Research On Session-based Recommendation With Deep Learning

The recommender system is one of the most important applications in the big data era.In this background,information overload is a normal state,both infor-mation providers and users face enormous challenges.As information providers,it is a difficult task to display the vast amount of information they stored to their users in a targeted manner.And as information users,how to find what they need from the massive data of the Internet is also a big challenge.The recommender system is an important.tool to help information users find the content they are interested in more efficiently.The larger the information resources,the greater the role of the recommender system.The core of the recommender system is recommendation algorithms.Most recommender system algorithms usually as-sume that user-related logs are available,such as the user's identity,clicked items and browsing logs.However,in many cases,the user's identity is invisible to the recommendation algorithm.Therefore,traditional personalized recommendation algorithms based on a specific user identity are no longer applicable.An alterna-tive recommendation system algorithm is called session-based recommendation.Given e-commerce scenarios that user profiles are invisible,session-based recom-mendation is proposed to generate recommendation results from short sessions.At present,the most powerful algorithms to solve the session-based recom-mendation problem are recurrent neural networks(RNN)with gated recurrent units(GRU).Take e-commerce websites as examples,this kind of model considers the item sequence clicked by a user as the initial input of RNN,and generates recommendations based on them.Then the user might click one of the recom-mendations,which is fed into RNN next,and the successive recommendations are produced based on the whole previous clicks.This process is completed until we meet the need of users or users stop browsing.Although existing researches which are based on RNN are more effective than most traditional methods,such as item-KNN,Markov chain,but these work only considers the user's sequential behavior in the current session,whereas the user's main purpose in the current session is not emphasized.In this paper,we propose a novel neural networks model,i.e.,Neural Attentive Recommendation Machine(NARM),to tackle this problem.Specifically,we explore a hybrid encoder with an attention mechanism to model the user's sequential behavior and capture the user's main purpose in the current session,which are combined as a unified session representation lat-er.We then compute the recommendation scores for each candidate item with a bi-linear matching scheme based on this unified session representation.We train NARM by jointly learning the item and session representations as well as their matchings.We carried out extensive experiments on two benchmark datasets.Our exper-imental results show that NARM outperforms state-of-the-art baselines in terms of recall@20 and MRR@20 on both datasets.Furthermore,we also find that NARM achieves a significant improvement on long sessions,which demonstrates its advantages in modeling the user's sequential behavior and main purpose si-multaneously.