|
Abstract:
|
Learning to rank is central to many information retrieval applications, ranging from document retrieval, expert search, computational advertising, to sentiment analysis. Taking document retrieval as an illustrating example, the primary goal is to rank a large collection of text documents given a text-based query, and retrieve the top-ranked documents. In this talk, I will present a query-specific learning to rank model that admits different ranking functions for different queries and also incorporates neighborhood structure among queries to improve the ranking performance. As opposed to most existing ranking models assuming a common ranking function for all queries, one key advantage of the proposed query-specific ranking model is that it can vary from query to query and thus accommodates the heterogeneity among different queries. The advantage is confirmed in a variety of simulated experiments as well as one large-scale real example on Yahoo! ranking challenge. If time permits, the asymptotic properties will also be discussed.
|
