Research On Session Sequence Based Recommendation Algorithm With Global Information

With the continuous development of e-commerce platforms,the number of online goods has reached an unprecedented scale,and the role of recommendation system is particularly prominent.Collaborative filtering algorithm is a widely used,classical and effective recommendation method.The main idea of collaborative filtering algorithm is to use the existing interactive data to predict the goods that current users may like.The existing collaborative filtering recommendation needs to maintain the interaction matrix between users and products,and the recommendation effect for new users and anonymous users is not good;In addition,the inconsistency between users' longterm and static preferences and their short-term and dynamic preferences also affects the accuracy of recommendation effect.In recent years,the commodity recommendation based on session sequence can make up for the above shortcomings of the traditional recommendation algorithm.The commodity recommendation based on session sequence predicts and recommends the next possible product item through the user's existing click sequence information.At present,the session sequence recommendation algorithm which can not only serve anonymous users,but also has better recommendation accuracy is an urgent problem to be studied.We propose a commodity session sequence recommendation algorithm based on global information for anonymous users.The main work contents are as follows:1.A global directed graph of items is proposed to represent the complex relationship between projects,in order to comprehensively consider the global item transfer relationship.Preprocess the original data set and import the processed data set into the Neo4 j graph database to build a global directed graph of commodity session sequence.The nodes in the graph are commodity items,and the directed arc between nodes represents the click order of commodity items,with the weight data corresponding to the click order on the arc.2.A scoring strategy of global preference propagation is proposed to obtain the global impact of multiple items in the session,in order to make full use of the historical items of the session.For multiple items in the session,and the global impact of the later item can be expanded by improving the parameters corresponding to item.When the recommended items are scored based on weight data and penetration data,cluster them first,and then the commodity items of each category are re labeled with appropriate size to avoid the adverse impact of too large difference in data values on the calculation.In the experimental part,various parameter combinations are used to verify the effectiveness of the preference propagation scoring strategy proposed in this paper.3.On Diginetica and Yoochoose standard datasets,SRGDG basis P@20compared with the traditional Item-KNN method,the recommended accuracy is improved by 30.25% and 6.12%,and basis MRR@20 the recommended accuracy improved by 33.88% and 15.04%.The experimental results show that the global directed graph search and scoring strategy proposed in this paper is effective.