The Theory and Practice of Co-active Search

Retrieval systems implementing query-based search routinely encounter difficulties related to lexical ambiguity. Frequent attempts have been made in the fields of Information Retrieval (IR), Artificial Intelligence (AI) and Natural Language Processing (NLP) to address this problem, yet it remains unsolved. This thesis introduces a novel methodology known as co-active search that expresses the users of a system as the paramount resource for resolving query term ambiguity. A co-active system resolves the word sense of a given query by recording and analysing navigational behaviour. Captured data is modelled in a multi-dimensional feature space and evaluated through cluster analysis. The resulting sense clusters are then used to auto-categorise sets of search results, enabling the user to quickly select results matching the appropriate word sense for a particular query term. This method addresses lexical ambiguity in a democratic and adaptive fashion, so that a particular word sense can evolve as common usage dictates. Also introduced is the SENSAI system, a concrete implementation of the coactive search approach. Various components of the system are described, and a full user test documented. Experimental results from this test suggest that the most appropriate way to deal with lexical ambiguity in query-based search is to devolve the sense resolution process to the users of that system that the business of determining what words mean should be left in the hands of the people using them.