Intelligent methods for information access in context: The role of topic descriptors and discriminators

Abstract

Successful access to information sources on the Web depends on effective methods for identifying the needs of a user and making relevant information resources available when needed. This paper formulates a theoretical framework for the study of context-drivenWeb search and proposes new methods for learning query terms based on the user task. These methods use an incrementally-retrieved, topic-dependent selection of Web documents for term-weight reinforcement reflecting the aptness of the terms in describing and discriminating the topic of the user context. Based on this framework, we propose an incremental search algorithm for information retrieval agents that has the potential to improve significantly over the traditional IR techniques. The new algorithm learns new descriptors by searching for terms that tend to occur often in relevant documents, and learns good discriminators by identifying terms that tend to occur only in the context of the given topic. We discuss the technical challenges posed by this new framework, outline our agent system architecture, and present an evaluation of the proposed techniques.