SDM

Abstract

Capturing users' precise preferences is a fundamental problem in large-scale recommender system. Currently, item-based Collaborative Filtering (CF) methods are common matching approaches in industry. However, they are not effective to model dynamic and evolving preferences of users. In this paper, we propose a new sequential deep matching (SDM) model to capture users' dynamic preferences by combining short-term sessions and long-term behaviors. Compared with existing sequence-aware recommendation methods, we tackle the following two inherent problems in realworld applications: (1) there could exist multiple interest tendencies in one session.(2) long-term preferences may not be effectively fused with current session interests. Long-term behaviors are various and complex, hence those highly related to the short-term session should be kept for fusion. We propose to encode behavior sequences with two corresponding components: multi-head self-attention module to capture multiple types of interests and long-short term gated fusion module to incorporate long-term preferences. Successive items are recommended after matching between sequential user behavior vector and item embedding vectors. Offline experiments on real-world datasets show the superior performance of the proposed SDM. Moreover, SDM has been successfully deployed on online large-scale recommender system at Taobao and achieves improvements in terms of a range of commercial metrics.

Itinerary-aware Personalized Deep Matching at Fliggy

“…During the ranking phase, a more complex ranking model is used to assign comparable scores to the candidate items, and top-items are recommended to users. After the big success achieved in computer vision domain [10,16,27], the deep neural network [9] has become the most popular technique in both academia and industry circles for solving the matching and ranking problems in RSs [6,7,11,17,19,32,34,35], due to its better representation and generalization ability compared with CF based solutions [19].…”

“…presented the ENCORE model [34], which is a neural item-relationship based recommender focusing on matching and presenting "complementary" items to users. Then, a sequential deep matching model, SDM [19], was proposed to capture users' dynamic preferences in the matching phase by combining users' short-term sessions and long-term behaviors. Currently, the idea of collaborative filtering is also connected with the deep neural network.…”

“…Then, the ranking phase is responsible for predicting probabilities of users interacting with these candidate items, and finally generates recommendations. The quality of candidate items retrieved in the matching phase plays a key role in the whole model [19]. Currently, inspired by the success of deep learning from computer vision domain [10] and natural language processing (NLP) [3], many deep matching networks (DMNs) that match items for users on the basis of deep neural networks have been successfully deployed in many industrial recommendation systems, due to their effectiveness and efficiency [6,17,19,32,34].…”

“…The quality of candidate items retrieved in the matching phase plays a key role in the whole model [19]. Currently, inspired by the success of deep learning from computer vision domain [10] and natural language processing (NLP) [3], many deep matching networks (DMNs) that match items for users on the basis of deep neural networks have been successfully deployed in many industrial recommendation systems, due to their effectiveness and efficiency [6,17,19,32,34]. However, without considering the challenges determined by business characteristics of OTPs, these networks become ineffective when the matching objects are travel-related items, which makes the personalized deep matching problem at Fliggy extremely challenging.…”

Group Recommendation via Self-Attention and Collaborative Metric Learning Model

“…In this paper, four evaluation indicators are used to evaluate the performance of the model: precision [31], recall [31], AUC [32] and NDCG [33]. Let K be the number of recommendation services where K is set to 5,10 and 15.…”