The prediction objectives of online advertisement ranking models are evolving from probabilistic metrics like conversion rate (CVR) to numerical business metrics like post-click gross merchandise volume (GMV). Unlike the well-studied delayed feedback problem in CVR prediction, delayed feedback modeling for GMV prediction remains unexplored and poses greater challenges, as GMV is a continuous target, and a single click can lead to multiple purchases that cumulatively form the label.

In industrial recommender systems, conversion rate (CVR) is often used for traffic allocation, but fails to fully reflect recommendation effectiveness as it does not account for refund rate (RFR). Thus, net conversion rate (NetCVR), the probability that a clicked item is purchased and not refunded, is proposed to better show true user satisfaction and business value. Unlike CVR, NetCVR prediction involves a more complex multi-stage cascaded delay feedback phenomenon. The two cascaded delays Click->Conversion and Conversion->Refund in NetCVR have opposite effects. Therefore, traditional CVR methods cannot be directly applied. At present, the lack of relevant open-source datasets and online continuous training schemes poses a challenge. To address these, we first introduce CAscadal Sequences of Conversion And Delayed rEfund (CASCADE), the first large-scale open dataset derived from Taobao app for online continuous NetCVR prediction. We further analyze CASCADE and derive three key insights: (1) NetCVR exhibits clear temporal patterns necessitating online continuous modeling; (2) Cascaded modeling CVR and RFR for NetCVR outperforms directly modeling NetCVR; and (3) delay time, which correlated with both CVR and RFR, is an important feature for NetCVR prediction. Based on these insights, we propose neT convErsion caScaded modeLing and debiAsing method (TESLA). This continuous method features a CVR-RFR cascaded architecture, stage-wise debiasing, and a delay-time-aware ranking loss for efficient NetCVR prediction. Experiments show that TESLA outperforms state-of-the-art methods on CASCADE, achieving an absolute improvement of 12.41% in RI-AUC and 14.94% in RI-PRAUC on NetCVR over the strongest baseline. We hope this work provides a new direction for online delayed feedback modeling in NetCVR prediction. Our code and dataset are available at https://github.com/alimama-tech/NetCVR.

Conversion rate (CVR) prediction is a core component of online advertising systems, where the attribution mechanisms-rules for allocating conversion credit across user touchpoints-fundamentally determine label generation and model optimization. While many industrial platforms support diverse attribution mechanisms (e.g., First-Click, Last-Click, Linear, and Data-Driven Multi-Touch Attribution), conventional approaches restrict model training to labels from a single production-critical attribution mechanism, discarding complementary signals in alternative attribution perspectives.

Despite the recognized potential of multimodal data to improve model accuracy, many large-scale industrial recommendation systems, including Taobao display advertising system, predominantly depend on sparse ID features in their models. In this work, we explore approaches to leverage multimodal data to enhance the recommendation accuracy. We start from identifying the key challenges in adopting multimodal data in a manner that is both effective and cost-efficient for industrial systems. To address these challenges, we introduce a two-phase framework, including: 1) the pre-training of multimodal representations to capture semantic similarity, and 2) the integration of these representations with existing ID-based models. Furthermore, we detail the architecture of our production system, which is designed to facilitate the deployment of multimodal representations. Since the integration of multimodal representations in mid-2023, we have observed significant performance improvements in Taobao display advertising system. We believe that the insights we have gathered will serve as a valuable resource for practitioners seeking to leverage multimodal data in their systems.

In machine learning systems, privileged features refer to the features that are available during offline training but inaccessible for online serving. Previous studies have recognized the importance of privileged features and explored ways to tackle online-offline discrepancies. A typical practice is privileged features distillation (PFD): train a teacher model using all features (including privileged ones) and then distill the knowledge from the teacher model using a student model (excluding the privileged features), which is then employed for online serving. In practice, the pointwise cross-entropy loss is often adopted for PFD. However, this loss is insufficient to distill the ranking ability for CTR prediction. First, it does not consider the non-i.i.d. characteristic of the data distribution, i.e., other items on the same page significantly impact the click probability of the candidate item. Second, it fails to consider the relative item order ranked by the teacher model's predictions, which is essential to distill the ranking ability. To address these issues, we first extend the pointwise-based PFD to the listwise-based PFD. We then define the calibration-compatible property of distillation loss and show that commonly used listwise losses do not satisfy this property when employed as distillation loss, thus compromising the model's calibration ability, which is another important measure for CTR prediction. To tackle this dilemma, we propose Calibration-compatible LIstwise Distillation (CLID), which employs carefully-designed listwise distillation loss to achieve better ranking ability than the pointwise-based PFD while preserving the model's calibration ability. We theoretically prove it is calibration-compatible. Extensive experiments on public datasets and a production dataset collected from the display advertising system of Alibaba further demonstrate the effectiveness of CLID.

Cascading architecture has been widely adopted in large-scale advertising systems to balance efficiency and effectiveness. In this architecture, the pre-ranking model is expected to be a lightweight approximation of the ranking model, which handles more candidates with strict latency requirements. Due to the gap in model capacity, the pre-ranking and ranking models usually generate inconsistent ranked results, thus hurting the overall system effectiveness. The paradigm of score alignment is proposed to regularize their raw scores to be consistent. However, it suffers from inevitable alignment errors and error amplification by bids when applied in online advertising. To this end, we introduce a consistency-oriented pre-ranking framework for online advertising, which employs a chunk-based sampling module and a plug-and-play rank alignment module to explicitly optimize consistency of ECPM-ranked results. A ΔNDCG-based weighting mechanism is adopted to better distinguish the importance of inter-chunk samples in optimization. Both online and offline experiments have validated the superiority of our framework. When deployed in Taobao display advertising system, it achieves an improvement of up to +12.3% CTR and +5.6% RPM.

Conversion rate (CVR) prediction is an essential task for e-commerce platforms. However, refunds frequently occur after conversion in online shopping systems, which drives us to pay attention to effective conversion for building healthier services. This paper defines the probability of item purchasing without any subsequent refund as an effective conversion rate (ECVR). A simple paradigm for ECVR prediction is to decompose it into two sub-tasks: CVR prediction and post-conversion refund rate (RFR) prediction. However, RFR prediction suffers from data sparsity (DS) and sample selection bias (SSB) issues, as refund behaviors are only available after user purchase. Furthermore, there is delayed feedback in both sequentially dependent conversion and refund events, named cascade delayed feedback (CDF). Previous studies mainly focus on tackling DS and SSB or delayed feedback for a single event. To jointly tackle these issues in ECVR prediction, we propose an Entire space CAscade Delayed feedback modeling (ECAD) method. Specifically, ECAD deals with DS and SSB by constructing two tasks including CVR and conversion&refund rate (CVRFR) predictions using the entire space modeling framework. In addition, it carefully schedules auxiliary tasks to leverage both conversion and refund time within data to alleviate CDF. Experiments on the offline industrial dataset and online A/B testing demonstrate the effectiveness of ECAD. ECAD has been deployed in the Xianyu recommender system of Alibaba, contributing to a significant improvement of ECVR.

Conversion rate (CVR) prediction is one of the core components in online recommender systems, and various approaches have been proposed to obtain accurate and well-calibrated CVR estimation. However, we observe that a well-trained CVR prediction model often performs sub-optimally during sales promotions. This can be largely ascribed to the problem of the data distribution shift, in which the conventional methods no longer work. To this end, we seek to develop alternative modeling techniques for CVR prediction. Observing similar purchase patterns across different promotions, we propose reusing the historical promotion data to capture the promotional conversion patterns. Herein, we propose a novel Historical Data Reuse (HDR) approach that first retrieves historically similar promotion data and then fine-tunes the CVR prediction model with the acquired data for better adaptation to the promotion mode. HDR consists of three components: an automated data retrieval module that seeks similar data from historical promotions, a distribution shift correction module that re-weights the retrieved data for better aligning with the target promotion, and a TransBlock module that quickly fine-tunes the original model for better adaptation to the promotion mode. Experiments conducted with real-world data demonstrate the effectiveness of HDR, as it improves both ranking and calibration metrics to a large extent. HDR has also been deployed on the display advertising system in Alibaba, bringing a lift of 9% RPM and 16% CVR during Double 11 Sales in 2022.

Despite the development of ranking optimization techniques, pointwise loss remains the dominating approach for click-through rate prediction. It can be attributed to the calibration ability of the pointwise loss since the prediction can be viewed as the click probability. In practice, a CTR prediction model is also commonly assessed with the ranking ability. To optimize the ranking ability, ranking loss (e.g., pairwise or listwise loss) can be adopted as they usually achieve better rankings than pointwise loss. Previous studies have experimented with a direct combination of the two losses to obtain the benefit from both losses and observed an improved performance. However, previous studies break the meaning of output logit as the click-through rate, which may lead to sub-optimal solutions. To address this issue, we propose an approach that can Jointly optimize the Ranking and Calibration abilities (JRC for short). JRC improves the ranking ability by contrasting the logit value for the sample with different labels and constrains the predicted probability to be a function of the logit subtraction. We further show that JRC consolidates the interpretation of logits, where the logits model the joint distribution. With such an interpretation, we prove that JRC approximately optimizes the contextualized hybrid discriminative-generative objective. Experiments on public and industrial datasets and online A/B testing show that our approach improves both ranking and calibration abilities. Since May 2022, JRC has been deployed on the display advertising platform of Alibaba and has obtained significant performance improvements.

Deep learning techniques have been applied widely in industrial recommendation systems. However, far less attention has been paid on the overfitting problem of models in recommendation systems, which, on the contrary, is recognized as a critical issue for deep neural networks. In the context of Click-Through Rate (CTR) prediction, we observe an interesting one-epoch overfitting problem: the model performance exhibits a dramatic degradation at the beginning of the second epoch. Such a phenomenon has been witnessed widely in real-world applications of CTR models. Thereby, the best performance is usually achieved by training with only one epoch. To understand the underlying factors behind the one-epoch phenomenon, we conduct extensive experiments on the production data set collected from the display advertising system of Alibaba. The results show that the model structure, the optimization algorithm with a fast convergence rate, and the feature sparsity are closely related to the one-epoch phenomenon. We also provide a likely hypothesis for explaining such a phenomenon and conduct a set of proof-of-concept experiments. We hope this work can shed light on the future research on training more epochs for better performance.

Feature interaction has been recognized as an important problem in machine learning, which is also very essential for click-through rate (CTR) prediction tasks. In recent years, Deep Neural Networks (DNNs) can automatically learn implicit nonlinear interactions from original sparse features, and therefore have been widely used in industrial CTR prediction tasks. However, the implicit feature interactions learned in DNNs cannot fully retain the complete representation capacity of the original and empirical feature interactions (e.g., cartesian product) without loss. For example, a simple attempt to learn the combination of feature A and feature B < A, B > as the explicit cartesian product representation of new features can outperform previous implicit feature interaction models including factorization machine (FM)-based models and their variations. This indicates there is still a big gap between explicit and implicit feature interaction models. However, to learn all the explicit feature interaction (cartesian product) representations requires a very large sample size along with N times of original parameter space (where N is quite large in most industrial applications). In this paper, we propose a Co-Action Network (CAN) to approximate the explicit pairwise feature interactions without introducing too many additional parameters. More specifically, giving feature A and its associated feature B, their feature interaction is modeled by learning two sets of parameters: 1) the embedding of feature A, and 2) a Multi-Layer Perceptron (MLP) to represent feature B. The approximated feature interaction can be obtained by passing the embedding of feature A through the MLP network of feature B. We refer to such pairwise feature interaction as feature co-action, and such a Co-Action Network unit can provide a very powerful capacity to fitting complex feature interactions. In addition, FM can be viewed as a special case of the CAN unit when the MLP is a single layer with only one output. Experimental results on public and industrial datasets show that CAN outperforms state-of-the-art CTR models and the cartesian product method. Moreover, CAN has been deployed in the display advertisement system in Alibaba, obtaining 12% improvement on CTR and 8% on Revenue Per Mille (RPM), which is a great improvement to the business. The code for experiments in this paper is open-sourced\footnotehttps://github.com/CAN-Paper/Co-Action-Network.

Traditional industry recommendation systems usually use data in a single domain to train models and then serve the domain. However, a large-scale commercial platform often contains multiple domains, and its recommendation system often needs to make click-through rate (CTR) predictions for multiple domains. Generally, different domains may share some common user groups and items, and each domain may have its own unique user groups and items. Moreover, even the same user may have different behaviors in different domains. In order to leverage all the data from different domains, a single model can be trained to serve all domains. However, it is difficult for a single model to capture the characteristics of various domains and serve all domains well. On the other hand, training an individual model for each domain separately does not fully use the data from all domains. In this paper, we propose the Star Topology Adaptive Recommender (STAR) model to train a single model to serve all domains by leveraging data from all domains simultaneously, capturing the characteristics of each domain, and modeling the commonalities between different domains. Essentially, the net- work of each domain consists of two factorized networks: one centered network shared by all domains and the domain-specific network tailored for each domain. For each domain, we combine these two factorized networks and generate a unified network by element-wise multiplying the weights of the shared network and those of the domain-specific network, although these two factorized networks can be combined using other functions, which is open for further research. Most importantly, STAR can learn the shared network from all the data and adapt domain-specific parameters according to the characteristics of each domain. The experimental results from production data validate the superiority of the proposed STAR model. Since late 2020, STAR has been deployed in the display advertising system of Alibaba, obtaining 8.0% improvement on CTR and 6.0% increase on RPM (Revenue Per Mille).

One of the difficulties of conversion rate (CVR) prediction is that the conversions can delay and take place long after the clicks. The delayed feedback poses a challenge: fresh data are beneficial to continuous training but may not have complete label information at the time they are ingested into the training pipeline. To balance model freshness and label certainty, previous methods set a short waiting window or even do not wait for the conversion signal. If conversion happens outside the waiting window, this sample will be duplicated and ingested into the training pipeline with a positive label. However, these methods have some issues. First, they assume the observed feature distribution remains the same as the actual distribution. But this assumption does not hold due to the ingestion of duplicated samples. Second, the certainty of the conversion action only comes from the positives. But the positives are scarce as conversions are sparse in commercial systems. These issues induce bias during the modeling of delayed feedback. In this paper, we propose DElayed FEedback modeling with Real negatives (DEFER) method to address these issues. The proposed method ingests real negative samples into the training pipeline. The ingestion of real negatives ensures the observed feature distribution is equivalent to the actual distribution, thus reducing the bias. The ingestion of real negatives also brings more certainty information of the conversion. To correct the distribution shift, DEFER employs importance sampling to weigh the loss function. Experimental results on industrial datasets validate the superiority of DEFER. DEFER have been deployed in the display advertising system of Alibaba, obtaining over 6.0% improvement on CVR in several scenarios. The code and data in this paper are now open-sourced (https://github.com/gusuperstar/defer.git).