Research On Session-Based Recommendation With Neural Networks

With the development of Internet technology,recommender systems have been widely used in e-commerce platforms and media streaming websites.At present,most recommender systems generate personalized recommendation results based on users' identity and long-term historical behavior.However,in many real application scenarios,users' identity and long-term historical behavior data are not available.In this case,the traditional recommendation algorithm will not work.We generate the recommendation results merely based on the user's click sequence in the current session,which is called session-based recommendation.Due to the uncertainty of user behavior and the limited information provided by browser sessions,session-based recommendation is still a challenging problem.Previous dominant methods to solve this problem are recurrent neural networks(RNN)based models.However,RNN's inherent sequential operation precludes parallelization and leads to expensive time cost.Recently,attention mechanisms have shown significant improvement on this issue.However,none of the existing attention-based methods explicitly take advantage of the position information,context information and local features in a sequence.In order to solve these problems,we propose some novel models based on the existing work.The main innovations of this thesis are as follows:1.We propose a position-aware context attention(PACA)model,which improves the recommendation performance by taking into account both the position information and the context information of items.PACA introduces positional vectors to model the position information and utilizes a pooling function to generate the context feature vectors.Extensive experiments on two real-world datasets show the effectiveness of our model.2.We propose a hierarchical convolutional attention network(HiCAN)model to further explore the local feature of user behavior.In session-based recommendation problems,we should not only consider the position information of user behavior sequence,but also take the local feature into account.We think that a user's several continuous clicked items can be used to extract common features.Our HiCAN combines the advantages of convolutional neural networks and attention networks.It can simultaneously capture global and local context dependencies and temporal information in a sequence.To our best knowledge,this is the first work to leverage the advantages of convolutional neural networks in session-based recommendation field.Extensive experiments on three real-world datasets also show the effectiveness of this model.