Finding Support Documents with a Logistic Regression Approach

Entity retrieval finds the relevant results for a user’s information needs at a finer unit called “entity”. To retrieve such entity, people usually first locate a small set of support documents which contain answer entities, and then further detect the answer entities in this set. In the literature, people view the support documents as relevant documents, and their findings as a conventional document retrieval problem. In this paper, we will state that finding support documents and that of relevant documents, although sounds similar, have important differences. Further, we propose a logistic regression approach to find support documents. Our experiment results show that the logistic regression method performs significantly better than a baseline system that treat the support document finding as a conventional document retrieval problem.