Entity Attribute Ranking Using Learning to Rank

Commercial search systems have recently begun using entity cards to support their ranked results lists. In this work, we propose an automated method for entity card building. We model this problem as a learning to rank approach applied to entity attributes. We introduce a new set of features that utilizes semantic information about entities as well as information from top-ranked documents using a generic search engine. The entity label is submitted as a search query to a general search engine and the top ranked documents are used to add context to the ranking process. The ranking model identifies and ranks the most important attributes (facts) of an entity. In order to experiment with our approach, we initially collected a dataset by exploiting Wikipedia infoboxes, whose ordering of attributes reflects the collaborative effort of a large community of Wikipedia users. Moreover, we are currently conducting a crowd-sourcing experiment to build a human labeled ground truth for our dataset. Our preliminary results demonstrate that our approach effectively replicates human ranking.