Junfeng Ge

Xu Yuan, Chen Xu, Qiwei Chen et al.

In this era of information explosion, a personalized recommendation system is convenient for users to get information they are interested in. To deal with billions of users and items, large-scale online recommendation services usually consist of three stages: candidate generation, coarse-grained ranking, and fine-grained ranking. The success of each stage depends on whether the model accurately captures the interests of users, which are usually hidden in users' behavior data. Previous research shows that users' interests are diverse, and one vector is not sufficient to capture users' different preferences. Therefore, many methods use multiple vectors to encode users' interests. However, there are two unsolved problems: (1) The similarity of different vectors in existing methods is too high, with too much redundant information. Consequently, the interests of users are not fully represented. (2) Existing methods model the long-term and short-term behaviors together, ignoring the differences between them. This paper proposes a Hierarchical Multi-Interest Co-Network (HCN) to capture users' diverse interests in the coarse-grained ranking stage. Specifically, we design a hierarchical multi-interest extraction layer to update users' diverse interest centers iteratively. The multiple embedded vectors obtained in this way contain more information and represent the interests of users better in various aspects. Furthermore, we develop a Co-Interest Network to integrate users' long-term and short-term interests. Experiments on several real-world datasets and one large-scale industrial dataset show that HCN effectively outperforms the state-of-the-art methods. We deploy HCN into a large-scale real world E-commerce system and achieve extra 2.5\% improvements on GMV (Gross Merchandise Value).

Qiwei Chen, Yue Xu, Changhua Pei et al.

Recent studies on Click-Through Rate (CTR) prediction has reached new levels by modeling longer user behavior sequences. Among others, the two-stage methods stand out as the state-of-the-art (SOTA) solution for industrial applications. The two-stage methods first train a retrieval model to truncate the long behavior sequence beforehand and then use the truncated sequences to train a CTR model. However, the retrieval model and the CTR model are trained separately. So the retrieved subsequences in the CTR model is inaccurate, which degrades the final performance. In this paper, we propose an end-to-end paradigm to model long behavior sequences, which is able to achieve superior performance along with remarkable cost-efficiency compared to existing models. Our contribution is three-fold: First, we propose a hashing-based efficient target attention (TA) network named ETA-Net to enable end-to-end user behavior retrieval based on low-cost bit-wise operations. The proposed ETA-Net can reduce the complexity of standard TA by orders of magnitude for sequential data modeling. Second, we propose a general system architecture as one viable solution to deploy ETA-Net on industrial systems. Particularly, ETA-Net has been deployed on the recommender system of Taobao, and brought 1.8% lift on CTR and 3.1% lift on Gross Merchandise Value (GMV) compared to the SOTA two-stage methods. Third, we conduct extensive experiments on both offline datasets and online A/B test. The results verify that the proposed model outperforms existing CTR models considerably, in terms of both CTR prediction performance and online cost-efficiency. ETA-Net now serves the main traffic of Taobao, delivering services to hundreds of millions of users towards billions of items every day.

Qiwei Chen, Changhua Pei, Shanshan Lv et al.

Click-Through Rate (CTR) prediction is one of the core tasks in recommender systems (RS). It predicts a personalized click probability for each user-item pair. Recently, researchers have found that the performance of CTR model can be improved greatly by taking user behavior sequence into consideration, especially long-term user behavior sequence. The report on an e-commerce website shows that 23\% of users have more than 1000 clicks during the past 5 months. Though there are numerous works focus on modeling sequential user behaviors, few works can handle long-term user behavior sequence due to the strict inference time constraint in real world system. Two-stage methods are proposed to push the limit for better performance. At the first stage, an auxiliary task is designed to retrieve the top-$k$ similar items from long-term user behavior sequence. At the second stage, the classical attention mechanism is conducted between the candidate item and $k$ items selected in the first stage. However, information gap happens between retrieval stage and the main CTR task. This goal divergence can greatly diminishing the performance gain of long-term user sequence. In this paper, inspired by Reformer, we propose a locality-sensitive hashing (LSH) method called ETA (End-to-end Target Attention) which can greatly reduce the training and inference cost and make the end-to-end training with long-term user behavior sequence possible. Both offline and online experiments confirm the effectiveness of our model. We deploy ETA into a large-scale real world E-commerce system and achieve extra 3.1\% improvements on GMV (Gross Merchandise Value) compared to a two-stage long user sequence CTR model.

Yufei Feng, Yu Gong, Fei Sun et al.

Recommender systems (RS) work effective at alleviating information overload and matching user interests in various web-scale applications. Most RS retrieve the user's favorite candidates and then rank them by the rating scores in the greedy manner. In the permutation prospective, however, current RS come to reveal the following two limitations: 1) They neglect addressing the permutation-variant influence within the recommended results; 2) Permutation consideration extends the latent solution space exponentially, and current RS lack the ability to evaluate the permutations. Both drive RS away from the permutation-optimal recommended results and better user experience. To approximate the permutation-optimal recommended results effectively and efficiently, we propose a novel permutation-wise framework PRS in the re-ranking stage of RS, which consists of Permutation-Matching (PMatch) and Permutation-Ranking (PRank) stages successively. Specifically, the PMatch stage is designed to obtain the candidate list set, where we propose the FPSA algorithm to generate multiple candidate lists via the permutation-wise and goal-oriented beam search algorithm. Afterwards, for the candidate list set, the PRank stage provides a unified permutation-wise ranking criterion named LR metric, which is calculated by the rating scores of elaborately designed permutation-wise model DPWN. Finally, the list with the highest LR score is recommended to the user. Empirical results show that PRS consistently and significantly outperforms state-of-the-art methods. Moreover, PRS has achieved a performance improvement of 11.0% on PV metric and 8.7% on IPV metric after the successful deployment in one popular recommendation scenario of Taobao application.

Yingqiang Ge, Shuchang Liu, Ruoyuan Gao et al.

As Recommender Systems (RS) influence more and more people in their daily life, the issue of fairness in recommendation is becoming more and more important. Most of the prior approaches to fairness-aware recommendation have been situated in a static or one-shot setting, where the protected groups of items are fixed, and the model provides a one-time fairness solution based on fairness-constrained optimization. This fails to consider the dynamic nature of the recommender systems, where attributes such as item popularity may change over time due to the recommendation policy and user engagement. For example, products that were once popular may become no longer popular, and vice versa. As a result, the system that aims to maintain long-term fairness on the item exposure in different popularity groups must accommodate this change in a timely fashion. Novel to this work, we explore the problem of long-term fairness in recommendation and accomplish the problem through dynamic fairness learning. We focus on the fairness of exposure of items in different groups, while the division of the groups is based on item popularity, which dynamically changes over time in the recommendation process. We tackle this problem by proposing a fairness-constrained reinforcement learning algorithm for recommendation, which models the recommendation problem as a Constrained Markov Decision Process (CMDP), so that the model can dynamically adjust its recommendation policy to make sure the fairness requirement is always satisfied when the environment changes. Experiments on several real-world datasets verify our framework's superiority in terms of recommendation performance, short-term fairness, and long-term fairness.

Wenhui Yu, Xiao Lin, Junfeng Ge et al.

Data sparsity is an inherent challenge in the recommender systems, where most of the data is collected from the implicit feedbacks of users. This causes two difficulties in designing effective algorithms: first, the majority of users only have a few interactions with the system and there is no enough data for learning; second, there are no negative samples in the implicit feedbacks and it is a common practice to perform negative sampling to generate negative samples. However, this leads to a consequence that many potential positive samples are mislabeled as negative ones and data sparsity would exacerbate the mislabeling problem. To solve these difficulties, we regard the problem of recommendation on sparse implicit feedbacks as a semi-supervised learning task, and explore domain adaption to solve it. We transfer the knowledge learned from dense data to sparse data and we focus on the most challenging case -- there is no user or item overlap. In this extreme case, aligning embeddings of two datasets directly is rather sub-optimal since the two latent spaces encode very different information. As such, we adopt domain-invariant textual features as the anchor points to align the latent spaces. To align the embeddings, we extract the textual features for each user and item and feed them into a domain classifier with the embeddings of users and items. The embeddings are trained to puzzle the classifier and textual features are fixed as anchor points. By domain adaptation, the distribution pattern in the source domain is transferred to the target domain. As the target part can be supervised by domain adaptation, we abandon negative sampling in target dataset to avoid label noise. We adopt three pairs of real-world datasets to validate the effectiveness of our transfer strategy. Results show that our models outperform existing models significantly.

Chen Xu, Quan Li, Junfeng Ge et al.

Features play an important role in the prediction tasks of e-commerce recommendations. To guarantee the consistency of off-line training and on-line serving, we usually utilize the same features that are both available. However, the consistency in turn neglects some discriminative features. For example, when estimating the conversion rate (CVR), i.e., the probability that a user would purchase the item if she clicked it, features like dwell time on the item detailed page are informative. However, CVR prediction should be conducted for on-line ranking before the click happens. Thus we cannot get such post-event features during serving. We define the features that are discriminative but only available during training as the privileged features. Inspired by the distillation techniques which bridge the gap between training and inference, in this work, we propose privileged features distillation (PFD). We train two models, i.e., a student model that is the same as the original one and a teacher model that additionally utilizes the privileged features. Knowledge distilled from the more accurate teacher is transferred to the student to improve its accuracy. During serving, only the student part is extracted and it relies on no privileged features. We conduct experiments on two fundamental prediction tasks at Taobao recommendations, i.e., click-through rate (CTR) at coarse-grained ranking and CVR at fine-grained ranking. By distilling the interacted features that are prohibited during serving for CTR and the post-event features for CVR, we achieve significant improvements over their strong baselines. During the on-line A/B tests, the click metric is improved by +5.0% in the CTR task. And the conversion metric is improved by +2.3% in the CVR task. Besides, by addressing several issues of training PFD, we obtain comparable training speed as the baselines without any distillation.