Asymmetric Item-Item Similarity Measure for Linked Open Data Enabled Collaborative Filtering

Abstract

The boom in Linked Open Data (LOD) has recently stimulated the research of a new generation of recommender systems—LOD-enabled recommender systems, in which the similarity measure for LOD is one of the core issues. The partitioned information content (PIC)-based semantic similarity (PICSS) is a newly developed symmetric similarity measure for LOD. However, recent studies have shown that asymmetric similarity measures are more effective than symmetric similarity measures in solving recommendation problems. In this paper we develop an asymmetric item-item similarity measure for LOD—the asymmetric PIC-based semantic similarity measure (APICSS), which applies our proposed two notions: the proportion of common PIC between two resources in the PIC of a resource and the PIC difference between two resources, on the basis of the notion of PIC. Experimental evaluation with the item-based collaborative filtering method on the MovieLens 100k dataset, the DBpedia 2016-04 release, and the DBpedia-MovieLens 100k dataset shows that our APICSS measure outperforms the PICSS measure in terms of both Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). The average RMSE accuracy has an increase of 1.58% and the maximum RMSE accuracy has an increase of 2.07%, compared to PICSS. The average MAE accuracy has an increase of 1.63% and the maximum MAE accuracy has an increase of 2.19%, compared to PICSS.