Ranking on Data Manifold with Sink Points

Abstract

Ranking is an important problem in various applications. Many natural language processing tasks involve ranking a set of items. Sometimes we want the top items to be not only good individually but also diverse collectively. These ranking approaches are used to avoid redundant information as possible.  Manifold Ranking with Sink Points (MRSP), is used to address the diversity and importance in ranking. We applied MRSP on two application tasks, update summarization and query recommendation. This ranking approach gives a strong performance of MRSP as compared to already existing ranking approaches.