Knowledge Graph Personalization by Estimation of Interest in Various Concepts

Abstract

In this paper, we propose a new personalization method for knowledge graphs using a graph convolutional neural network. As personalized items user interests are treated. Hobbies are used as an example of interest in this paper. Personalized data is obtained from labeled concepts rather than text. In the proposed method, the knowledge graph constructed from the Japanese WordNet and ALAGIN word co-occurrence dictionaries and GCN (Graph Convolutional Networks) are employed. Whether or a user is interested in each concept in the knowledge graph is estimated by three elements: interest data; graph structure of the hobby as an example; and GCN. As a result of the evaluation experiment, the proposed method obtained good results on predicting unknown subjects of interests.