Efficient dialogue strategy to find users' intended items from information query results
Kazunori Komatani, Tatsuya Kawahara, Ryosuke K. Ito, Hiroshi G. Okuno · 2002
We address a dialogue framework that narrows down the user's query results obtained by an information retrieval system. The follow-up dialogue to constrain query results is significant especially with the speech interfaces such as telephones because a lot of query results cannot be presented to the user. The proposed dialogue framework generates guiding questions based on an information theoretic criterion to eliminate retrieved candidates by a spontaneous query without assuming a semantic slot structure. We first describe its concept on general information query tasks, and then deal with a query task on the appliance manual where structured task knowledge is available. A hierarchical confirmation strategy is proposed by making use of a tree structure of the manual, and then three cost functions for selecting optimal question nodes are compared. Experimental evaluation demonstrates that the proposed system helps users find their intended items more efficiently.