Proposal on Generation of Synthetic Data for Book Recommendation Using Knowledge Graph Embedding

Abstract

This paper proposes a method of generating synthetic data for book recommendation based on knowledge graph embedding. Collaborative filtering (CF) is a key technology used in recommendation systems, but there are some problems, such as the cold-start problem, which are caused by the lack of ratings by new users and to items. In addition, privacy concerns have become a major issue in recent years. To solve these problems, the approach of generating synthetic data based on the statistical characteristics of a real data for machine learning and recommendation has been researched. The proposed method constructs a knowledge graph from the Goodreads dataset consisting of user's reading history and synthesizes a rating matrix by simulating the user's reading behavior based on the link prediction using TransE. The effectiveness of the proposed method is shown through a comparison with the actual rating matrix in a book recommendation task.