A ranking framework for entity oriented search using Markov Random fields

In this work we present a general model for entity ranking that is based on the Markov Random Field approach for modeling various types of dependencies between the query and the entity. We show that this model actually extends existing approaches for entity ranking while aggregating all pieces of relevance evidences in a unified way. We evaluated the performance of our model using the INEX datasets. Our results show that our ranking model significantly out- performs leading INEX systems in the tracks of 2007 and 2008, and is equivalent to the best results achieved in the 2009 track.