Personalized recommendation method based on collaborative knowledge graph

Abstract

The invention provides a personalized recommendation method based on a knowledge graph, and the method comprises the steps: S1, combining historical interaction data of users in a user set and articles in an article set and an original knowledge graph into a collaborative knowledge graph, in the collaborative knowledge graph, connecting entities corresponding to the user and the interactive article thereof in the knowledge graph by using directed edges; S2, expressing each entity and relationship in the collaborative knowledge graph by using quaternion vectors; S3, embedding the quaternion ofthe entity into a path along the collaborative knowledge graph by adopting an attention mechanism to perform preference propagation and aggregation; S4, constructing a preference score prediction function to calculate preference scores of each user and different articles on the basis of the collaborative knowledge graph after preference propagation and aggregation are completed; S5, adopting a loss function to jointly optimize an embedding and preference score prediction function of the collaborative knowledge graph; and S6, predicting the preference score of the user for the new article by adopting the optimized preference score prediction function to obtain a new article recommendation list for the user.

Description

technical field [0001] The present invention relates to the field of multimedia and natural language processing, specifically to the field of personalized recommendation of knowledge graphs, and more specifically to a method of personalized recommendation based on collaborative knowledge graphs. Background technique [0002] With the rapid development of the Internet, the amount of data available to people is increasing exponentially. In the face of information overload, it is difficult for users to select items that they are really interested in from many candidate items. In order to improve user experience and provide personalized suggestions to users, recommender systems are widely used. The most commonly used algorithm in the recommendation system is the collaborative filtering algorithm, which calculates user similarity based on historical user-item interaction information to make recommendations. This algorithm is usually effective, but its performance is greatly redu...

AI Technical Summary

Problems solved by technology

Although the above methods have improved the performance of recommendation algorithms, the existing recommendation algorithms based on knowledge graphs use real vectors to represent users, items, entities, and relationships, which have limited expressive power and do not have inherent antisymmetry. Limit the modeling between entities and relationships in the knowledge graph, affecting the final recommendation effect

Images