Deep Learning-Based Sequential Recommender Systems: Concepts, Algorithms, and Evaluations

Abstract

In the field of sequential recommendation, deep learning methods have received a lot of attention in the past few years and surpassed traditional models such as Markov chain-based and factorization-based ones. However, DL-based methods also have some critical drawbacks, such as insufficient modeling of user representation and ignoring to distinguish the different types of interactions (i.e., user behavior) among users and items. In this view, this survey focuses on DL-based sequential recommender systems by taking the aforementioned issues into consideration. Specifically, we illustrate the concept of sequential recommendation, propose a categorization of existing algorithms in terms of three types of behavioral sequence, summarize the key factors affecting the performance of DL-based models, and conduct corresponding evaluations to demonstrate the effects of these factors. We conclude this survey by systematically outlining future directions and challenges in this field.Index Terms-sequential recommendation, session-based recommendation, sequential data, deep learning, influential factors ! • Hui Fang is with the 1. We searched on arXiv.org with keywords related to the sequential recommendation and DL techniques in March 2019.2. Fig. 3: A schematic diagram of the sequence recommendation. c i : behavior type, o i : behavior object. A behavior is represented by a 2-tuple. A behavior sequence (i.e., behavior trajectory) is a time ordered list of 2-tuples.The input behavior sequence {a 1 , a 2 , a 3 , ..., a t } is polymorphic, which can thus be divided into three types: experience-based, transaction-based and interaction-based behavior sequence, whose details are introduced as follows:

TRec: Sequential Recommender Based On Latent Item Trend Information

“…Sequential Recommendation System (SRS) is a hot research topic in the research area of recommendation systems [1]. The main-stream methods bifurcate into two categories: traditional non-neural approaches, and neural-based or deep-learning based approaches.…”

“…Traditional Approaches Most traditional recommender systems utilize collaborative filtering based methods. Specifically, they incline to utilize a user's historical interactions to learn her static preference with the assumption that all user-item interactions in the historical sequence are equally important [1]. In real world scenario, user behavior is usually not determined merely upon her all-time preference, a basic observation is that the interest of user changes all the time.…”

“…Recommender system is born to address such information overload problems. [1]. Collaborative filtering-based (CF) [2] approaches and content-based (CB) approaches [3] are two groups of mainstream traditional recommendation methods.…”

“…For example, based on the former kungfu lover's interaction history, a traditional factorization-based recommender might give a high prediction score on a target kung-fu movie. As a result, however, we can see that both the classical matrix factorization approaches [1], [6], [7] and the recent deep learning based approaches fail to find a way to model items' changing popularity. However in the context of sequential recommender systems, the item is still regarded as a static representation.…”

DeepAPF: Deep Attentive Probabilistic Factorization for Multi-site Video Recommendation

“…Most of recommender systems use one-site viewing data to predict user preferences on videos [Zhou et al, 2010;Qian et al, 2014;Bu et al, 2016]. In particular, some recent works apply Deep Neural Networks (DNNs) into the recommendation methods Zheng et al, 2017;Chen et al, 2017;Wang et al, 2017;Zhu et al, 2018;Zhou et al, 2018;Gao et al, 2019;Fang et al, 2019]. For example, Covington et al [Covington et al, 2016] propose a deep neural network for YouTube video recommendation.…”