Commodity recommendation method based on meta-learning and knowledge graph

Abstract

The invention discloses a commodity recommendation method based on meta learning and a knowledge graph, and the method comprises the steps: 1) obtaining the historical interaction record data of a user commodity, and constructing a user-commodity interaction bipartite graph as the training data of a recommendation model; 2) obtaining commodity attributes and association features among the attributes, and constructing a background knowledge graph by using priori knowledge; 3) constructing a recommendation model based on meta learning and a knowledge graph, and selecting a proper loss function to optimize model parameters and feature vectors; and 4) using the recommendation model to carry out score prediction on the possibility that the non-interacted commodities generate interaction in the future, and finally obtaining a commodity recommendation result according to a score sequence. According to the method, the technology based on meta-learning and the knowledge graph is adopted, the background knowledge graph of commodity attributes is fused, and commodities which the user may be interested in can be accurately recommended for the user only by using a small amount of training data to train a recommendation system.

Description

technical field [0001] The invention belongs to the technical field of natural language processing, in particular to a product recommendation method based on meta-learning and knowledge graph. Background technique [0002] With the rapid development of the Internet and information computing, massive amounts of data have been derived. We have entered an era of information explosion, and massive amounts of information are generated every moment. However, this information is not all personal concern. It is also becoming more and more difficult to find useful information for oneself in the information of On the other hand, the producers of information are also racking their brains to send the information that users are interested in to users, and everyone has different interests, so a recommendation system that can achieve thousands of people and thousands of faces emerges as the times require. Simply put, the recommendation system is to determine the user's interest based on t...

AI Technical Summary

Problems solved by technology

However, there are often some problems in the recommendation method based on collaborative filtering: for new users who join the system, because there is no historical record in the database, their interests and preferences cannot be excavated and personalized recommendations cannot be accurately made, which is called cold Start-up problem; due to the large increase in the number of users and the number of products, the number of products that a single user can interact with is limited, making the interaction matrix contain a large number of blank elements, which is called the data sparse problem

The embedding-based method mainly uses the graph embedding method to carry out vector modeling of various entities and associations in the map, and then expands the semantic information expressed by the original products and users. However, this method focuses on building strict semantic associations. The model often ignores the attribute information of the node itself in the knowledge graph, so that it is impossible to accurately model the user's preference for the node content attribute, resulting in a decrease in the accuracy of the recommendation; the path-based method focuses on mining the various information between users and products based on the graph. Connect the relationship, extract the path carrying high-level information and input it into the prediction model, but because the choice of the path has a great impact on the final performance, and the definition of the path requires a lot of manual operation and certain domain knowledge, in practical situations , it is difficult to get the optimal connection path, so that the role of the knowledge map in the recommendation method cannot be fully utilized

Images