192 as the most useful attribute for confirming the target case and asks the user what location she prefers. When asked to explain the relevance of location, Top Case points out that if the preferred location is Tyrol, this will increase the sim ilarity of the target case from 0. 28 to 0.44 and eliminate 866 of the 972 competing cases. When the user chooses Alps instead as the pre- ferred location, the target case changes to Case 574, but now there are only 82 competing cases. The user's answers to the next two questions are enough for Top Case to confirm Case 574 as the recommended case. When asked to explain its recommendation, Top Case points out that the recommended case differs from the user's query only in price and that her preferences with respect to the remaining attributes can- not affect the recommendation 4.3. Discussion a known limitation of similarity-based retrieval is that the most sim ilar case may not be the one that is most acceptable to the user(e.g McSherry, 2003b). It must also be recognised that the case recom- mended by Top Case may not be acceptable to the user even though guaranteed to remain the most similar case no matter how she chooses to extend her query. In future research we plan to investigate an approach to addressing this issue in which the dialogue is allowed to continue beyond the initial recommendation so that the user can extend or revise her query to include one or more constraints that must be satisfied It is worth noting that all three cases presented by Top Case in response to the user's initial query have the same values for holiday type and month; in fact there are 19 other cases that exactly match the user's initial query. Inseparability of competing cases is a com mon problem associated with incomplete queries(McSherry, 2002c) that highlights the importance of Top Case having the ability to take the initiative to help users discriminate between alternatives that are equally good in terms of their initial requirements. Related Work Allowing the user to enter an initial query to be incrementally extended is a feature that Top Case shares with mixed-initiative CCbr tools for fault diagnosis such as Na code (Aha et al192 D. MCSHERRY as the most useful attribute for confirming the target case and asks the user what location she prefers. When asked to explain the relevance of location, Top Case points out that if the preferred location is Tyrol, this will increase the sim￾ilarity of the target case from 0.28 to 0.44 and eliminate 866 of the 972 competing cases. When the user chooses Alps instead as the pre￾ferred location, the target case changes to Case 574, but now there are only 82 competing cases. The user’s answers to the next two questions are enough for Top Case to confirm Case 574 as the recommended case. When asked to explain its recommendation, Top Case points out that the recommended case differs from the user’s query only in price and that her preferences with respect to the remaining attributes can￾not affect the recommendation. 4.3. Discussion A known limitation of similarity-based retrieval is that the most sim￾ilar case may not be the one that is most acceptable to the user (e.g. McSherry, 2003b). It must also be recognised that the case recom￾mended by Top Case may not be acceptable to the user even though it is guaranteed to remain the most similar case no matter how she chooses to extend her query. In future research we plan to investigate an approach to addressing this issue in which the dialogue is allowed to continue beyond the initial recommendation so that the user can extend or revise her query to include one or more constraints that must be satisfied. It is worth noting that all three cases presented by Top Case in response to the user’s initial query have the same values for holiday type and month; in fact there are 19 other cases that exactly match the user’s initial query. Inseparability of competing cases is a com￾mon problem associated with incomplete queries (McSherry, 2002c) that highlights the importance of Top Case having the ability to take the initiative to help users discriminate between alternatives that are equally good in terms of their initial requirements. 5. Related Work Allowing the user to enter an initial query to be incrementally extended is a feature that Top Case shares with mixed-initiative CCBR tools for fault diagnosis such as NaCoDAE (Aha et al