An approach for intention-driven, dialogue-based Web search
Web search engines facilitate the achievement of Web-mediated tasks, including information retrieval, Web page navigation, and online transactions. These tasks often involve goals that pertain to multiple topics, or domains. Current search engines are not suitable for satisfying complex, multi-domain needs due to their lack of interactivity and knowledge. This thesis presents a novel intention-driven, dialogue-based Web search approach that uncovers and combines users' multi-domain goals to provide helpful virtual assistance. The intention discovery procedure uses a hierarchy of Partially Observable Markov Decision Process-based dialogue managers and a backing knowledge base to systematically explore the dialogue's information space, probabilistically refining the perception of user goals. The search approach has been implemented in IDS, a search engine for online gift shopping. A usability study comparing IDS-based searching with Google-based searching found that the IDS-based approach takes significantly less time and effort, and results in higher user confidence in the retrieved results.