Research And Aplication Of Collaborative Filtering Algorithm Based On Knowledge Graph

With the continuous development and expansion of various online websites such as movies,music,articles,and shopping,it has become more difficult for users to find information,and some information cannot be accessed.How to mine interest information from a large amount of scattered information makes users more concerned.Although the traditional collaborative filtering recommendation algorithm can recommend the information of interest to the user from the user's perspective and the item's perspective,the data sparsity makes its recommendation effect not ideal,and the traditional recommendation system does not consider changes in user interest.In view of the above problems,this paper proposes an improved collaborative filtering algorithm,which uses knowledge graphs as a supplementary tool for item semantics,and incorporates time items that represent the decay of user interest.This article will conduct the following research work on data sparsity and interest attenuation:Aiming at the problem of data sparsity in traditional recommendation algorithms,this paper proposes to use knowledge graph as a semantic complement tool for items and apply it to collaborative filtering algorithms.In the research of knowledge graph,this article mainly involves two points,one is the construction of knowledge graph,and the second is to realize the application of knowledge graph to collaborative filtering algorithm.First of all,this article will study the construction of the film knowledge graph for the application field of the algorithm,mainly research the pattern layer and data layer of the knowledge graph.In the model layer,the corresponding concepts are extracted according to the knowledge features in the movie,the entities and relationships of the movie are divided,and the ontology database in the movie field is established.Then,the attributes and relationships of specific instances of the corresponding ontology are extracted from the related movie websites through the relationship attributes of the movies to supplement the data layer of the knowledge graph.Finally,the standard triplet data is imported into the neo4j graph database for storage,so as to realize the establishment of the knowledge graph.Next,according to the method of how the knowledge graph is applied to the collaborative filtering algorithm,this paper proposes a similarity negative sampling(TransE-SNS)training model based on the classic knowledge graph knowledge representation method-TransE model,through k-means clustering method Divide the entities of knowledge triples into clusters.When collecting negative examples,only the entities of the same cluster are used for replacement,thereby improving the training quality of negative examples.Through model training,the entities and relationships are embedded in the low-dimensional semantic vector space,which enriches the association of entities and relationships in the low-dimensional space.By calculating the semantic similarity of items and the similarity of item ratings,the two are merged to obtain the optimal ratio of knowledge graph embedded in collaborative filtering algorithm.Aiming at the problem that the user's interest is changing,this paper proposes an improved algorithm that takes into account the real-time problem of interest based on the application of knowledge graph to collaborative filtering algorithm.Newton's law of cooling function is used to fit the Ebbinghaus forgetting function.By introducing the time term of the thermal decay of the article applicable to this article,the article similar formula finally applied in this article is obtained.Through score prediction,the top N items with high similarity are used as neighbor items,and then the ranking is used to recommend the target users.Compared with the traditional collaborative filtering recommendation algorithm,the improved algorithm proposed in this paper has obvious improvement effect in application performance.