188 D MCSHERRY on the criteria we have shown to be essential to ensure that the recommendation will remain unchanged in any approach to attribute Our evaluation was based on the Travel case library (www ai-cbr. org), a standard benchmark containing more than 1,000 cases, and the pc case library (McGinty and Smyth, 2002). The results showed inn to be more effective in reducing dialogue length than any of the other attribute-selection strategies. Its performance on the pc ase library was close to optimal, reducing the number of questions asked by up to 63% and by 35% on average relative to a full-length query. It also gave the best performance on the Travel case library, reducing dialogue length by up to 63% and by 25% on average 4. Explanation in Top Case We now present an approach to explanation of the recommendation process in iNN in which explanations are automatically generated with no requirement for domain knowledge other than the similar ity knowledge and cases already available to the system. We demon- strate the approach in a mixed-initiative recommender system called Top Case which can explain the relevance of any question the user is asked in strategic terms, recognise when the dialogue can be safely terminated, and justify its recommendations on the grounds that any remaining attributes cannot affect the outcome. An example recom mendation dialogue based on a well-known case library in the travel domain is used to illustrate the approach 4.1. Explanation engineering An initial query entered by the user is incrementally extended in Top Case by asking the user to specify preferred values for attributes not mentioned in her initial query. On each recommendation cycle, the user is asked for the preferred value of the most useful attribute for confirming the target case and shown the competing cases that are now most similar to her query. The user can terminate the recommen- lation dialogue at any stage by selecting one of the cases she is shown as the product she prefers. Otherwise, query elicitation continues until Top Case has determined that its recommendation will be the same no matter how the user chooses to extend her query. At this point, the dialogue is terminated and the user is informed that the target case188 D. MCSHERRY on the criteria we have shown to be essential to ensure that the recommendation will remain unchanged in any approach to attribute selection. Our evaluation was based on the Travel case library (www. ai-cbr.org), a standard benchmark containing more than 1,000 cases, and the PC case library (McGinty and Smyth, 2002). The results showed iNN to be more effective in reducing dialogue length than any of the other attribute-selection strategies. Its performance on the PC case library was close to optimal, reducing the number of questions asked by up to 63% and by 35% on average relative to a full-length query. It also gave the best performance on the Travel case library, reducing dialogue length by up to 63% and by 25% on average. 4. Explanation in Top Case We now present an approach to explanation of the recommendation process in iNN in which explanations are automatically generated with no requirement for domain knowledge other than the similar￾ity knowledge and cases already available to the system. We demon￾strate the approach in a mixed-initiative recommender system called Top Case which can explain the relevance of any question the user is asked in strategic terms, recognise when the dialogue can be safely terminated, and justify its recommendations on the grounds that any remaining attributes cannot affect the outcome. An example recom￾mendation dialogue based on a well-known case library in the travel domain is used to illustrate the approach. 4.1. Explanation engineering An initial query entered by the user is incrementally extended in Top Case by asking the user to specify preferred values for attributes not mentioned in her initial query. On each recommendation cycle, the user is asked for the preferred value of the most useful attribute for confirming the target case and shown the competing cases that are now most similar to her query. The user can terminate the recommen￾dation dialogue at any stage by selecting one of the cases she is shown as the product she prefers. Otherwise, query elicitation continues until Top Case has determined that its recommendation will be the same no matter how the user chooses to extend her query. At this point, the dialogue is terminated and the user is informed that the target case