A Combined Linkage Model Based on Knowledge Graphs and Recommender Systems

To provide rich and structured knowledge information for recommender system items, previous methods focus on linking items of recommender systems with Knowledge Graphs (KGs) without considering the impact of linkage ratio on recommendation performance. To bridge this gap, we propose a combined linkage model (CLM) based on existing KGs for generating more linkage data. In the CLM, we design a combination of entity-item linkage with Freebase as the main component and Yago as the secondary component. However, due to Freebase API has been closed forever by Google, researchers cannot retrieve KGs entities with item titles as queries. A Reverse Verification algorithm is proposed to produce linked entity-item ID pairs with Freebase. Based on linked entity-item ID pairs with Freebase and Yago, we compared the linkage ratio with existing research and the recommended performance of four KG-based recommendation methods. The experiments showed that the performance of four methods using generated datasets is better than that using the original datasets, which can provide users with more accurate recommendation results.