1st, 2nd, 1st and overall best at the ESWC14 LOD-enabled Recommender Systems Challenge

Petar Ristoski (University of Mannheim), Eneldo Loza Mencia and Heiko Paulheim (University of Mannheim, former KE member) achieved jointly two first positions (tasks 1 and 3), one second (task 2) and the overall Best Performing Approach among 9 participants with their solution (solution "helloWorld" in the final tables) at the ESWC 2014 Semantic Web Evaluation - Linked Open Data-enabled Recommender Systems Challenge.

Petar and Heiko receiving the award certificate from Valentina Presutti and Tommaso Di Noia

The goal was to predict the ratings of users given to books by maximally exploiting any information available in the Semantic Web about the books. Task 1 considered cold-start situations and the goal was to predict a 1-to-5 rating of user for a given book. Task 2 was concerned with providing a ranking of recommended books which a given user will likely like. Task 3 was similar to task 2, but with the additional constraint of returning recommendations with high diversity (w.r.t. authors and subjects of the books).

Our winning solution uses different features from DBpedia and RDF Book Mashup to create book recommendations with a variety of strategies, reaching from collaborative filtering and recommender system techniques to classical classification algorithms, which are combined using stacking and rank aggregation of the individual predictions. The approach is described in more detail in:
Petar Ristoski, Eneldo Loza Mencia, Heiko Paulheim. A Hybrid Multi-Strategy Recommender System Using Linked Open Data