In the realm of multi-behavior sequential recommendation (MBSR), the complexity and heterogeneity of user interactions pose substantial challenges for sequence modeling. Existing studies involve significant efforts in combining different modules to learn more expressive multi-behavior sequence representations or designing strategies to extract user preferences related to the target behavior. Despite their effectiveness, these methods neglect a thorough analysis of how behavioral information shapes the probability distribution for next-item prediction, which is crucial for accurately modeling user preferences. To this end, we first analyze the learning distribution of MBSR, shedding light on the significance of target behavior in next-item prediction. Building upon this insight, we propose a Behavior-Aware Regularization approach for multi-behavior sequential Rec ommendation (BAR4Rec), where we introduce a regularization loss function to preserve the intrinsic constraints of target behavior. In this way, the target probability distribution is extracted from the whole distribution and naturally evolves into a more compatible and tractable form, thus facilitating model design and training. We evaluate the proposed method on three real-world datasets, and the results validate the efficacy of our approach.

Locality sensitive hashing (LSH) is a representative approach for nearest neighbor (NN) search in high-dimensional spaces, which is able to answer c-approximate NN (c-ANN) queries in sublinear time with constant probability. Existing advanced LSH methods leverage a plurality of novel techniques such as query-aware dynamic bucketing, virtual rehashing, and efficient indexing to achieve state-of-the-art performance. However, they rely on similar random LSH functions, which provides distance estimations that are irrelevant to the given data distribution. Therefore, the quality of the searched candi-dates is suboptimal. In this study, we reformulate the c-ANN query from the perspective of data distribution. Specifically, we propose a novel distribution-aware c-ANN query, which can guarantee the quality of searched results from the query distribution perspective. We introduce an accurately unbiased distance estimator into LSH methods, which can provide more precise distance estimations by modeling the data distribution. We also conduct rigorous theoretical analysis to prove that our methods can correctly answer the distribution-aware c-ANN query with at least a constant probability. Experiments on seven real datasets with different sizes and dimensionalities indicate that the proposed method can achieve better performance than existing LSH methods in terms of efficiency and effectiveness.

The two-tower model has become prevalent in recommender systems for its computational efficiency and robust predictive capabilities. The model usually employs two independent neural networks to encode user and item data separately, and predicts the similarity score with inner product or cosine functions, depending on which the Top-k ranked item list is generated. The optimization process typically involves a multi-label classification objective, often guided by surrogate loss functions like Softmax and One-vs-All (OvA), to enhance the recommendation performance. Despite both Softmax and OvA losses being Bayes-consistent, empirical observations reveal a significant performance gap in evaluation metrics, suggesting limitations in Bayes-consistency for analyzing loss effectiveness. To address this, we introduce ℋ-consistency into the discussion, which provides non-asymptotic and hypothesis-specific guarantees for Top-k classification within the two-tower model's hypothesis space. Through theoretical analysis, we demonstrate that Softmax and Cosine Contrastive Loss exhibit ℋ-consistency, while the OvA loss does not, explaining the observed performance discrepancies. Our findings bridge the gap between theoretical properties and practical outcomes, offering deeper insights into the optimization of two-tower models and contributing to the development of more effective recommendation systems.

Deep learning has been widely applied in recommender systems, which has recently achieved revolutionary progress. However, most existing learning-based methods assume that the user and item distributions remain unchanged between the training phase and the test phase. However, the distribution of user and item features can naturally shift in real-world scenarios, potentially resulting in a substantial decrease in recommendation performance. This phenomenon can be formulated as an Out-Of-Distribution (OOD) recommendation problem. To address this challenge, we propose a novelDualTest-Time-Training framework forOODRecommendation, termedDT3OR. In DT3OR, we incorporate a model adaptation mechanism during the test-time phase to carefully update the recommendation model, allowing the model to adapt specially to the shifting user and item features. To be specific, we propose a self-distillation task and a contrastive task to assist the model learning both the user’s invariant interest preferences and the variant user/item characteristics during the test-time phase, thus facilitating a smooth adaptation to the shifting features. Furthermore, we provide theoretical analysis to support the rationale behind our dual test-time training framework. To the best of our knowledge, this paper is the first work to address OOD recommendation via a test-time-training strategy. We conduct experiments on five datasets with various backbones. Comprehensive experimental results have demonstrated the effectiveness of DT3OR compared to other state-of-the-art baselines.

Collaborative filtering (CF) works have demonstrated the robust capabilities of Shallow Autoencoders on implicit feedback, showcasing highly competitive performance with other reasonable approaches (e.g., iALS and VAE-CF). However, despite their dual advantages of high performance and simple construction, EASE still exhibits several major shortcomings that must be addressed. To be more precise, the scalability of EASE is limited by the number of items, which determines the storage and inversion cost of a large dense matrix; the square-loss optimization objective does not consistently meet the recommendation task’s requirement for predicting personalized rankings, resulting in suboptimal outcomes; the regularization coefficients are sensitive and require recalibration with different datasets, leading to an exhaustive and time-consuming fine-tuning process. In order to address these obstacles, we propose a novel approach called Similarity-Structure Aware Shallow Autoencoder (AutoS \(^2\) AE) that aims to enhance both recommendation accuracy and model efficiency. Our method introduces three similarity structures: Co-occurrence, KNN, and NSW graphs, which replace the large dense matrix in EASE with a sparse structure, thus facilitating model compression. Additionally, we optimize the model by incorporating a low-rank training component into the matrix and applying a weighted square loss for improved ranking-oriented approximations. To automatically tune the hyperparameters, we further design two validation losses on the validation set for guidance and update the hyperparameters using the gradients of these validation losses. Both theoretical analyses regarding the introduction of similarity structures and empirical evaluations on multiple real-world datasets demonstrate the effectiveness of our proposed method, which significantly outperforms competing baselines.

Searching for similar trajectories from a database is an important way for extracting human-understandable knowledge. However, due to the huge volume of trajectories and high computation complexity of distance between trajectories, it is difficult to search for exact results, which motivates the research of approximating approaches. In this study, we propose a learning to hash method for trajectory similarity computation and search, called Traj2Hash, which consists of a two-channel trajectory encoder and a hash layer to encode trajectories into Euclidean and Hamming space, respectively. The embeddings of trajectories obtained from the encoder are capable of preserving the reverse symmetric property and more representative due to the reverse augmentation and the lower-bound induced read-out layer. Moreover, we design a decomposed grid representation in the encoder to make the model lighter and better. In the model training phase, we combine a weighted mean squared error loss and a ranking-based hashing loss to enable the model similarity-aware and representations self-structured, respectively, in which a fast trajectory triplet generation method is leveraged to enrich the training corpus. Extensive experiments conducted on real data offer evidence of the effectiveness and efficiency of the proposed model.

The popularity of mobile devices has led to the increased attention of Spatial Crowdsourcing (SC), a framework that assigns location-sensitive tasks to mobile workers. Task recommendation is crucial in helping workers discover attractive tasks. Existing studies have focused on modeling workers' preferences from past task-performing patterns, but their performance is sub-optimal due to the strong coupling of sequentiality, spatiality, and temporality. Moreover, achieving the highest preference-based utility of workers in most of the existing task recommendation studies is inferior to the benefits of the SC platform and the satisfaction of workers in a long range, due to the lower task coverage rate and the poor diversity in a worker's recommended list. To address these problems, we propose a Diversity-Coverage Balanced Task Recommendation (DCBTaskRec) framework. Specifically, we first introduce a decoupled worker preference learning model that adopts self-attention networks as the backbone and decouples the modeling of multiple factors in attention scores. Additionally, we provide an optimal diveristy-aware approach to maximize the recommendation diversity while keeping high preference-based utility of workers to satisfy the multiple tastes of workers. From the side of the SC platform, we also provide two approaches (i.e., greedy coverage-aware approach and diversity-coverage balanced approach) to achieve high coverage and provide a trade-off between diversity and coverage, respectively. Extensive experiments offer insight into the effectiveness of the proposed framework.

Online recommenders have attained growing interest and created great revenue for businesses. Given numerous users and items, incremental update becomes a mainstream paradigm for learning large-scale models in industrial scenarios, where only newly arrived data within a sliding window is fed into the model, meeting the strict requirements of quick response. However, this strategy would be prone to overfitting to newly arrived data. When there exists a significant drift of data distribution, the long-term information would be discarded, which harms the recommendation performance. Conventional methods address this issue through native model-based continual learning methods, without analyzing the data characteristics for online recommenders. To address the aforementioned issue, we propose an incremental update framework for online recommenders with Data-Driven Prior (DDP), which is composed of Feature Prior (FP) and Model Prior (MP). The FP performs the click estimation for each specific value to enhance the stability of the training process. The MP incorporates previous model output into the current update while strictly following the Bayes rules, resulting in a theoretically provable prior for the robust update. In this way, both the FP and MP are well integrated into the unified framework, which is model-agnostic and can accommodate various advanced interaction models. Extensive experiments on two publicly available datasets as well as an industrial dataset demonstrate the superior performance of the proposed framework. © 2023 Copyright held by the owner/author(s). Publication rights licensed to ACM.

Recommendation retrievers commonly retrieve user potentially preferred items from numerous items, where the query and item representation are learned according to the dual encoders with the log-softmax loss. Under real scenarios, the number of items becomes considerably large, making it exceedingly difficult to calculate the partition function with the whole item corpus. Negative sampling, which samples a subset from the item corpus, is widely used to accelerate the model training. Among different samplers, the in-batch sampling is commonly adopted for online recommendation retrievers, which regards the other items within the mini-batch as the negative samples for the given query, owing to its time and memory efficiency. However, the sample selection bias occurs due to the skewed feedback, harming the retrieval quality. In this paper, we propose a negative sampling approach named Batch-Mix Negative Sampling (BMNS), which adopts batch mixing operation to generate additional negatives for model training. Concretely, BMNS first generates new negative items with the sampled mix coefficient from the Beta distribution, after which a tailored correct strategy guided by frequency is designed to match the sampled softmax loss. In this way, the effort of re-encoding items out of the mini-batch is reduced while also improving the representation space of the negative set. The empirical experiments on four real-world datasets demonstrate BMNS is superior to the competitive negative inbatch sampling method.

A dozen recommendation libraries have recently been developed to accommodate popular recommendation algorithms for reproducibility. However, they are almost simply a collection of algorithms, overlooking the modularization of recommendation algorithms and their usage in practical scenarios. Algorithmic modularization has the following advantages: 1) helps to understand the effectiveness of each algorithm; 2) easily assembles new algorithms with well-performed modules by either drag-and-drop programming or automatic machine learning; 3) enables reinforcement between algorithms since one algorithm may act as a module of another algorithm. To this end, we develop a highly-modularized recommender system -- RecStudio, in which any recommendation algorithm is categorized into either a ranker or a retriever. In the RecStudio library, we implement 90 recommendation algorithms with the pure Pytorch, covering both common algorithms in other libraries and complex algorithms involving multiple recommendation models. RecStudio is featured from several perspectives, such as index-supported efficient recommendation and evaluation, GPU-accelerated negative sampling, hyperparameter learning on the validation, and cooperation between the retriever and ranker. RecStudio is also equipped with a web service, where the recommendation pipeline can be quickly established and visually evaluated on selected datasets, and the evaluation results are automatically archived and visualized in a leaderboard. The project and documents are released at http://recstudio.org.cn.

The Embarrassingly Shallow Autoencoders (EASE and SLIM) are strong recommendation methods based on implicit feedback, compared to competing methods like iALS and VAE-CF. However, EASE suffers from several major shortcomings. First, the training and inference of EASE can not scale with the increasing number of items since it requires storing and inverting a large dense matrix; Second, though its optimization objective – the square loss– can yield a closed-form solution, it is not consistent with recommendation goal – predicting a personalized ranking on a set of items, so that its performance is far from optimal w.r.t ranking-oriented recommendation metrics. Finally, the regularization coefficients are sensitive w.r.t recommendation accuracy and vary a lot across different datasets, so the fine-tuning of these parameters is important yet time-consuming. To improve training and inference efficiency, we propose a Similarity-Structure Aware Shallow Autoencoder on top of three similarity structures, including Co-Occurrence, KNN and NSW. We then optimize the model with a weighted square loss, which is proven effective for ranking-based recommendation but still capable of deriving closed-form solutions. However, the weight in the loss can not be learned in the training set and is similarly sensitive w.r.t the accuracy to regularization coefficients. To automatically tune the hyperparameters, we design two validation losses on the validation set for guidance, and update the hyperparameters with the gradient of the validation losses. We finally evaluate the proposed method on multiple real-world datasets and show that it outperforms seven competing baselines remarkably, and verify the effectiveness of each part in the proposed method.

Deep recommender systems (DRS) are intensively applied in modern web services. To deal with the massive web contents, DRS employs a two-stage workflow: retrieval and ranking, to generate its recommendation results. The retriever aims to select a small set of relevant candidates from the entire items with high efficiency; while the ranker, usually more precise but time-consuming, is supposed to further refine the best items from the retrieved candidates. Traditionally, the two components are trained either independently or within a simple cascading pipeline, which is prone to poor collaboration effect. Though some latest works suggested to train retriever and ranker jointly, there still exist many severe limitations: item distribution shift between training and inference, false negative, and misalignment of ranking order. As such, it remains to explore effective collaborations between retriever and ranker.

Query cost estimation, which is to estimate the query plan cost and query execution cost, is of utmost importance to query optimizers. Query plan cost estimation heavily relies on accurate cardinality estimation, and query execution cost estimation gives good hints on query latency, both of which are challenging in database management systems. Despite decades of research, existing studies either over-simplify the models only using histograms and polynomial calculation that leads to inaccurate estimates, or over-complicate them by using cumbersome neural networks with the requirements for large amounts of training data hence poor computational efficiency. Besides, most of the studies ignore the diversity of query plan structures. In this work, we propose a plan-based query cost estimation framework, called Saturn, which can eStimate cardinality and latency accurately and efficiently, for any query plan structures. Saturn first encodes each query plan tree into a compressed vector by using a traversal-based query plan autoencoder to cope with diverse plan structures. The compressed vectors can be leveraged to distinguish different query types, which is highly useful for downstream tasks. Then a pseudo label generator is designed to acquire all cardinality and latency labels with the execution part of the query plans in the training workload, which can significantly reduce the overhead of collecting the real cardinality and latency labels. Finally, a chain-wise transfer learning module is proposed to estimate the cardinality and latency of the query plan in a pipeline paradigm, which further enhances the efficiency. An extensive empirical study on benchmark data offers evidence that Saturn outperforms the state-of-the-art proposals in terms of accuracy, efficiency, and generalizability for query cost estimation.

Join order selection plays an important role in DBMS query optimizers. The problem aims to find the optimal join order with the minimum cost, and usually becomes an NP-hard problem due to the exponentially increasing search space. Recent advanced studies attempt to use deep reinforcement learning (DRL) to generate better join plans than the ones provided by conventional query optimizers. However, DRL-based methods require time-consuming training, which is not suitable for online applications that need frequent periodic re-training. In this paper, we propose a novel framework, namely efficient Join Order selection learninG with Graph-basEd Representation (JOGGER). We firstly construct a schema graph based on the primary-foreign key relationships, from which table representations are well learned to capture the correlations between tables. The second component is the state representation, where a graph convolutional network is utilized to encode the query graph and a tailored-tree-based attention module is designed to encode the join plan. To speed up the convergence of DRL training process, we exploit the idea of curriculum learning, in which queries are incrementally added into the training set according to the level of difficulties. We conduct extensive experiments on JOB and TPC-H datasets, which demonstrate the effectiveness and efficiency of the proposed solutions.

Due to the widespread presence of implicit feedback, recommendation based on them has been a long-standing research problem in academia and industry. However, it suffers from the extremely-sparse problem, since each user only interacts with a few items. One well-known and good-performing method is to treat each user's all uninteracted items as negative with low confidence. The method intrinsically imposes an implicit regularization to penalize large deviation of each user's preferences for uninteracted items from a constant. However, these methods have to assume a constant-rating prior to uninteracted items, which may be questionable. In this paper, we propose a novel ring-based regularization to penalize significant differences of each user's preferences between each item and some other items. The ring structure, described by an item graph, determines which other items are selected for each item in the regularization. The regularization not only averts the introduction of the prior ratings but also implicitly penalizes the remarkable preference differences for all items according to theoretical analysis. However, optimizing the recommenders with the regularization still suffers from computational challenges, so we develop a scalable alternating least square algorithm by carefully designing gradient computation. Therefore, as long as connecting each item with a sublinear/constant number of other items in the item graph, the overall learning algorithm could be comparably efficient to the existing algorithms. The proposed regularization is extensively evaluated with several public recommendation datasets, where the results show that the regularization could lead to considerable improvements in recommendation performance.

Recommendation is prevalently studied for implicit feedback recently, but it seriously suffers from the lack of negative samples, which has a significant impact on the training of recommendation models. Existing negative sampling is based on the static or adaptive probability distributions. Sampling from the adaptive probability receives more attention, since it tends to generate more hard examples, to make recommender training faster to converge. However, item sampling becomes much more time-consuming particularly for complex recommendation models. In this paper, we propose an Adaptive Sampling method based on Importance Resampling (AdaSIR for short), which is not only almost equally efficient and accurate for any recommender models, but also can robustly accommodate arbitrary proposal distributions. More concretely, AdaSIR maintains a contextualized sample pool of fixed-size with importance resampling, from which items are only uniformly sampled. Such a simple sampling method can be proved to provide approximately accurate adaptive sampling under some conditions. The sample pool plays two extra important roles in (1) reusing historical hard samples with certain probabilities; (2) estimating the rank of positive samples for weighting, such that recommender training can concentrate more on difficult positive samples. Extensive empirical experiments demonstrate that AdaSIR outperforms state-of-the-art methods in terms of sampling efficiency and effectiveness.

Variational AutoEncoder (VAE) has been extended as a representative nonlinear method for collaborative filtering. However, the bottleneck of VAE lies in the softmax computation over all items, such that it takes linear costs in the number of items to compute the loss and gradient for optimization. This hinders the practical use due to millions of items in real-world scenarios. Importance sampling is an effective approximation method, based on which the sampled softmax has been derived. However, existing methods usually exploit the uniform or popularity sampler as proposal distributions, leading to a large bias of gradient estimation. To this end, we propose to decompose the inner-product-based softmax probability based on the inverted multi-index, leading to sublinear-time and highly accurate sampling. Based on the proposed proposals, we develop a fast Variational AutoEncoder (FastVAE) for collaborative filtering. FastVAE can outperform the state-of-the-art baselines in terms of both sampling quality and efficiency according to the experiments on three real-world datasets.

One-class collaborative filtering (OCCF) problems are ubiquitous in real-world recommendation systems, such as news recommendation, but suffer from data sparsity and lack of negative items. To address the challenge, the state-of-the-art algorithm assigns uninteracted items with smaller weights of being negative and performs low-rank approximation over the user-item interaction matrix. However, the prior ratings are usually suggested to be zero but may not be well-defined. To avert the direct utilization of prior ratings for uninteracted items, we propose a novel ranking-based implicit regularizer by hypothesizing that users’ preference scores for uninteracted items should not deviate a lot from each other. The regularizer is then used in a ranking-based OCCF framework to penalize large differences of preference scores between uninteracted items. To efficiently optimize model parameters in this framework, we develop the scalable alternating least square algorithm and coordinate descent algorithm, whose time complexity is linearly proportional to the data size. Finally, we extensively evaluate the proposed algorithms on six public real-world datasets. The results show that the proposed regularizer significantly improves the recommendation quality of ranking-based OCCF algorithms, such as BPRMF and RankALS. Moreover, the ranking-based framework with the proposed regularizer outperforms the state-of-the-art recommendation algorithms for implicit feedback.

The factorization-based models have achieved great success in online advertisements and recommender systems due to the capability of efficiently modeling combinational features. These models encode feature interactions by the vector product between feature embedding. Despite the improvement of generalization, the memory consumption of these models grows significantly, because they usually take hundreds to thousands of large categorical features as input. Several existing works try to reduce the memory footprint by hashing, randomized embedding composition, and dimensionality search, but they suffer from either substantial performance degradation or limited memory compression. To this end, in this paper, we propose an extremely memory-efficient Factorization Machine (xLightFM), where each category embedding is composited with latent vectors selected from codebooks. Based on the characteristics of each categorical feature, we further propose to adapt the codebook size with the neural architecture search techniques for compositing the embedding of each categorical feature. This further pushes the limits of memory compression while incurring negligible degradation or even some improvements in prediction performance. We extensively evaluate the proposed algorithm with two real-world datasets. The results demonstrate that xLightFM can outperform the state-of-the-art lightweight factorization-based methods in terms of both prediction quality and memory footprint, and achieve more than 18x and 27x memory compression compared to the vanilla FM on these two datasets, respectively.

Advertising creatives are ubiquitous in E-commerce advertisements and aesthetic creatives may improve the click-through rate (CTR) of the products. Nowadays smart advertisement platforms provide the function of compositing creatives based on source materials provided by advertisers. Since a great number of creatives can be generated, it is difficult to accurately predict their CTR given a limited amount of feedback. Factorization machine (FM), which models inner product interaction between features, can be applied for the CTR prediction of creatives. However, interactions between creative elements may be more complex than the inner product, and the FM-estimated CTR may be of high variance due to limited feedback. To address these two issues, we propose an Automated Creative Optimization (AutoCO) framework to model complex interaction between creative elements and to balance between exploration and exploitation. Specifically, motivated by AutoML, we propose one-shot search algorithms for searching effective interaction functions between elements. We then develop stochastic variational inference to estimate the posterior distribution of parameters based on the reparameterization trick, and apply Thompson Sampling for efficiently exploring potentially better creatives. We evaluate the proposed method with both a synthetic dataset and two public datasets. The experimental results show our method can outperform competing baselines with respect to cumulative regret. The online A/B test shows our method leads to a 7% increase in CTR compared to the baseline.