Collaborative Filtering Based on Semantic Distance Among Items

• Salmo M. S. Júnior USP
• Marcelo G. Manzato USP

Resumo

This paper proposes a new method to compute the semantic distance among items in collaborative filtering based on k-nearest neighbors. This approach predicts the rating that a user u would give to an item i calculating the similarity between i and other items rated by u. This items’ similarity is obtained using a semantic distance metric proposed in this paper. The technique exploits ontologies available on the Web through the Linked Open Data. This is possible because they have semantic descriptions, structured by links, that define a knowledge domain. The equation to calculate the semantic distance is an extension of a related work. We propose to assign weight to links to show the specificity of item’s categories. Our proposal was evaluated with a movies dataset and it was shown that significant improvements can be achieved when compared to the baseline without weighted links.