Best bets: thousands of queries in search of a client

Abstract

A number of applications require selecting targets for specific contents on the basis of criteria defined by the contents providers rather than selecting documents in response to user queries, as in ordinary information retrieval. We present a class of retrieval systems, called Best Bets, that generalize Information Filtering and encompass a variety of applications including editorial suggestions, promotional campaigns and targeted advertising, such as Google AdWords™. We developed techniques for implementing Best Bets systems addressing performance issues for large scale deployment as efficient query search, incremental updates and dynamic ranking.