebook mechanism for each market to quantize collaborative embeddings. The first layer utilizes a global federated codebook, updated via aggregation to capture universally shared collaborative patterns, while the second layer maintains a local codebook to learn market-specific semantics. Finally, the learned discrete codes, which integrate both general and specific collaborative signals, are incorporated into downstream CTR models to enhance prediction accuracy across all markets. Extensive experiments on benchmark datasets demonstrate that FedMM significantly improves recommendation performance with privacy guarantees.

Multi-scenario multi-task recommendation (MSMTR) systems must address recommendation demands across diverse scenarios while simultaneously optimizing multiple objectives, such as click-through rate and conversion rate. Existing MSMTR models typically consist of four information units: scenario-shared, scenario-specific, task-shared, and task-specific networks. These units interact to generate four types of relationship information flows, directed from scenario-shared or scenario-specific networks to task-shared or task-specific networks. However, these models face two main limitations: 1) They often rely on complex architectures, such as mixture-of-experts (MoE) networks, which increase the complexity of information fusion, model size, and training cost. 2) They extract all available information flows without filtering out irrelevant or even harmful content, introducing potential noise. Regarding these challenges, we propose a lightweight Automated Information Flow Selection (AutoIFS) framework for MSMTR. To tackle the first issue, AutoIFS incorporates low-rank adaptation (LoRA) to decouple the four information units, enabling more flexible and efficient information fusion with minimal parameter overhead. To address the second issue, AutoIFS introduces an information flow selection network that automatically filters out invalid scenario-task information flows based on model performance feedback. It employs a simple yet effective pruning function to eliminate useless information flows, thereby enhancing the impact of key relationships and improving model performance. Finally, we evaluate AutoIFS and confirm its effectiveness through extensive experiments on two public benchmark datasets and an online A/B test.

Large language models (LLMs) have been incorporated into numerous industrial applications. Meanwhile, a vast array of API assets is scattered across various functions in the financial domain. An online financial question-answering system can leverage both LLMs and private APIs to provide timely financial analysis and information. The key is equipping the LLM model with function calling capability tailored to a financial scenario. However, a generic LLM requires customized financial APIs to call and struggles to adapt to the financial domain. Additionally, online user queries are diverse and contain out-of-distribution parameters compared with the required function input parameters, which makes it more difficult for a generic LLM to serve online users. In this paper, we propose a data-driven pipeline to enhance function calling in LLM for our online, deployed financial QA, comprising dataset construction, data augmentation, and model training. Specifically, we construct a dataset based on a previous study and update it periodically, incorporating queries and an augmentation method named AugFC. The addition of user query-related samples will exploit our financial toolset in a data-driven manner, and AugFC explores the possible parameter values to enhance the diversity of our updated dataset. Then, we train an LLM with a two-step method, which enables the use of our financial functions. Extensive experiments on existing offline datasets, as well as the deployment of an online scenario, illustrate the superiority of our pipeline. The related pipeline has been adopted in the financial QA of YuanBao. https://yuanbao.tencent.com/chat/, one of the largest chat platforms in China.

Lifelong behavior modeling for single-domain has been widely investigated in industry click-through (CTR) prediction. However, some domains do not always have rich historical behaviors in online platforms, so cross-domain lifelong behavior modeling is overlooked. This paper proposes a novel retrieval augmented lifelong cross-domain net (RAL-CDNet) to address the challenges in cross-domain lifelong behavior modeling. There are three components in RAL-CDNet, i.e., cross-domain retrieval unit, cross-domain alignment unit, and cross-net. As the general search unit in the previous study, a cross-domain retrieval unit features a retrieval augmented paradigm that utilizes a pre-trained language model to learn the intrinsic textual information of user behaviors and generates the sequential behaviors from the source domain based on sequential behaviors in the target domain. The retrieval augmented behaviors can achieve consistency and capture accurate hidden interest for target domain CTR prediction. Furthermore, we propose the cross-domain alignment unit to align the embeddings across domains by adding a semantic-guided contrastive loss and auxiliary task loss in the source domain. This allows the embeddings to be consistent across domains and have enough source information to capture the cross-domain relation. Finally, the cross-net utilizes two-level attention techniques to enhance the final prediction in the target domain. We conduct extensive experiments on both a public dataset and an industrial dataset from the WeChat advertising platform to demonstrate the effectiveness of RAL-CDNet in terms of offline and online metrics.

Multi-scenario recommendation (MSR) is a popular training paradigm in industrial platforms for uniformly integrating information from multiple scenarios and serving them simultaneously. A key challenge in MSR research is accurately identifying the commonalities and distinctive information between scenarios. Currently, most existing MSR methods focus on implicitly extracting this information from the architectural level. However, this continues to increase the complexity and training overhead of MSR. Furthermore, the custom components responsible for extracting implicit information in each MSR method are too dependent on the specific MSR architecture and are not easily reused in other methods. Given these challenges, we first show in a motivating experiment that it may be beneficial to explicitly select a reasonable set of shared instances that can affect parameter optimization in all scenarios during the training of MSR, i.e., to explicitly obtain the critical information required for MSR from the data level. Then, this paper proposes SSIM with an adaptive selection network. Specifically, SSIM can be integrated with existing MSR methods in a lightweight way to adaptively select an informative and shareable subset of instances from each scenario to improve recommendations. In particular, the selected multi-scenario shared subset has extraordinary reusability and can be easily saved to benefit model training of various future MSR models. Finally, we evaluate SSIM and demonstrate its effectiveness through experiments on two public multi-scenario benchmarks and an online A/B test.

Fund allocation has been an increasingly important problem in the financial domain. In reality, we aim to allocate the funds to buy certain assets within a certain future period. Naive solutions such as prediction-only or Predict-then-Optimize approaches suffer from goal mismatch. Additionally, the introduction of the SOTA time series forecasting model inevitably introduces additional uncertainty in the predicted result. To solve both problems mentioned above, we introduce a Risk-aware Time-Series Predict-and-Allocate (RTS-PnO) framework, which holds no prior assumption on the forecasting models. Such a framework contains three features: (i) end-to-end training with objective alignment measurement(ii) adaptive forecasting uncertainty calibration, and (iii) agnostic towards forecasting models. The evaluation of RTS-PnO is conducted over both online and offline experiments. For offline experiments, eight datasets from three categories of financial applications are used: Currency, Stock, and Cryptos. RTS-PnO consistently outperforms other competitive baselines. The online experiment is conducted on the Cross-Border Payment business at FiT, Tencent, and an 8.4% decrease in regret is witnessed when compared with the product-line approach. The code for the offline experiment is available.

In recent years, the introduction of knowledge graphs (KGs) has significantly advanced recommender systems by facilitating the discovery of potential associations between items. However, existing methods still face several limitations. First, most KGs suffer from missing facts or limited scopes. Second, existing methods convert textual information in KGs into IDs, resulting in the loss of natural semantic connections between different items. Third, existing methods struggle to capture high-order connections in the global KG. To address these limitations, we propose a novel method called CoLaKG, which leverages large language models (LLMs) to improve KG-based recommendations. The extensive knowledge and remarkable reasoning capabilities of LLMs enable our method to supplement missing facts in KGs, and their powerful text understanding abilities allow for better utilization of semantic information. Specifically, CoLaKG extracts useful information from KGs at both local and global levels. By employing the item-centered subgraph extraction and prompt engineering, it can accurately understand the local information. In addition, through the semantic-based retrieval module, each item is enriched by related items from the entire knowledge graph, effectively harnessing global information. Furthermore, the local and global information are effectively integrated into the recommendation model through a representation fusion module and a retrieval-augmented representation learning module, respectively. Extensive experiments on four real-world datasets demonstrate the superiority of our method.

Multi-scenario recommendation (MSR) has become a core component of various online platforms, but its increasing model size has also brought attention to its efficiency optimization. An important effort is to find effective and efficient feature embedding layers for MSR, and existing work focuses on scenario-level feature selection, i.e., all instance embeddings in the same scenario get the same filtering results on the feature set, and the filtering results are different for different scenarios. However, this ignores the information redundancy of the dimension set and the individuality of different instances in the same scenario. To address these limitations, we propose a multi-scenario instance embedding learning (MultiEmb) framework that implements exclusive feature-dimension redundant information removal for different instances within a scenario to obtain the optimal individual embeddings. The core of our MultiEmb is to introduce an instance embedding selection network to effectively complete the above challenging tasks, in which a set of feature selection and dimension selection adaptive components are equipped for each scenario, and their combination completes the optimal embedding selection for each instance. Finally, we evaluate MultiEmb through extensive experiments on two public multi-scenario benchmarks and demonstrate its effectiveness, compatibility, transferability, etc.

The evolution of previous Click-Through Rate (CTR) models has mainly been driven by proposing complex components, whether shallow or deep, that are adept at modeling feature interactions. However, there has been less focus on improving fusion design. Instead, two naive solutions, stacked and parallel fusion, are commonly used. Both solutions rely on pre-determined fusion connections and fixed fusion operations. It has been repetitively observed that changes in fusion design may result in different performances, highlighting the critical role that fusion plays in CTR models. While there have been attempts to refine these basic fusion strategies, these efforts have often been constrained to specific settings or dependent on specific components. Neural architecture search has also been introduced to partially deal with fusion design, but it comes with limitations. The complexity of the search space can lead to inefficient and ineffective results. To bridge this gap, we introduce OptFusion, a method that automates the learning of fusion, encompassing both the connection learning and the operation selection. We have proposed a one-shot learning algorithm tackling these tasks concurrently. Our experiments are conducted over three large-scale datasets. Extensive experiments prove both the effectiveness and efficiency of OptFusion in improving CTR model performance. Our code implementation is available here https://github.com/kexin-kxzhang/OptFusion.

Customer Lifetime Value (CLTV) prediction is a critical task in business applications, such as customer relationship management (CRM), online marketing, etc. Accurately predicting CLTV is challenging in real-world business scenarios, as the distribution of CLTV is complex and mutable. Firstly, there is a large number of users without any consumption consisting of a long-tailed part that is too complex to fit. Secondly, the small set of high-value users spent orders of magnitude more than a typical user leading to a wide range of the CLTV distribution which is hard to capture in a single distribution. Existing approaches for CLTV estimation either assume a prior probability distribution and fit a single group of distribution-related parameters for all samples, or directly learn from the posterior distribution with manually predefined buckets in a heuristic manner. However, all these methods fail to handle complex and mutable distributions. In this paper, we propose a novel optimal distribution selection model (OptDist) for CLTV prediction, which utilizes an adaptive optimal sub-distribution selection mechanism to improve the accuracy of complex distribution modeling. Specifically, OptDist trains several candidate sub-distribution networks in the distribution learning module (DLM) for modeling the probability distribution of CLTV. Then, a distribution selection module (DSM) is proposed to select the sub-distribution for each sample, thus making the selection automatically and adaptively. Besides, we design an alignment mechanism that connects both modules, which effectively guides the optimization. We conduct extensive experiments on both two public and one private dataset to verify that OptDist outperforms state-of-the-art baselines. Furthermore, OptDist has been deployed on a large-scale financial platform for customer acquisition marketing campaigns and the online experiments also demonstrate the effectiveness of OptDist.

Uplift modeling has been widely employed in online marketing by predicting the response difference between the treatment and control groups, so as to identify the sensitive individuals toward interventions like coupons or discounts. Compared with traditional conversion uplift modeling,revenue uplift modeling exhibits higher potential due to its direct connection with the corporate income. However, previous works can hardly handle the continuous long-tail response distribution in revenue uplift modeling. Moreover, they have neglected to optimize the uplift ranking among different individuals, which is actually the core of uplift modeling. To address such issues, in this paper, we first utilize the zero-inflated lognormal (ZILN) loss to regress the responses and customize the corresponding modeling network, which can be adapted to different existing uplift models. Then, we study the ranking-related uplift modeling error from the theoretical perspective and propose two tighter error bounds as the additional loss terms to the conventional response regression loss. Finally, we directly model the uplift ranking error for the entire population with a listwise uplift ranking loss. The experiment results on offline public and industrial datasets validate the effectiveness of our method for revenue uplift modeling. Furthermore, we conduct large-scale experiments on a prominent online fintech marketing platform, Tencent FiT, which further demonstrates the superiority of our method in real-world applications.

As user behaviors become complicated on business platforms, online recommendations focus more on how to touch the core conversions, which are highly related to the interests of platforms. These core conversions are usually continuous targets, such as watch time, revenue, and so on, whose predictions can be enhanced by previous discrete conversion actions. Therefore, multi-task learning (MTL) can be adopted as the paradigm to learn these hybrid targets. However, existing works mainly emphasize investigating the sequential dependence among discrete conversion actions, which neglects the complexity of dependence between discrete conversions and the final continuous conversion. Moreover, simultaneously optimizing hybrid tasks with stronger task dependence will suffer from volatile issues where the core regression task might have a larger influence on other tasks. In this paper, we study the MTL problem with hybrid targets for the first time and propose the model named Hybrid Targets Learning Network (HTLNet) to explore task dependence and enhance optimization. Specifically, we introduce label embedding for each task to explicitly transfer the label information among these tasks, which can effectively explore logical task dependence. We also further design the gradient adjustment regime between the final regression task and other classification tasks to enhance the optimization. Extensive experiments on two offline public datasets and one real-world industrial dataset are conducted to validate the effectiveness of HTLNet. Moreover, online A/B tests on the financial recommender system also show that our model has improved significantly. Our implementation is available here1.

In modern online platforms, incentives (e.g., discounts, bonus) are essential factors that enhance user engagement and increase platform revenue. Over recent years, uplift modeling has been introduced as a strategic approach to assign incentives to individual customers. Especially in many real-world applications, online platforms can only incentivize customers with specific budget constraints. This problem can be reformulated as the multi-choice knapsack problem (MCKP). The objective of this optimization is to select the optimal incentive for each customer to maximize the return on investment (ROI). Recent works in this field frequently tackle the budget allocation problem using a two-stage approach. However, this solution is confronted with the following challenges: (1) The causal inference methods often ignore the domain knowledge in online marketing, where the expected response curve of a customer should be monotonic and smooth as the incentive increases. (2) There is an optimality gap between the two stages, resulting in inferior sub-optimal allocation performance due to the loss of the incentive recommendation information for the uplift prediction under the limited budget constraint. To address these challenges, we propose a novel End-to-End Cost-Effective Incentive Recommendation (E3IR) model under the budget constraint. Specifically, our methods consist of two modules, i.e., the uplift prediction module and the differentiable allocation module. In the uplift prediction module, we construct prediction heads to capture the incremental improvement between adjacent treatments with the marketing domain constraints (i.e., monotonic and smooth). We incorporate integer linear programming (ILP) as a differentiable layer input in the differentiable allocation module. Furthermore, we conduct extensive experiments on public and real product datasets, demonstrating that our E3IR improves allocation performance compared to existing two-stage approaches.

Estimating the individual treatment effect (ITE) from observational data is a crucial research topic that holds significant value across multiple domains. How to identify hidden confounders poses a key challenge in ITE estimation. Recent studies have incorporated the structural information of social networks to tackle this challenge, achieving notable advancements. However, these methods utilize graph neural networks to learn the representation of hidden confounders in Euclidean space, disregarding two critical issues: (1) the social networks often exhibit a scale-free structure, while Euclidean embeddings suffer from high distortion when used to embed such graphs, and (2) each ego-centric network within a social network manifests a treatment-related characteristic, implying significant patterns of hidden confounders. To address these issues, we propose a novel method called Treatment-Aware Hyperbolic Representation Learning (TAHyper). Firstly, TAHy-per employs the hyperbolic space to encode the social networks, thereby effectively reducing the distortion of confounder representation caused by Euclidean embeddings. Secondly, we design a treatment-aware relationship identification module that enhances the representation of hidden confounders by identifying whether an individual and her neighbors receive the same treatment. Extensive experiments on two benchmark datasets are conducted to demonstrate the superiority of our method. The code is available at https://github.com/ziqiangcui/TAHyper.

Multi-behavior recommender systems (MBRS) have been commonly deployed on real-world industrial platforms for their superior advantages in understanding user preferences and mitigating data sparsity. However, the cascade graph modeling paradigm adopted in mainstream MBRS usually assumes that users will refer to all types of behavioral knowledge they have when making decisions about target behaviors, i.e., use all types of behavioral interactions indiscriminately when modeling and predicting target behaviors for each user. We call this a full decision chain constraint and argue that it may be too strict by ignoring that different types of behavioral knowledge have varying importance for different users. In this paper, we propose a novel automated decision chain selection (AutoDCS) framework to relax this constraint, which can consider each user's unique decision dependencies and select a reasonable set of behavioral knowledge to activate for the prediction of target behavior. Specifically, AutoDCS first integrates some existing MBRS methods in a base cascade module to obtain a set of behavior-aware embeddings. Then, a bilateral matching gating mechanism is used to select an exclusive set of behaviors for the current user-item pair to form a decision chain, and the corresponding behavior-augmented embeddings are selectively activated. Subsequently, AutoDCS combines the behavior-augmented and original behavior-aware embeddings to predict the target behavior. Finally, we evaluate AutoDCS and demonstrate its effectiveness through experiments over four public multi-behavior benchmarks.

Multi-scenario recommender systems (MSRSs) have been increasingly used in real-world industrial platforms for their excellent advantages in mitigating data sparsity and reducing maintenance costs. However, conventional MSRSs usually use all relevant features indiscriminately and ignore that different kinds of features have varying importance under different scenarios, which may cause confusion and performance degradation. In addition, existing feature selection methods for deep recommender systems may lack the exploration of scenario relations. In this paper, we propose a novel automated multi-scenario feature selection (MultiFS) framework to bridge this gap, which is able to consider scenario relations and utilize a hierarchical gating mechanism to select features for each scenario. Specifically, MultiFS first efficiently obtains feature importance across all the scenarios through a scenario-shared gate. Then, some scenario-specific gate aims to identify feature importance to individual scenarios from a subset of the former with lower importance. Subsequently, MultiFS imposes constraints on the two gates to make the learning mechanism more feasible and combines the two to select exclusive features for different scenarios. We evaluate MultiFS and demonstrate its ability to enhance the multi-scenario model performance through experiments over two public multi-scenario benchmarks.

As a key component in online marketing, uplift modeling aims to accurately capture the degree to which different treatments motivate different users, such as coupons or discounts, also known as the estimation of individual treatment effect (ITE). In an actual business scenario, the options for treatment may be numerous and complex, and there may be correlations between different treatments. In addition, each marketing instance may also have rich user and contextual features. However, existing methods still fall short in both fully exploiting treatment information and mining features that are sensitive to a particular treatment. In this paper, we propose an explicit feature interaction-aware uplift network (EFIN) to address these two problems. Our EFIN includes four customized modules: 1) a feature encoding module encodes not only the user and contextual features, but also the treatment features; 2) a self-interaction module aims to accurately model the user's natural response with all but the treatment features; 3) a treatment-aware interaction module accurately models the degree to which a particular treatment motivates a user through interactions between the treatment features and other features, i.e., ITE; and 4) an intervention constraint module is used to balance the ITE distribution of users between the control and treatment groups so that the model would still achieve a accurate uplift ranking on data collected from a non-random intervention marketing scenario. We conduct extensive experiments on two public datasets and one product dataset to verify the effectiveness of our EFIN. In addition, our EFIN has been deployed in a credit card bill payment scenario of a large online financial platform with a significant improvement.

Multi-task learning for various real-world applications usually involves tasks with logical sequential dependence. For example, in online marketing, the cascade behavior pattern of impression \rightarrow click \rightarrow conversion is usually modeled as multiple tasks in a multi-task manner, where the sequential dependence between tasks is simply connected with an explicitly defined function or implicitly transferred information in current works. These methods alleviate the data sparsity problem for long-path sequential tasks as the positive feedback becomes sparser along with the task sequence. However, the error accumulation and negative transfer will be a severe problem for downstream tasks. Especially, at the beginning stage of training, the optimization for parameters of former tasks is not converged yet, and thus the information transferred to downstream tasks is negative. In this paper, we propose a prior information merged model (PIMM), which explicitly models the logical dependence among tasks with a novel prior information merged (PIM) module for multiple sequential dependence task learning in a curriculum manner. Specifically, the PIM randomly selects the true label information or the prior task prediction with a soft sampling strategy to transfer to the downstream task during the training. Following an easy-to-difficult curriculum paradigm, we dynamically adjust the sampling probability to ensure that the downstream task will get the effective information along with the training. The offline experimental results on both public and product datasets verify that PIMM outperforms state-of-the-art baselines. Moreover, we deploy the PIMM in a large-scale FinTech platform, and the online experiments also demonstrate the effectiveness of PIMM.

Tabular data is one of the most common data storage formats behind many real-world web applications such as retail, banking, and e-commerce. The success of these web applications largely depends on the ability of the employed machine learning model to accurately distinguish influential features from all the predetermined features in tabular data. Intuitively, in practical business scenarios, different instances should correspond to different sets of influential features, and the set of influential features of the same instance may vary in different scenarios. However, most existing methods focus on global feature selection assuming that all instances have the same set of influential features, and few methods considering instance-wise feature selection ignore the variability of influential features in different scenarios. In this paper, we first introduce a new perspective based on the influence function for instance-wise feature selection, and give some corresponding theoretical insights, the core of which is to use the influence function as an indicator to measure the importance of an instance-wise feature. We then propose a new solution for discovering instance-wise influential features in tabular data (DIWIFT), where a self-attention network is used as a feature selection model and the value of the corresponding influence function is used as an optimization objective to guide the model. Benefiting from the advantage of the influence function, i.e., its computation does not depend on a specific architecture and can also take into account the data distribution in different scenarios, our DIWIFT has better flexibility and robustness. Finally, we conduct extensive experiments on both synthetic and real-world datasets to validate the effectiveness of our DIWIFT.

Click-through prediction (CTR) models transform features into latent vectors and enumerate possible feature interactions to improve performance based on the input feature set. Therefore, when selecting an optimal feature set, we should consider the influence of both features and their interaction. However, most previous works focus on either feature field selection or only select feature interaction based on the fixed feature set to produce the feature set. The former restricts search space to the feature field, which is too coarse to determine subtle features. They also do not filter useless feature interactions, leading to higher computation costs and degraded model performance. The latter identifies useful feature interaction from all available features, resulting in many redundant features in the feature set. In this paper, we propose a novel method named OptFS to address these problems. To unify the selection of features and their interaction, we decompose the selection of each feature interaction into the selection of two correlated features. Such a decomposition makes the model end-to-end trainable given various feature interaction operations. By adopting feature-level search space, we set a learnable gate to determine whether each feature should be within the feature set. Because of the large-scale search space, we develop a learning-by-continuation training scheme to learn such gates. Hence, OptFS generates the feature set containing features that improve the final prediction results. Experimentally, we evaluate OptFS on three public datasets, demonstrating OptFS can optimize feature sets which enhance the model performance and further reduce both the storage and computational cost.

Although reinforcement learning (RL) techniques are regarded as promising solutions for interactive recommender systems (IRS), such solutions still face three main challenges, namely, i) time inefficiency when handling large discrete action space in IRS, ii) inability to deal with the cold-start scenarios in IRS, iii) data inefficiency during training the RL-based methods. To tackle these challenges, we propose a generic tree-structured RL framework taking both policy-based and value-based approaches into consideration. We propose to construct a balanced tree over representations of the items, such that picking an item is formulated as seeking a suitable path from the root to a leaf node in the balanced tree, which dramatically reduces the time complexity of item recommendation. Further, for cold-start scenarios where prior information of the items is unavailable, we initialize a random balanced tree as the starting point and then refine the tree structure based on the learned item representations. Besides, we also incorporate a user modeling component to explicitly model the environment, which can be utilized in the training phase to improve data efficiency. Extensive experiments on two real-world datasets are conducted and demonstrate that our framework can achieve superior recommendation performance and provide time and data efficiency improvement over state-of-the-art methods in both warm-start and cold-start IRS scenarios.

The bias problems in recommender systems are an important challenge. In this paper, we focus on solving the bias problems via uniform data. Previous works have shown that simple modeling with a uniform data can alleviate the bias problems and improve the performance. However, the uniform data is usually few and expensive to collect in a real product. In order to use the valuable uniform data more effectively, we propose a novel and general knowledge distillation framework for counterfactual recommendation with four specific methods, including label-based distillation, feature-based distillation, sample-based distillation and model structure-based distillation. Moreover, we discuss the relation between the proposed framework and the previous works. We then conduct extensive experiments on both public and product datasets to verify the effectiveness of the proposed four methods. In addition, we explore and analyze the performance trends of the proposed methods on some key factors, and the changes in the distribution of the recommendation lists. Finally, we emphasize that counterfactual modeling with uniform data is a rich research area, and list some interesting and promising research topics worthy of further exploration. Note that the source codes are available athttps://github.com/dgliu/TKDE_KDCRec.

Feature embedding learning and feature interaction modeling are two crucial components of deep models for Click-Through Rate (CTR) prediction in recommender systems. Most existing deep CTR models suffer from the following three problems. First, feature interactions are either manually designed or simply enumerated. However, not all the feature interactions are useful for the prediction task and useless feature interactions may introduce noisy signals thus causing overfitting. Second, all the feature interactions are modeled with an identical interaction function, whereas different interaction functions introduce different inductive biases to better capture various feature interaction patterns. Third, in most existing models, different features share the same embedding size. However, model size can be further optimized without sacrificing performance by differentiating embedding sizes for individual features, as the amount of information contained in each feature varies much. To address the three issues mentioned above, we proposeAutomatic Interaction Machine (AIM)with three core components, namely, Feature Interaction Search (FIS), Interaction Function Search (IFS) and Embedding Dimension Search (EDS), respectively. To tackle the first problem, FIS component automatically identifies different orders of essential feature interactions with useless ones pruned. Taking care of the second problem, IFS component selects appropriate interaction functions for each individual feature interaction in a learnable way. Moreover, to avoid learning conflict among different interaction functions, IFS proposes function-wise embeddings via performing multiple embeddings for each feature, where each feature embedding corresponds to one possible interaction function. However, utilizing multiple embeddings for each feature may make the model size affordably large if we keep the same embedding size as utilizing shared embedding (i.e., each feature shares the same embedding for different interaction functions). To solve this third problem, EDS automatically selects proper embedding size for each feature. Such a flexible embedding size adaptation is able to reduce the large amount of embedding parameters introduced by function-wise embeddings. Offline experiments on three large-scale datasets (two public benchmarks, one private dataset) validate that AIM can significantly improve various FM-based models. AIM has been deployed in the recommendation service of a mainstream app market, where a three-week online A/B test demonstrated the superiority of AIM, improving DeepFM model by 4.4% in terms of CTR.

Debiased recommendation with a randomized dataset has shown very promising results in mitigating system-induced biases. However, it still lacks more theoretical insights or an ideal optimization objective function compared with the other more well-studied routes without a randomized dataset. To bridge this gap, we study the debiasing problem from a new perspective and propose to directly minimize the upper bound of an ideal objective function, which facilitates a better potential solution to system-induced biases. First, we formulate a new ideal optimization objective function with a randomized dataset. Second, according to the prior constraints that an adopted loss function may satisfy, we derive two different upper bounds of the objective function: a generalization error bound with triangle inequality and a generalization error bound with separability. Third, we show that most existing related methods can be regarded as the insufficient optimization of these two upper bounds. Fourth, we propose a novel method called debiasing approximate upper bound ( DUB ) with a randomized dataset, which achieves a more sufficient optimization of these upper bounds. Finally, we conduct extensive experiments on a public dataset and a real product dataset to verify the effectiveness of our DUB.

Click-through rate (CTR) estimation plays a crucial role in modern online personalization services. It is essential to capture users’ drifting interests by modeling sequential user behaviors to build an accurate CTR estimation model. However, as the users accumulate a large amount of behavioral data on the online platforms, the current CTR models have to truncate user behavior sequences and utilize the most recent behaviors, which leads to a problem that sequential patterns such as periodicity or long-term dependency are not contained in the recent behaviors but in far back history. However, it is non-trivial to model the entire user sequence by directly using it for two reasons. Firstly, the very long input sequences will make online inference time and system load infeasible. Secondly, the very long sequences contain much noise, thus making it difficult for CTR models to capture useful patterns effectively. To tackle this issue, we consider it from the input data perspective instead of designing more sophisticated yet complex models. As the entire user behavior sequence contains much noise, it is unnecessary to input the entire sequence. Instead, we could just retrieve only a small part of it as the input to the CTR model. In this article, we propose the U ser B ehavior R etrieval (UBR) framework which aims at learning to retrieve the most informative user behaviors according to each CTR estimation request. Retrieving only a small set of behaviors could alleviate the two problems of utilizing very long sequences (i.e., inference efficiency and noisy input). The distinguishing property of UBR is that it supports arbitrary and learnable retrieval functions instead of utilizing a fixed pre-defined function, which is different from the current retrieval-based methods. Offline evaluations on three large-scale real-world datasets demonstrate the superiority and efficacy of the UBR framework. We further deploy UBR at the Huawei App Store, where it achieves 6.6% of eCPM gain in the online A/B test and now serves the main traffic in the Huawei App Store advertising scenario.

Debiased recommendation has recently attracted increasing attention from both industry and academic communities. Traditional models mostly rely on the inverse propensity score (IPS), which can be hard to estimate and may suffer from the high variance issue. To alleviate these problems, in this article, we propose a novel debiased recommendation framework based on user feature balancing. The general idea is to introduce a projection function to adjust user feature distributions, such that the ideal unbiased learning objective can be upper bounded by a solvable objective purely based on the offline dataset. In the upper bound, the projected user distributions are expected to be equal given different items. From the causal inference perspective, this requirement aims to remove the causal relation from the user to the item, which enables us to achieve unbiased recommendation, bypassing the computation of IPS. To efficiently balance the user distributions upon each item pair, we propose three strategies, including clipping, sampling, and adversarial learning to improve the training process. For more robust optimization, we deploy an explicit model to capture the potential latent confounders in recommendation systems. To the best of our knowledge, this article is the first work on debiased recommendation based on confounder balancing. In the experiments, we compare our framework with many state-of-the-art methods based on synthetic, semi-synthetic, and real-world datasets. Extensive experiments demonstrate that our model is effective in promoting the recommendation performance.

How to effectively mitigate the bias of feedback in recommender systems is an important research topic. In this article, we first describe the generation process of the biased and unbiased feedback in recommender systems via two respective causal diagrams, where the difference between them can be regarded as the source of system-induced biases. We then define this difference as a confounding bias and propose a new perspective on debiased representation learning to alleviate it. Specifically, for the case with biased feedback alone, we derive the conditions that need to be satisfied to obtain a debiased representation from the causal diagrams. Then, we propose a novel framework called debiased information bottleneck (DIB) to optimize these conditions and then find a tractable solution for it. The proposed framework constrains the model to learn a biased embedding vector with independent biased and unbiased components in the training phase, and uses only the unbiased component in the test phase to deliver more accurate recommendations. We further propose a variant of DIB by relaxing the independence between the biased and unbiased components. Finally, we conduct extensive experiments on a public dataset and a real product dataset to verify the effectiveness of the proposed framework.

CTR prediction is essential for modern recommender systems. Ranging from early factorization machines to deep learning based models in recent years, existing CTR methods focus on capturing useful feature interactions or mining important behavior patterns. Despite the effectiveness, we argue that these methods suffer from the risk of label sparsity (i.e., the user-item interactions are highly sparse with respect to the feature space), label noise (i.e., the collected user-item interactions are usually noisy), and the underuse of domain knowledge (i.e., the pairwise correlations between samples). To address these challenging problems, we propose a novel Multi-Interest Self-Supervised learning (MISS) framework which enhances the feature embeddings with interest-level self-supervision signals. With the help of two novel CNN-based multi-interest extractors, self-supervision signals are discovered with full considerations of different interest representations (point-wise and union-wise), interest dependencies (short-range and long-range), and interest correlations (inter-item and intra-item). Based on that, contrastive learning losses are further applied to the augmented views of interest representations, which effectively improves the feature representation learning. Furthermore, our proposed MISS frame-work can be used as an “plug-in” component with existing CTR prediction models and further boost their performances. Extensive experiments on three large-scale datasets show that MISS significantly outperforms the state-of-the-art models, by up to 13.55% in AUC, and also enjoys good compatibility with representative deep CTR models.

Click-through rate prediction is one of the core tasks in commercial recommender systems. It aims to predict the prob-ability of a user clicking a particular item given user and item features. As feature interactions bring in non-linearity, they are widely adopted to improve the performance of CTR prediction models. Therefore, effectively modelling feature interactions has attracted much attention in both the research and industry field. The current approaches can generally be categorized into three classes: (i) naïve methods, which do not model feature interactions and only use original features; (ii) memorized methods, which memorize feature interactions by explicitly viewing them as new features and assigning trainable embeddings; (iii) factorized meth-ods, which learn latent vectors for original features and implicitly model feature interactions through factorization functions. Studies have shown that modelling feature interactions by one of these methods alone are suboptimal due to the unique characteristics of different feature interactions. To address this issue, we first propose a general framework called OptInter which finds the most suitable modelling method for each feature interaction. Different state-of-the-art deep CTR models can be viewed as instances of OptInter. To realize the functionality of OptInter, we also introduce a learning algorithm that automatically searches for the optimal modelling method. We conduct extensive experiments on four large datasets, including three public and one private. Experimental results demonstrate the effectiveness of OptInter. Because our OptInter finds the optimal modelling method for each feature interaction, our experiments show that OptInter improves the best performed state-of-the-art baseline deep CTR models by up to 2.21%. Compared to the memorized method, which also outperforms baselines, we reduce up to 91% parameters. In addition, we conduct several ablation studies to investigate the influence of different components of OptInter. Finally, we provide interpretable discussions on the results of OptInter.

Deep learning-based recommendation has become a widely adopted technique in various online applications. Typically, a deployed model undergoes frequent re-training to capture users' dynamic behaviors from newly collected interaction logs. However, the current model training process only acquires users' feedbacks as labels, but fails to take into account the errors made in previous recommendations. Inspired by the intuition that humans usually reflect and learn from mistakes, in this paper, we attempt to build a self-correction continual learning loop (dubbed ReLoop) for recommender systems. In particular, a new customized loss is employed to encourage every new model version to reduce prediction errors over the previous model version during training. Our ReLoop learning framework enables a continual self-correction process in the long run and thus is expected to obtain better performance over existing training strategies. Both offline experiments and an online A/B test have been conducted to validate the effectiveness of ReLoop.

This paper presents a neural information retrieval pipeline that integrates cooperative learning of query reformulation and neural retrieval models. Our pipeline first exploits an automatic query reformulator to reformulate the user-issued query and then submits the reformulated query to the neural retrieval model. We simultaneously optimize the quality of reformulated queries and ranking performance with an alternate training strategy where query reformulator and neural retrieval model learn from the feedback of each other. Besides, we incorporate knowledge information into automatic query reformulation. The reformulated queries are further improved and contribute to a better ranking performance of the following neural retrieval model. We study two representative neural retrieval models KNRM and BERT in our pipeline. Experiments on two datasets show that our pipeline consistently improves the retrieval performance of the original neural retrieval models while only increases negligible time on automatic query reformulation.

Knowledge tracing, which estimates students' knowledge states by predicting the probability that they correctly answer questions, is an essential task for online learning platforms. It has gained much attention in the decades due to its importance to downstream tasks like learning material arrangement, etc. The previous deep learning-based methods trace students' knowledge states with the explicitly intra-student information, i.e., they only consider the historical information of individuals to make predictions. However, they neglect the inter-student information, which contains the response correctness of other students who have similar question-answering experiences, may offer some valuable clues. Based on this consideration, we propose a method called Collaborative Knowledge Tracing (CoKT) in this paper, which sufficiently exploits the inter-student information in knowledge tracing. It retrieves the sequences of peer students who have similar question-answering experiences to obtain the inter-student information, and integrates the inter-student information with the intra-student information to trace students' knowledge states and predict their correctness in answering questions. We validate the effectiveness of our method on four real-world datasets and compare it with 11 baselines. The experimental results reveal that CoKT achieves the best performance.

Dynamic Treatment Recommendation (DTR) is a sequence of tailored treatment decision rules which can be grouped as individual sub-tasks. As the reward signals in DTR are hard to design, Imitation Learning (IL) has achieved great success as it is effective in mimicking doctors' behaviors from their demonstrations without explicit reward signals. As a patient may have several different symptoms, the behaviors in doctors' demonstrations can often be grouped to handle individual symptoms. However, a single flat policy learned by IL is difficult to mimic doctors' demonstrations with such hierarchical structure, where low-level behaviors are switching from one symptom to another controlled by high-level decisions. Due to this observation, we consider Hierarchical Imitation Learning methods as good solutions for DTR. In this paper, we propose a novel Subgoal conditioned HIL framework (short for SHIL), where a high-level policy sequentially sets a subgoal for each sub-task without prior knowledge, and the low-level policy for sub-tasks is learned to reach the subgoal. To get rid of prior knowledge, a self-supervised learning method is proposed to learn an effective representation for each subgoal. More specifically, we carefully designed to encourage diverse representations among different subgoals. To demonstrate that SHIL is able to learn meaningful high-level policy and low-level policy that accurately reproduces complex doctors' demonstrations, we conduct experiments on a real-world medical data from health care domain, MIMIC-III. Compared with state-of-the-art baselines, SHIL improves the likelihood of patient survival by a significant margin and provides explainable recommendation with hierarchical structure.

Self-supervised learning, especially contrastive learning, has made an outstanding contribution to the development of many deep learning research fields. Recently, researchers in the acoustic signal processing field noticed its success and leveraged contrastive learning for better music representation. Typically, existing approaches maximize the similarity between two distorted audio segments sampled from the same music. In other words, they ensure a semantic agreement at the music level. However, those coarse-grained methods neglect some inessential or noisy elements at the frame level, which may be detrimental to the model to learn the effective representation of music. Towards this end, this paper proposes a novel Positive-nEgative frame mask for Music Representation based on the contrastive learning framework, abbreviated as PEMR. Concretely, PEMR incorporates a Positive-Negative Mask Generation module, which leverages transformer blocks to generate frame masks on Log-Mel spectrogram. We can generate self-augmented negative and positive samples by masking important components or inessential components, respectively. We devise a novel contrastive learning objective to accommodate both self-augmented positives/negatives sampled from the same music. We conduct experiments on four public datasets. The experimental results of two music-related downstream tasks, music classification and cover song identification, demonstrate the generalization ability and transferability of music representation learned by PEMR.

Learning to rank from logged user feedback, such as clicks or purchases, is a central component of many real-world information systems. Different from human-annotated relevance labels, the user feedback is always noisy and biased. Many existing learning to rank methods infer the underlying relevance of query–item pairs based on different assumptions of examination, and still optimize a relevance based objective. Such methods rely heavily on the correct estimation of examination, which is often difficult to achieve in practice. In this work, we propose a general framework U-rank+ for learning to rank with logged user feedback from the perspective of graph matching. We systematically analyze the biases in user feedback, including examination bias and selection bias. Then, we take both biases into consideration for unbiased utility estimation that directly based on user feedback, instead of relevance. In order to maximize the estimated utility in an efficient manner, we design two different solvers based on Sinkhorn and LambdaLoss for U-rank+ . The former is based on a standard graph matching algorithm, and the latter is inspired by the traditional method of learning to rank. Both of the algorithms have good theoretical properties to optimize the unbiased utility objective while the latter is proved to be empirically more effective and efficient in practice. Our framework U-rank+ can deal with a general utility function and can be used in a widespread of applications including web search, recommendation, and online advertising. Semi-synthetic experiments on three benchmark learning to rank datasets demonstrate the effectiveness of U-rank+ . Furthermore, our proposed framework has been deployed on two different scenarios of a mainstream App store, where the online A/B testing shows that U-rank+ achieves an average improvement of 19.2% on click-through rate and 20.8% improvement on conversion rate in recommendation scenario, and 5.12% on platform revenue in online advertising scenario over the production baselines.

Effectively modeling feature interactions is crucial for CTR prediction in industrial recommender systems. The state-of-the-art deep CTR models with parallel structure (e.g., DCN) learn explicit and implicit feature interactions through independent parallel networks. However, these models suffer from trivial sharing issues, namely insufficient sharing in hidden layers and excessive sharing in network input, limiting the model's expressiveness and effectiveness. Therefore, to enhance information sharing between explicit and implicit feature interactions, we propose a novel deep CTR model EDCN. EDCN introduces two advanced modules, namely bridge module and regulation module, which work collaboratively to capture the layer-wise interactive signals and learn discriminative feature distributions for each hidden layer of the parallel networks. Furthermore, two modules are lightweight and model-agnostic, which can be generalized well to mainstream parallel deep CTR models. Extensive experiments and studies are conducted to demonstrate the effectiveness of EDCN on two public datasets and one industrial dataset. Moreover, the compatibility of two modules over various parallel-structured models is verified, and they have been deployed onto the online advertising platform in Huawei, where a one-month A/B test demonstrates the improvement over the base parallel-structured model by 7.30% and 4.85% in terms of CTR and eCPM, respectively.

Collaborative filtering (CF) is a widely studied research topic in recommender systems. The learning of a CF model generally depends on three major components, namely interaction encoder, loss function, and negative sampling. While many existing studies focus on the design of more powerful interaction encoders, the impacts of loss functions and negative sampling ratios have not yet been well explored. In this work, we show that the choice of loss function as well as negative sampling ratio is equivalently important. More specifically, we propose the cosine contrastive loss (CCL) and further incorporate it to a simple unified CF model, dubbed SimpleX. Extensive experiments have been conducted on 10 benchmark datasets and compared with 28 existing CF models in total. Surprisingly, the results show that, under our CCL loss and a large negative sampling ratio, SimpleX can surpass most sophisticated state-of-the-art models by a large margin (e.g., max 48.5% improvement in [email protected] over LightGCN). We believe that SimpleX could not only serve as a simple strong baseline to foster future research on CF, but also shed light on the potential research direction towards improving loss function and negative sampling.

With the recent success of graph convolutional networks (GCNs), they have been widely applied for recommendation, and achieved impressive performance gains. The core of GCNs lies in its message passing mechanism to aggregate neighborhood information. However, we observed that message passing largely slows down the convergence of GCNs during training, especially for large-scale recommender systems, which hinders their wide adoption. LightGCN makes an early attempt to simplify GCNs for collaborative filtering by omitting feature transformations and nonlinear activations. In this paper, we take one step further to propose an ultra-simplified formulation of GCNs (dubbed UltraGCN), which skips infinite layers of message passing for efficient recommendation. Instead of explicit message passing, UltraGCN resorts to directly approximate the limit of infinite-layer graph convolutions via a constraint loss. Meanwhile, UltraGCN allows for more appropriate edge weight assignments and flexible adjustment of the relative importances among different types of relationships. This finally yields a simple yet effective UltraGCN model, which is easy to implement and efficient to train. Experimental results on four benchmark datasets show that UltraGCN not only outperforms the state-of-the-art GCN models but also achieves more than 10x speedup over LightGCN.

Top-N recommendation, which aims to learn user ranking-based preference, has long been a fundamental problem in a wide range of applications. Traditional models usually motivate themselves by designing complex or tailored architectures based on different assumptions. However, the training data of recommender system can be extremely sparse and imbalanced, which poses great challenges for boosting the recommendation performance. To alleviate this problem, in this paper, we propose to reformulate the recommendation task within the causal inference framework, which enables us to counterfactually simulate user ranking-based preferences to handle the data scarce problem. The core of our model lies in the counterfactual question: "what would be the user's decision if the recommended items had been different?''. To answer this question, we firstly formulate the recommendation process with a series of structural equation models (SEMs), whose parameters are optimized based on the observed data. Then, we actively indicate many recommendation lists (called intervention in the causal inference terminology) which are not recorded in the dataset, and simulate user feedback according to the learned SEMs for generating new training samples. Instead of randomly intervening on the recommendation list, we design a learning-based method to discover more informative training samples. Considering that the learned SEMs can be not perfect, we, at last, theoretically analyze the relation between the number of generated samples and the model prediction error, based on which a heuristic method is designed to control the negative effect brought by the prediction error. Extensive experiments are conducted based on both synthetic and real-world datasets to demonstrate the effectiveness of our framework.

Click-through rate (CTR) prediction is a critical task for many applications, as its accuracy has a direct impact on user experience and platform revenue. In recent years, CTR prediction has been widely studied in both academia and industry, resulting in a wide variety of CTR prediction models. Unfortunately, there is still a lack of standardized benchmarks and uniform evaluation protocols for CTR prediction research. This leads to non-reproducible or even inconsistent experimental results among existing studies, which largely limit the practical value and potential impact of their research. In this work, we aim to perform open benchmarking for CTR prediction and present a rigorous comparison of different models in a reproducible manner. To this end, we ran over 7,000 experiments for more than 12,000 GPU hours in total to re-evaluate 24 existing models on multiple dataset settings. Surprisingly, our experiments show that with sufficient hyper-parameter search and model tuning, many deep models have smaller differences than expected. The results also reveal that making real progress on the modeling of CTR prediction is indeed a very challenging research task. We believe that our benchmarking work could not only allow researchers to gauge the effectiveness of new models conveniently but also make them fairly compare with the state of the arts. We have publicly released the benchmarking tools, evaluation protocols, and experimental settings of our work to promote reproducible research in this field.

Click-Through Rate (CTR) prediction is critical for industrial recommender systems, where most deep CTR models follow an Embedding & Feature Interaction paradigm. However, the majority of methods focus on designing network architectures to better capture feature interactions while the feature embedding, especially for numerical features, has been overlooked. Existing approaches for numerical features are difficult to capture informative knowledge because of the low capacity or hard discretization based on the offline expertise feature engineering. In this paper, we propose a novel embedding learning framework for numerical features in CTR prediction (AutoDis) with high model capacity, end-to-end training and unique representation properties preserved. AutoDis consists of three core components: meta-embeddings, automatic discretization and aggregation. Specifically, we propose meta-embeddings for each numerical field to learn global knowledge from the perspective of field with a manageable number of parameters. Then the differentiable automatic discretization performs soft discretization and captures the correlations between the numerical features and meta-embeddings. Finally, distinctive and informative embeddings are learned via an aggregation function. Comprehensive experiments on two public and one industrial datasets are conducted to validate the effectiveness of AutoDis. Moreover, AutoDis has been deployed onto a mainstream advertising platform, where online A/B test demonstrates the improvement over the base model by 2.1% and 2.7% in terms of CTR and eCPM, respectively. In addition, the code of our framework is publicly available in MindSpore.

CTR prediction, which aims to estimate the probability that a user will click an item, plays a crucial role in online advertising and recommender system. Feature interaction modeling based and user interest mining based methods are the two kinds of most popular techniques that have been extensively explored for many years and have made great progress for CTR prediction. However, (1) feature interaction based methods which rely heavily on the co-occurrence of different features, may suffer from the feature sparsity problem (i.e., many features appear few times); (2) user interest mining based methods which need rich user behaviors to obtain user's diverse interests, are easy to encounter the behavior sparsity problem (i.e., many users have very short behavior sequences). To solve these problems, we propose a novel module named Dual Graph enhanced Embedding, which is compatible with various CTR prediction models to alleviate these two problems. We further propose a Dual Graph enhanced Embedding Neural Network(DG-ENN) for CTR prediction. Dual Graph enhanced Embedding exploits the strengths of graph representation with two carefully designed learning strategies (divide-and-conquer, curriculum-learning-inspired organized learning) to refine the embedding. We conduct comprehensive experiments on three real-world industrial datasets. The experimental results show that our proposed DG-ENN significantly outperforms state-of-the-art CTR prediction models. Moreover, when applying to state-of-the-art CTR prediction models, Dual graph enhanced embedding always obtains better performance. Further case studies prove that our proposed dual graph enhanced embedding could alleviate the feature sparsity and behavior sparsity problems. Our framework will be open-source based on MindSpore in the near future.

Prediction over tabular data is an essential task in many data science applications such as recommender systems, online advertising, medical treatment, etc. Tabular data is structured into rows and columns, with each row as a data sample and each column as a feature attribute. Both the columns and rows of the tabular data carry useful patterns that could improve the model prediction performance. However, most existing models focus on the cross-column patterns yet overlook the cross-rowpatterns as they deal with single samples independently. In this work, we propose a general learning framework named Retrieval & Interaction Machine (RIM) that fully exploits both cross-row and cross-column patterns among tabular data. Specifically, RIM first leverages search engine techniques to efficiently retrieve useful rows of the table to assist the label prediction of the target row, then uses feature interaction networks to capture the cross-column patterns among the target row and the retrieved rows so as to make the final label prediction. We conduct extensive experiments on 11 datasets of three important tasks, i.e., CTR prediction (classification), top-n recommendation (ranking) and rating prediction (regression). Experimental results show that RIM achieves significant improvements over the state-of-the-art and various baselines, demonstrating the superiority and efficacy of RIM.

Interactive recommendation has drawn widespread attention from both academia and industry due to its effectiveness in real-world mobile applications. Instead of receiving message passively, customers can exploit further with less effort through generated queries. Usually, such systems mainly contain two main components: query generation and item recommendation. In this paper, we propose a novel framework that models both queries and items in shared latent embedding space via a dual sequence transformer structure, which captures customer's potential interest from the prospect of reconciling the historical queries and corresponding customers interactions. We propose a click-through-rate model to generate query candidates, and a session search model for further more precise information. Comprehensive offline and online experiments are conducted, and the results demonstrate that our proposed dual-sequence-transformer based model can better utilize interaction and improve the accuracy of recommendations.

How to effectively mitigate the bias of feedback in recommender systems is an important research topic. In this paper, we first describe the generation process of the biased and unbiased feedback in recommender systems via two respective causal diagrams, where the difference between them can be regarded as the source of bias. We then define this difference as a confounding bias, which can be regarded as a collection of some specific biases that have previously been studied. For the case with biased feedback alone, we derive the conditions that need to be satisfied to obtain a debiased representation from the causal diagrams. Based on information theory, we propose a novel method called debiased information bottleneck (DIB) to optimize these conditions and then find a tractable solution for it. In particular, the proposed method constrains the model to learn a biased embedding vector with independent biased and unbiased components in the training phase, and uses only the unbiased component in the test phase to deliver more accurate recommendations. Finally, we conduct extensive experiments on a public dataset and a real product dataset to verify the effectiveness of the proposed method and discuss its properties.

Because of the superior feature representation ability of deep learning, various deep Click-Through Rate (CTR) models are deployed in the commercial systems by industrial companies. To achieve better performance, it is necessary to train the deep CTR models on huge volume of training data efficiently, which makes speeding up the training process an essential problem. Different from the models with dense training data, the training data for CTR models is usually high-dimensional and sparse. To transform the high-dimensional sparse input into low-dimensional dense real-value vectors, almost all deep CTR models adopt the embedding layer, which easily reaches hundreds of GB or even TB. Since a single GPU cannot afford to accommodate all the embedding parameters, when performing distributed training, it is not reasonable to conduct the data-parallelism only. Therefore, existing distributed training platforms for recommendation adopt model-parallelism. Specifically, they use CPU (Host) memory of servers to maintain and update the embedding parameters and utilize GPU worker to conduct forward and backward computations. Unfortunately, these platforms suffer from two bottlenecks: (1) the latency of pull & push operations between Host and GPU; (2) parameters update and synchronization in the CPU servers. To address such bottlenecks, in this paper, we propose the ScaleFreeCTR: a MixCache-based distributed training system for CTR models. Specifically, in SFCTR, we also store huge embedding table in CPU but utilize GPU instead of CPU to conduct embedding synchronization efficiently. To reduce the latency of data transfer between both GPU-Host and GPU-GPU, the MixCache mechanism and Virtual Sparse Id operation are proposed. Comprehensive experiments are conducted to demonstrate the effectiveness and efficiency of SFCTR. In addition, our system will be open-source based on MindSpore in the near future.

Personalized news recommendation aims to alleviate information overload and help users find news of their interests. Accurately matching candidate news and users' interests is the key to news recommendation. Most existing methods separately encode each user and news into vectors by news contents and then match the two vectors. However, a user's interest may differ in each news or each topic of one news. It's necessary to dynamically learn user and news vector and model their interaction. In this work, we present Recurrent Reasoning Memory Network over BERT (RMBERT) for news recommendation. Compared with other methods, our approach can leverage the ability of content modeling from BERT. Moreover, the recurrent reasoning memory network which performs a series of attention based reasoning steps can dynamically learn user and news vector and model their interaction in each step. As a result, our approach can better model user's interests. We conduct extensive experiments on a real-world news recommendation dataset and the results show that our approach significantly outperforms existing state-of-the-art methods.

Due to the popularity of video contents on the Internet, the information retrieval between videos and texts has attracted broad interest from researchers, which is a challenging cross-modal retrieval task. A common solution is to learn a joint embedding space to measure the cross-modal similarity. However, many existing approaches either pay more attention to textual information, video information, or cross-modal matching methods, but less to all three. We believe that a good video-text retrieval system should take into account all three points, fully exploiting the semantic information of both modalities and considering a comprehensive match. In this paper, we propose a Hierarchical Cross-Modal Graph Consistency Learning Network (HCGC) for video-text retrieval task, which considers multi-level graph consistency for video-text matching. Specifically, we first construct a hierarchical graph representation for the video, which includes three levels from global to local: video, clips and objects. Similarly, the corresponding text graph is constructed according to the semantic relationships among sentence, actions and entities. Then, in order to learn a better match between the video and text graph, we design three types of graph consistency (both direct and indirect): inter-graph parallel consistency, inter-graph cross consistency and intra-graph cross consistency. Extensive experimental results on different video-text datasets demonstrate the effectiveness of our approach on both text-to-video and video-to-text retrieval.

To better exploit search logs and model users' behavior patterns, numerous click models are proposed to extract users' implicit interaction feedback. Most traditional click models are based on the probabilistic graphical model (PGM) framework, which requires manually designed dependencies and may oversimplify user behaviors. Recently, methods based on neural networks are proposed to improve the prediction accuracy of user behaviors by enhancing the expressive ability and allowing flexible dependencies. However, they still suffer from the data sparsity and cold-start problems. In this paper, we propose a novel graph-enhanced click model (GraphCM) for web search. Firstly, we regard each query or document as a vertex, and propose novel homogeneous graph construction methods for queries and documents respectively, to fully exploit both intra-session and inter-session information for the sparsity and cold-start problems. Secondly, following the examination hypothesis, we separately model the attractiveness estimator and examination predictor to output the attractiveness scores and examination probabilities, where graph neural networks and neighbor interaction techniques are applied to extract the auxiliary information encoded in the pre-constructed homogeneous graphs. Finally, we apply combination functions to integrate examination probabilities and attractiveness scores into click predictions. Extensive experiments conducted on three real-world session datasets show that GraphCM not only outperforms the state-of-art models, but also achieves superior performance in addressing the data sparsity and cold-start problems.

The two-tower architecture has been widely applied for learning item and user representations, which is important for large-scale recommender systems. Many two-tower models are trained using various in-batch negative sampling strategies, where the effects of such strategies inherently rely on the size of mini-batches. However, training two-tower models with a large batch size is inefficient, as it demands a large volume of memory for item and user contents and consumes a lot of time for feature encoding. Interestingly, we find that neural encoders can output relatively stable features for the same input after warming up in the training process. Based on such facts, we propose a simple yet effective sampling strategy called Cross-Batch Negative Sampling (CBNS), which takes advantage of the encoded item embeddings from recent mini-batches to boost the model training. Both theoretical analysis and empirical evaluations demonstrate the effectiveness and the efficiency of CBNS.

Personalized news recommendation is a critical technology to help users find interested news, and how to precisely match users' interests and candidate news lies in the core of news recommendation. Existing studies generally learn user's interest vector by aggregating his/her browsed news and then match it with the candidate news vector, which may lose the textual semantic matching signals for recommendation. In this paper, we propose an Attentive Multi-field Matching (AMM) framework for news recommendation which captures the semantic matching representations between each browsed news and candidate news, and then aggregates them as final user-news matching signal. In addition, our method incorporates multi-field information and designs a within-field and cross-field matching mechanism, which leverages complementary information from different fields (e.g., titles, abstracts and bodies) and obtain the multi-field matching representations. To achieve a comprehensive semantic understanding, we employ the most popular language model BERT to learn the matching representation of each browsed-candidate news pair, and incorporate the attention mechanism in aggregating procedure to characterize the importance of each matching representation for the final user-news matching signal. Experiments on the real world datasets validate the effectiveness of AMM.

Modern information retrieval systems, including web search, ads placement, and recommender systems, typically rely on learning from user feedback. Click models, which study how users interact with a ranked list of items, provide a useful understanding of user feedback for learning ranking models. Constructing ”right” dependencies is the key of any successful click model. However, probabilistic graphical models (PGMs) have to rely on manually assigned dependencies, and oversimplify user behaviors. Existing neural network based methods promote PGMs by enhancing the expressive ability and allowing flexible dependencies, but still suffer from exposure bias and inferior estimation. In this paper, we propose a novel framework, Adversarial Imitation Click Model (AICM), based on imitation learning. Firstly, we explicitly learn the reward function that recovers users’ intrinsic utility and underlying intentions. Secondly, we model user interactions with a ranked list as a dynamic system instead of one-step click prediction, alleviating the exposure bias problem. Finally, we minimize the JS divergence through adversarial training and learn a stable distribution of click sequences, which makes AICM generalize well across different distributions of ranked lists. A theoretical analysis has indicated that AICM reduces the exposure bias from O(T2) to O(T). Our studies on a public web search dataset show that AICM not only outperforms state-of-the-art models in traditional click metrics but also achieves superior performance in addressing the exposure bias and recovering the underlying patterns of click sequences.

Relevance measures the relation between query and document which contains several different dimensions, e.g., semantic similarity, topical relatedness, cognitive relevance (the relations in the aspect of knowledge), usefulness, timeliness, utility and so on. However, existing retrieval models mainly focus on semantic similarity and cognitive relevance while ignore other possible dimensions to model relevance. Topical relatedness, as an important dimension to measure relevance, is not well studied in existing neural information retrieval. In this paper, we propose a Topic Enhanced Knowledge-aware retrieval Model (TEKM) that jointly learns semantic similarity, knowledge relevance and topical relatedness to estimate relevance between query and document. We first construct a neural topic model to learn topical information and generate topic embeddings of a query. Then we combine the topic embeddings with a knowledge-aware retrieval model to estimate different dimensions of relevance. Specifically, we exploit kernel pooling to soft match topic embeddings with word and entity in a unified embedding space to generate fine-grained topical relatedness. The whole model is trained in an end-to-end manner. Experiments on a large-scale publicly available benchmark dataset show that TEKM outperforms existing retrieval models. Further analysis also shows how topic relatedness is modeled to improve traditional retrieval model with semantic similarity and knowledge relevance.

With the rapid development of mobile app ecosystem, mobile apps have grown greatly popular. The explosive growth of apps makes it difficult for users to find apps that meet their interests. Therefore, it is necessary to recommend user with a personalized set of apps. However, one of the challenges is data sparsity, as users’ historical behavior data are usually insufficient. In fact, user’s behaviors from different domains in app store regarding the same apps are usually relevant. Therefore, we can alleviate the sparsity using complementary information from correlated domains. It is intuitive to model users’ behaviors using graph, and graph neural networks have shown the great power for representation learning. In this article, we propose a novel model, Deep Multi-Graph Embedding (DMGE), to learn cross-domain app embedding. Specifically, we first construct a multi-graph based on users’ behaviors from different domains, and then propose a multi-graph neural network to learn cross-domain app embedding. Particularly, we present an adaptive method to balance the weight of each domain and efficiently train the model. Finally, we achieve cross-domain app recommendation based on the learned app embedding. Extensive experiments on real-world datasets show that DMGE outperforms other state-of-art embedding methods.

Tag-aware recommender systems (TRS) utilize rich tagging records to better depict user portraits and item features. Recently, many efforts have been done to improve TRS with neural networks. However, these solutions rustically rely on the tag-based features for recommendation, which is insufficient to ease the sparsity, ambiguity and redundancy issues introduced by tags, thus hindering the recommendation performance. In this paper, we propose a novel tag-aware recommendation model named Tag Graph Convolutional Network (TGCN), which leverages the contextual semantics of multi-hop neighbors in the user-tag-item graph to alleviate the above issues. Specifically, TGCN first employs type-aware neighbor sampling and aggregation operation to learn the type-specific neighborhood representations. Then we leverage attention mechanism to discriminate the importance of different node types and creatively employ Convolutional Neural Network (CNN) as type-level aggregator to perform vertical and horizontal convolutions for modeling multi-granular feature interactions. Besides, a TransTag regularization function is proposed to accurately identify user's substantive preference. Extensive experiments on three public datasets and a real industrial dataset show that TGCN significantly outperforms state-of-the-art baselines for tag-aware top-N recommendation.

Learning to rank with implicit feedback is one of the most important tasks in many real-world information systems where the objective is some specific utility, e.g., clicks and revenue. However, we point out that existing methods based on probabilistic ranking principle do not necessarily achieve the highest utility. To this end, we propose a novel ranking framework called U-rank that directly optimizes the expected utility of the ranking list. With a position-aware deep click-through rate prediction model, we address the attention bias considering both query-level and item-level features. Due to the item-specific attention bias modeling, the optimization for expected utility corresponds to a maximum weight matching on the item-position bipartite graph. We base the optimization of this objective in an efficient Lambdaloss framework, which is supported by both theoretical and empirical analysis. We conduct extensive experiments for both web search and recommender systems over three benchmark datasets and two proprietary datasets, where the performance gain of U-rank over state-of-the-arts is demonstrated. Moreover, our proposed U-rank has been deployed on a large-scale commercial recommender and a large improvement over the production baseline has been observed in an online A/B testing.

Click-through rate prediction is an important task in commercial recommender systems and it aims to predict the probability of a user clicking on an item. The event of a user clicking on an item is accompanied by several user and item features. As modelling the feature interactions effectively can lead to better predictions, it has been the focus of many recent approaches including deep learning-based models. However, the existing approaches either (i) model all possible feature interactions for a given order, or (ii) manually select which feature interactions to model. Besides, they use the same network structure or function to model all the feature interactions while ignoring the difference of complexity among them. To address these issues, we propose a neural architecture search based approach called AutoFeature that automatically finds essential feature interactions and selects an appropriate structure to model each of these interactions. Specifically, we first define a flexible architecture search space for the CTR prediction task which covers many popular designs such as PIN, PNN and DeepFM, etc., and enables higher-order interactions. Then we propose an efficient neural architecture search algorithm that recursively refines the search space by partitioning it into several subspaces and samples from higher quality ones. Extensive experiments on multiple CTR prediction benchmarks show the superiority of our AutoFeature over the state-of-the-art baselines. In addition, our experiments show that the learned architectures use fewer flops/parameters and hence can efficiently incorporate higher-order feature interactions. This further boosts the performance. Finally, we show that AutoFeature can find meaningful feature interactions.

Re-ranking is a critical task for large-scale commercial recommender systems. Given the initial ranked lists, top candidates are re-ranked to improve the accuracy of the ranking results. However, existing re-ranking strategies are sub-optimal due to (i) most prior works do not consider explicit item relationships, like being substitutable or complementary, which may mutually influence the user satisfaction on other items in the lists, and (ii) they usually apply an identical re-ranking strategy for all users, with personalized user preferences and intents ignored. To resolve the problem, we construct a heterogeneous graph to fuse the initial scoring information and item relationships information. We develop a graph neural network based framework, IRGPR, to explicitly model transitive item relationships by recursively aggregating relational information from multi-hop neighborhoods. We also incorporate a novel intent embedding network to embed personalized user intents into the propagation. We conduct extensive experiments on real-world datasets, demonstrating the effectiveness of IRGPR in re-ranking. Further analysis reveals that modeling the item relationships and personalized intents are particularly useful for improving the performance of re-ranking.

Recently, deep learning-based models have been widely studied for click-through rate (CTR) prediction and lead to improved prediction accuracy in many industrial applications. However, current research focuses primarily on building complex network architectures to better capture sophisticated feature interactions and dynamic user behaviors. The increased model complexity may slow down online inference and hinder its adoption in real-time applications. Instead, our work targets at a new model training strategy based on knowledge distillation (KD). KD is a teacher-student learning framework to transfer knowledge learned from a teacher model to a student model. The KD strategy not only allows us to simplify the student model as a vanilla DNN model but also achieves significant accuracy improvements over the state-of-the-art teacher models. The benefits thus motivate us to further explore the use of a powerful ensemble of teachers for more accurate student model training. We also propose some novel techniques to facilitate ensembled CTR prediction, including teacher gating and early stopping by distillation loss. We conduct comprehensive experiments against 12 existing models and across three industrial datasets. Both offline and online A/B testing results show the effectiveness of our KD-based training strategy.

Learning feature interactions is crucial for click-through rate (CTR) prediction in recommender systems. In most existing deep learning models, feature interactions are either manually designed or simply enumerated. However, enumerating all feature interactions brings large memory and computation cost. Even worse, useless interactions may introduce noise and complicate the training process. In this work, we propose a two-stage algorithm called Automatic Feature Interaction Selection (AutoFIS). AutoFIS can automatically identify important feature interactions for factorization models with computational cost just equivalent to training the target model to convergence. In the search stage, instead of searching over a discrete set of candidate feature interactions, we relax the choices to be continuous by introducing the architecture parameters. By implementing a regularized optimizer over the architecture parameters, the model can automatically identify and remove the redundant feature interactions during the training process of the model. In the re-train stage, we keep the architecture parameters serving as an attention unit to further boost the performance. Offline experiments on three large-scale datasets (two public benchmarks, one private) demonstrate that AutoFIS can significantly improve various FM based models. AutoFIS has been deployed onto the training platform of Huawei App Store recommendation service, where a 10-day online A/B test demonstrated that AutoFIS improved the DeepFM model by 20.3% and 20.1% in terms of CTR and CVR respectively.

Personalized recommender systems are playing an increasingly important role for online consumption platforms. Because of the multitude of relationships existing in recommender systems, Graph Neural Networks (GNNs) based approaches have been proposed to better characterize the various relationships between a user and items while modeling a user's preferences. Previous graph-based recommendation approaches process the observed user-item interaction graph as a ground-truth depiction of the relationships between users and items. However, especially in the implicit recommendation setting, all the unobserved user-item interactions are usually assumed to be negative samples. There are missing links that represent a user's future actions. In addition, there may be spurious or misleading positive interactions. To alleviate the above issue, in this work, we take a first step to introduce a principled way to model the uncertainty in the user-item interaction graph using the Bayesian Graph Convolutional Neural Network framework. We discuss how inference can be performed under our framework and provide a concrete formulation using the Bayesian Probabilistic Ranking training loss. We demonstrate the effectiveness of our proposed framework on four benchmark recommendation datasets. The proposed method outperforms state-of-the-art graph-based recommendation models. Furthermore, we conducted an offline evaluation on one industrial large-scale dataset. It shows that our proposed method outperforms the baselines, with the potential gain being more significant for cold-start users. This illustrates the potential practical benefit in real-world recommender systems.

Click-through rate (CTR) prediction is an important topic in mobile recommendation systems and computational advertising. As previous research indicates, a key point to maximize CTR is feature conjunction for making the training data more informative. Despite great progress, existing methods still fail to choose suitable settings of feature conjunction for the given data. In particular, a linear model on the pair-wise feature conjunction may overfit the training set if the data set is highly sparse. For such data, a model based on low-rank latent matrices are shown to be more appropriate. Unfortunately, practitioners now face difficulties to decide when to use which. In this paper, we propose an adaptive framework to address the feature conjunction problem. Our proposed framework adaptively chooses effective models to do feature conjunction according to data properties. We offer a case for building feature conjunction based on feature-pair frequency. Efficient training, as well as parameter selection, are thoroughly investigated. We conduct comprehensive online and offline experiments to demonstrate the effectiveness of the adaptive model over existing models for CTR prediction.

Most commercial industrial recommender systems have built their closed feedback loops. Though it is helpful in item recommendation and model training, the closed feedback loop may lead to the so-called bias problems, including the position bias, selection bias and popularity bias. The recommendation models trained with biased may hurt the user experiences by recommending homogenous items. How to control the biases in the closed feedback loop has become one of major challenges in modern recommender systems. This talk discusses the counterfactual learning technologies for tackling the bias problem in recommendation.

Recently, GAN-based collaborative filtering methods have gained increasing attention in recommendation tasks which can learn remarkable user and item representation. However, these existing GAN-based methods mainly suffer from two limitations: (1) Their trainings are not comprehensive given the fact that the discriminator may be trained misleadingly and over-early converging since the generator may accidentally sample real items as fake ones, resulting in the emergence of contradicting labels for the same items. (2) They fail to consider implicit friends (users with the same interests.), leading to severe limitations of recommendation performance. In this paper, we propose BiGAN, an innovative bidirectional adversarial recommendation model which can alleviate the limitations mentioned above in recommendation tasks. It consists of two GANs, namely ForwardGAN and BackwardGAN. Specifically, ForwardGAN learns to generate a group of possible interacted items given a specific user, it aims to ensure that the discriminator Df can be trained effectively. Furthermore, BackwardGAN fully exploits implicit friends with similar behaviors, then propagates them back to ForwardGAN, where a similarity exploration strategy is implemented to gain more outstanding user representation. Therefore, two GANs are trained jointly in a circle, where the augment of one GAN will enhance another one, leading to the promising user and item representation. In the experimental part, we demonstrate that our model is superior to other state-of-the-art recommenders.

Predicting tags for a given item and leveraging tags to assist item recommendation are two popular research topics in the field of recommender system. Previous studies mostly focus only one of them to make contributions. However, we believe that these tasks are inherently correlated with each other: tags can provide additional information to profile items for more accurate recommendation; user behaviors can help to infer item relationships to benefit the item tagging process. In order to take the advantages of such mutually influential signals, we propose to integrate item tagging and tag-based recommendation into a unified model. We firstly design a basic framework, where the user-item interaction signals are leveraged to supervise the item tagging process. Then we extend the basic model with a bootstrapping technique to circulate such mutual improvements between different tasks. We conduct extensive experiments based on real-word datasets to demonstrate our model's superiorities.

Factorization Machine (FM)-based models can only reveal the relationship between a pair of features. With all feature embeddings fed to a MLP, DNN-based factorization models which combine FM with multi-layer perceptron (MLP) can only reveal the relationship among some features implicitly. Some other DNN-based methods apply CNN to generate feature interactions. However, (1) they model feature interactions at the bit-wise (where only part of an embedding is utilized to generate feature interactions), which can not express the semantics of features comprehensively, (2) they can only model the interactions among the neighboring features. To deal with aforementioned problems, this paper proposes a Multi-Branch Convolutional Network (MBCN) which includes three branches: the standard convolutional layer, the dilated convolutional layer and the bias layer. MBCN is able to explicitly model feature interactions with arbitrary orders at the vector-wise, which fully express context-aware feature semantics. Extensive experiments on three public benchmark datasets are conducted to demonstrate the superiority of MBCN, compared to the state-of-the-art baselines for context-aware top-k recommendation.

Recommender systems are feedback loop systems, which often face bias problems such as popularity bias, previous model bias and position bias. In this paper, we focus on solving the bias problems in a recommender system via a uniform data. Through empirical studies in online and offline settings, we observe that simple modeling with a uniform data can alleviate the bias problems and improve the performance. However, the uniform data is always few and expensive to collect in a real product. In order to use the valuable uniform data more effectively, we propose a general knowledge distillation framework for counterfactual recommendation that enables uniform data modeling through four approaches: (1) label-based distillation focuses on using the imputed labels as a carrier to provide useful de-biasing guidance; (2) feature-based distillation aims to filter out the representative causal and stable features; (3) sample-based distillation considers mutual learning and alignment of the information of the uniform and non-uniform data; and (4) model structure-based distillation constrains the training of the models from the perspective of embedded representation. We conduct extensive experiments on both public and product datasets, demonstrating that the proposed four methods achieve better performance over the baseline models in terms of AUC and NLL. Moreover, we discuss the relation between the proposed methods and the previous works. We emphasize that counterfactual modeling with uniform data is a rich research area, and list some interesting and promising research topics worthy of further exploration. Note that the source codes are available at \urlhttps://github.com/dgliu/SIGIR20_KDCRec.

Modelling feature interactions is key in Click-Through Rate (CTR) predictions. State-of-the-art models usually include explicit feature interactions to better model non-linearity in a deep network, but enumerating all feature combinations of high orders is not efficient and brings challenges to network optimization. In this work, we use AutoML to seek useful high-order feature interactions to train on without manual feature selection. For this purpose, an end-to-end model, AutoGroup, is proposed, which casts the selection of feature interactions as a structural optimization problem. In a nutshell, AutoGroup first automatically groups useful features into a number of feature sets. Then, it generates interactions of any order from these feature sets using a novel interaction function. The main contribution of AutoGroup is that it performs both dimensionality reduction and feature selection which are not seen in previous models. Offline experiments on three public large-scale benchmark datasets demonstrate the superior performance and efficiency of AutoGroup over state-of-the-art models. Furthermore, a ten-day online A/B test verifies that AutoGroup can be reliably deployed in production and outperform the commercial baseline by 10% on average in terms of CTR and CVR.

Tagging has been recognized as a successful practice to boost relevance matching for information retrieval (IR), especially when items lack rich textual descriptions. A lot of research has been done for either multi-label text categorization or image annotation. However, there is a lack of published work that targets at item tagging specifically for IR. Directly applying a traditional multi-label classification model for item tagging is sub-optimal, due to the ignorance of unique characteristics in IR. In this work, we propose to formulate item tagging as a link prediction problem between item nodes and tag nodes. To enrich the representation of items, we leverage the query logs available in IR tasks, and construct a query-item-tag tripartite graph. This formulation results in a TagGNN model that utilizes heterogeneous graph neural networks with multiple types of nodes and edges. Different from previous research, we also optimize both full tag prediction and partial tag completion cases in a unified framework via a primary-dual loss mechanism. Experimental results on both open and industrial datasets show that our TagGNN approach outperforms the state-of-the-art multi-label classification approaches.

Personalized recommendation plays an important role in many online services. Substantial research has been dedicated to learning embeddings of users and items to predict a user's preference for an item based on the similarity of the representations. In many settings, there is abundant relationship information, including user-item interaction history, user-user and item-item similarities. In an attempt to exploit these relationships to learn better embeddings, researchers have turned to the emerging field of Graph Convolutional Neural Networks (GCNs), and applied GCNs for recommendation. Although these prior works have demonstrated promising performance, directly apply GCNs to process the user-item bipartite graph is suboptimal because the GCNs do not consider the intrinsic differences between user nodes and item nodes. Additionally, existing large-scale graph neural networks use aggregation functions such as sum/mean/max pooling operations to generate a node embedding that considers the nodes' neighborhood (i.e., the adjacent nodes in the graph), and these simple aggregation strategies fail to preserve the relational information in the neighborhood. To resolve the above limitations, in this paper, we propose a novel framework NIA-GCN, which can explicitly model the relational information between neighbor nodes and exploit the heterogeneous nature of the user-item bipartite graph. We conduct empirical studies on four public benchmarks, demonstrating a significant improvement over state-of-the-art approaches. Furthermore, we generalize our framework to a commercial App store recommendation scenario. We observe significant improvement on a large-scale commercial dataset, demonstrating the practical potential for our proposed solution as a key component of a large scale commercial recommender system. Furthermore, online experiments are conducted to demonstrate that NIA-GCN outperforms the baseline by 10.19% and 9.95% in average in terms of CTR and CVR during ten-day AB test in a mainstream App store.

Recommender system is one of the most successful machine learning technologies for commerce. However, it can reinforce the closed feedback loop problem, where the recommender system generates items to users, then the further recommendation model is trained with the data that users' feedback to the items. Such self-reinforcing pattern can cause data bias problems. There are several debiasing methods, inverse-propensity-scoring (IPS) is a practical one for industry product. Since it is relatively easy to reweight training samples, and ameliorate the distribution shift problem. However,because of deterministic policy problem and confoundings in real-world data, it is hard to predict propensity score accurately. Inspired by the sample reweight work for robust deep learning, we propose a novel influence function based method for recommendation modeling, and analyze how the influence function corrects the bias. In the experiments, our proposed method achieves better performance against the state-of-the-art approaches.

Interactive recommender system (IRS) has drawn huge attention because of its flexible recommendation strategy and the consideration of optimal long-term user experiences. To deal with the dynamic user preference and optimize accumulative utilities, researchers have introduced reinforcement learning (RL) into IRS. However, RL methods share a common issue of sample efficiency, i.e., huge amount of interaction data is required to train an effective recommendation policy, which is caused by the sparse user responses and the large action space consisting of a large number of candidate items. Moreover, it is infeasible to collect much data with explorative policies in online environments, which will probably harm user experience. In this work, we investigate the potential of leveraging knowledge graph (KG) in dealing with these issues of RL methods for IRS, which provides rich side information for recommendation decision making. Instead of learning RL policies from scratch, we make use of the prior knowledge of the item correlation learned from KG to (i) guide the candidate selection for better candidate item retrieval, (ii) enrich the representation of items and user states, and (iii) propagate user preferences among the correlated items over KG to deal with the sparsity of user feedback. Comprehensive experiments have been conducted on two real-world datasets, which demonstrate the superiority of our approach with significant improvements against state-of-the-arts.

The research of reinforcement learning (RL) based recommendation method has become a hot topic in recommendation community, due to the recent advance in interactive recommender systems. The existing RL recommendation approaches can be summarized into a unified framework with three components, namely embedding component (EC), state representation component (SRC) and policy component (PC). We find that EC cannot be nicely trained with the other two components simultaneously. Previous studies bypass the obstacle through a pre-training and fixing strategy, which makes their approaches unlike a real end-to-end fashion. More importantly, such pre-trained and fixed EC suffers from two inherent drawbacks: (1) Pre-trained and fixed embeddings are unable to model evolving preference of users and item correlations in the dynamic environment; (2) Pre-training is inconvenient in the industrial applications. To address the problem, in this paper, we propose an End-to-end Deep Reinforcement learning based Recommendation framework (EDRR). In this framework, a supervised learning signal is carefully designed for smoothing the update gradients to EC, and three incorporating ways are introduced and compared. To the best of our knowledge, we are the first to address the training compatibility between the three components in RL based recommendations. Extensive experiments are conducted on three real-world datasets, and the results demonstrate the proposed EDRR effectively achieves the end-to-end training purpose for both policy-based and value-based RL models, and delivers better performance than state-of-the-art methods.

Recommender systems often face heterogeneous datasets containing highly personalized historical data of users, where no single model could give the best recommendation for every user. We observe this ubiquitous phenomenon on both public and private datasets and address the model selection problem in pursuit of optimizing the quality of recommendation for each user. We propose a meta-learning framework to facilitate user-level adaptive model selection in recommender systems. In this framework, a collection of recommenders is trained with data from all users, on top of which a model selector is trained via meta-learning to select the best single model for each user with the user-specific historical data. We conduct extensive experiments on two public datasets and a real-world production dataset, demonstrating that our proposed framework achieves improvements over single model baselines and sample-level model selector in terms of AUC and LogLoss. In particular, the improvements may lead to huge profit gain when deployed in online recommender systems.

Personalized recommendation is ubiquitous, playing an important role in many online services. Substantial research has been dedicated to learning vector representations of users and items with the goal of predicting a user's preference for an item based on the similarity of the representations. Techniques range from classic matrix factorization to more recent deep learning based methods. However, we argue that existing methods do not make full use of the information that is available from user-item interaction data and the similarities between user pairs and item pairs. In this work, we develop a graph convolution-based recommendation framework, named Multi-Graph Convolution Collaborative Filtering (Multi-GCCF), which explicitly incorporates multiple graphs in the embedding learning process. Multi-GCCF not only expressively models the high-order information via a bipartite user-item interaction graph, but integrates the proximal information by building and processing user-user and item-item graphs. Furthermore, we consider the intrinsic difference between user nodes and item nodes when performing graph convolution on the bipartite graph. We conduct extensive experiments on four publicly accessible benchmarks, showing significant improvements relative to several state-of-the-art collaborative filtering and graph neural network-based recommendation models. Further experiments quantitatively verify the effectiveness of each component of our proposed model and demonstrate that the learned embeddings capture the important relationship structure.

User response prediction is a crucial component for personalized information retrieval and filtering scenarios, such as recommender system and web search. The data in user response prediction is mostly in a multi-field categorical format and transformed into sparse representations via one-hot encoding. Due to the sparsity problems in representation and optimization, most research focuses on feature engineering and shallow modeling. Recently, deep neural networks have attracted research attention on such a problem for their high capacity and end-to-end training scheme. In this article, we study user response prediction in the scenario of click prediction. We first analyze a coupled gradient issue in latent vector-based models and propose kernel product to learn field-aware feature interactions. Then, we discuss an insensitive gradient issue in DNN-based models and propose Product-based Neural Network, which adopts a feature extractor to explore feature interactions. Generalizing the kernel product to a net-in-net architecture, we further propose Product-network in Network (PIN), which can generalize previous models. Extensive experiments on four industrial datasets and one contest dataset demonstrate that our models consistently outperform eight baselines on both area under curve and log loss. Besides, PIN makes great click-through rate improvement (relatively 34.67%) in online A/B test.

For Click-Through Rate (CTR) prediction, Field-aware Factorization Machines (FFM) have exhibited great effectiveness by considering field information. However, it is also observed that FFM suffers from the overfitting problem in many practical scenarios. In this paper, we propose a Field-aware Probabilistic Embedding Neural Network (FPENN) model with both good generalization ability and high accuracy. FPENN estimates the probability distribution of the field-aware embedding rather than using the single point estimation (the maximum a posteriori estimation) to prevent overfitting. Both low-order and high-order feature interactions are considered to improve the accuracy. FPENN consists of three components, i.e., FPE component, Quadratic component and Deep component. FPE component outputs probabilistic embedding to the other two components, where various confidence levels for feature embeddings are incorporated to enhance the robustness and the accuracy. Quadratic component is designed for extracting low-order feature interactions, while Deep component aims at capturing high-order feature interactions. Experiments are conducted on two benchmark datasets, Avazu and Criteo. The results confirm that our model alleviates the overfitting problem while having a higher accuracy.

It is well known that the key issue of online marketing is to accurately find the target user groups for the corresponding advertisements. Traditionally, the advertising products target user groups based on search keywords (e.g. AdWords), page visiting (e.g. AdSense), and etc. In this work, we explore a new targeting strategy - targeting users based on their downloaded apps. Specifically, we make use of a subset of the data from the Huawei App Store, consisting of 20,169,033 users and 122,875 apps with 453,346,250 downloads during one year. For each marketing job, the advertiser only need to label a small set of apps, usually around 10 apps, that the target users might be interested in. Our system xRank will automatically find a list of top potential target users for the advertiser. We implement xRank with very efficient performance on the top of Hadoop to be capable for a real web-scale dataset, and then conducted our system to several real marketing tasks. The results show that, for each marketing task, with only a few labels, xRank can effectively find a precise target group of users, and can also significantly improved the effectiveness of our online marketing compared to the rule-based approaches in the current system.

With the increasing prevalence of versatile mobile devices and the fast deployment of broadband mobile networks, a huge volume of Mobile Broadband (MBB) data has been generated over time. The MBB data naturally contain rich information of a large number of mobile users, covering a considerable fraction of whole population nowadays, including the mobile applications they are using at different locations and time; the MBB data may present the unprecedentedly large knowledge base of human behavior which has highly recognized commercial and social value. However, the storage, management and analysis of the huge and fast growing volume of MBB data post new and significant challenges to the industrial practitioners and research community. In this demonstration, we present a new, MBB data tailored, distributed analytic system named OceanST which has addressed a series of problems and weaknesses of the existing systems, originally designed for more general purpose and capable to handle MBB data to some extent. OceanST is featured by ( i ) efficiently loading of ever-growing MBB data, ( ii ) a bunch of spatiotemporal aggregate queries and basic analysis APIs frequently found in various MBB data application scenarios, and ( iii ) sampling-based approximate solution with provable accuracy bound to cope with huge volume of MBB data. The demonstration will show the advantage of OceanST in a cluster of 5 machines using 3TB data.