An item-item similarity approach based on linked open data semantic relationship

Nowadays, persons produce more data than they can process, leading to an overload of information. Due to this fact, we need to improve the systems of information processing and retrieval. Moreover, recommendation systems are important tools for leading to such an improvement. Recommendation systems based on collaborative filtering may have degraded performance due to the problem known as cold start. In an attempt to solve this problem, we should recommend items based on their content. The content-based recommendation systems exploit external data sources, such as web documents or the Linked Open Data, to compare item contents. However, a few proposals make use of the semantic relationship among the items. In this work, we exploit the semantic relationship provided by Linked Open Data (LOD) to search out automatically relevant item features and propose an approach for measuring the similarity between items. This similarity measure leads to a ranked list of similar items. To evaluate our approach, we performed experiments on two domains: museums and movies. Our approach produced competitive results compared to those approaches that manually define the relevant features.