Research On Session-based Recommendation Algorithm With Neural Network

In recent years,with the rapid development of computer hardware,Internet,big data and other technologies,users are surrounded by massive amounts of information.It is very difficult for users to find the information they need in the growing information database.In order to allow users to spend a small amount of time in the system to find the items or information they need,the recommendation system came into being.The emergence of the recommendation system solves the problem of "information overload" to a great extent.It is a very effective method of information filtering.It can learn the user's interest preferences based on the user's relevant information or historical browsing records,so as to provide users with Recommend information and items of interest or need.This article focuses on related research on session information recommendation for user interaction with the system.The main work is as follows:(1)Session-based recommendation is an emerging topic in the field of recommendation,and in recent years has attracted great attention from academia and industry.The biggest advantage of session-based recommendation is that it does not need to mine the user's own information,but only needs to process the item click event sequence information during the user's interaction with the system.Aiming at the problem that the traditional session-based recommendation algorithm does not use the feature information of the item,this thesis proposes a new session-based recommendation algorithm-SR-I2V(Session-based Recommendation with Item2Vec),which first learns the embedding vector of items,then click on the item that has occurred to extract the intent feature vector by the progressive intent formula,and finally recommend by calculating the vector similarity.The SR-I2 V algorithm not only effectively solves the problems of cold start and data sparseness in the recommendation algorithm,but also the experimental results have been improved to a large extent compared to other recommendation algorithms,such as the recall rate and the reciprocal ranking of comments.The effectiveness and feasibility of the recommendation algorithm.(2)Since the SR-12 V recommendation algorithm uses a progressive intent formula to extract the user's intention,it cannot extract the user's main intention very accurately.In order to solve this problem,this thesis then proposes a novel recommendation algorithm,that is,a session-based recommendation algorithm fused with attention mechanism(SRA)to solve this problem.The algorithm explores a hybrid encoder with attention mechanism to model the user's sequence behavior and capture the user's main intention in the current session,and then merge it into a unified session representation.Then,based on this unified representation of session,a bilinear matching scheme is used to calculate a recommendation score for each candidate item.SRA is trained by jointly learning item and session representation and their matching items.We conducted extensive experiments on two public datasets.Experimental results show that SRA is better than the common baseline on both data sets.It proves the advantages of SRA in simultaneously modeling the user's sequence behavior and main intentions.