Research On Recommendation Algorithm Based On Category Combination And Semantic Relationship

With the rapid development of information technology in recent years,the data resources in the Internet are also growing at an exponential rate.The recommendation system can analyze the users' historical behavior from the massive information,and help users to find the content they need quickly and accurately.Collaborative filtering is the most effective algorithm in the recommendation system,which recommends items for target users based on neighbor users' preferences.But the recommendation performance of this method is still limited by cold start and data sparse.So more and more researchers proposed targeted solutions by adding a variety of information of users and items into the recommendation system.Two kinds of recommendation algorithms are proposed by using item category information and studying semantic relationship in knowledge graph separately,aiming to improve the recommendation accuracy.The association between categories is only considered or organized into flat or hierarchical structures in traditional category-driven methods,and the complex relationships between items and categories are neglected.Therefore,a random walk recommendation algorithm based on category combination space is proposed.Firstly,an item category combination space representing by Haas diagram is constructed,and the one-to-many relationship between the item and the category is mapped into one-to-one.Then,the semantic relationships and the semantic distances between category combinations are defined,which can describe the changes on user dynamic preferences more qualitatively and quantitatively.Afterwards,the random walk is used to establish the user personalized category preference model combining with various information.The information includes the user's preferences changes,user jump behaviors,jump times,time and scores,and more,all of these are obtained on user browsing process under category combination space.Finally,the user-based collaborative filtering recommendation is completed according to the user's personalized preference.Experiments on real data sets show that the proposed algorithm is superior to the traditional category-driven recommendation algorithms in sorting recommendation.In order to get the semantic structure relationships between users and items in the knowledge graph,a temporal recommendation model based on semantic relationships is proposed.This method can mine the potential semantic associations by relation extraction.Firstly,a lightweight knowledge graph based on single-domain knowledge is designed.Then the interactions between/within users and items are defined as three semantic structure relationships,including the social relationships between users,explicit interactions between users and items,and implicit information between items.Afterwards,the influence of popular items is added into the recommendation.The users' dynamic preferences and the items' short-term characteristics are obtained by deep learning technology.The experiment results show that the prediction accuracy of the recommendation algorithm is enhanced,and the model is more efficient and applicable without additional knowledge.