Effective explanations of recommendations: user-centered design Export

@inproceedings{citeulike:2430455, abstract = {This paper characterizes general properties of useful, or Effective , explanations of recommendations. It describes a methodology based on focus groups, in which we elicit what helps moviegoers decide whether or not they would like a movie. Our results highlight the importance of personalizing explanations to the individual user, as well as considering the source of recommendations, user mood, the effects of group viewing, and the effect of explanations on user expectations.}, address = {New York, NY, USA}, author = {Tintarev, Nava and Masthoff, Judith}, booktitle = {RecSys '07: Proceedings of the 2007 ACM conference on Recommender systems}, citeulike-article-id = {2430455}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1297231.1297259}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1297231.1297259}, comment = {The recommender systems community is reaching a consensus that accuracy metrics such as mean average error (MAE), precision and recall, can only partially evaluate a recommender system [9]. User satisfaction and derivatives thereof such as serendipity [8], diversity [12] and trust [3] are increasingly seen as important. Explanations of recommendations can play an important role in improving the user experience. However, the definition of a good explanation is still largely open and depends on the general aim of the recommender system. Good explanations could help inspire user trust and loyalty, increase satisfaction, make it quicker and easier for users to find what they want, and persuade them to try or purchase a recommended item. A second limitation of the experiment of Herlocker et al. is that the explanations based on movie properties such as favorite actor/actress were not personalized for the participants, but rather for the main author of the paper [6]. This may have resulted in the relatively poor, yet significant, acceptance for explanations using this type of information. It would seem plausible that a movie feature such as favorite actor/actress is more important to some users than others, and that it would depend on each user's disposition toward the particular actor/actress. The features mentioned varied largely between subjects, the most commonly mentioned feature was good in its genre (e.g. this movie is funny, when talking about a comedy) followed by script complexity and mood. We confirmed that participants differ in the features they use when describing their favorite movies. Participants believed that their mood is likely to influence the genre they choose to see, and as a secondary effect, what features they consider important. Currently, we are developing a prototype which generates explanations for movie recommendations. The user model used in this system weighs the movies features elicited by our exploratory corpus analysis and focus groups according to specified user utility. Relevant meta-data is extracted from the Amazon e- Commerce Service (ECS) for this purpose. Textual recommendations are generated by a flexible natural language generation system. This flexibility allows us to modify parameters such as which features to mention, and how to describe them. }, doi = {10.1145/1297231.1297259}, isbn = {978-1-59593-730--8}, keywords = {evaluation, explanation, recommender, user-centered, user-model}, location = {Minneapolis, MN, USA}, pages = {153--156}, posted-at = {2009-10-05 17:03:56}, priority = {0}, publisher = {ACM}, title = {Effective explanations of recommendations: user-centered design}, url = {http://dx.doi.org/10.1145/1297231.1297259}, year = {2007} }

TY - CONF ID - citeulike:2430455 L3 - citeulike-article-id:2430455 N2 - This paper characterizes general properties of useful, or Effective , explanations of recommendations. It describes a methodology based on focus groups, in which we elicit what helps moviegoers decide whether or not they would like a movie. Our results highlight the importance of personalizing explanations to the individual user, as well as considering the source of recommendations, user mood, the effects of group viewing, and the effect of explanations on user expectations. CY - Minneapolis, MN, USA SN - 978-1-59593-730--8 AD - New York, NY, USA EP - 156 TI - Effective explanations of recommendations: user-centered design PB - ACM T2 - RecSys '07: Proceedings of the 2007 ACM conference on Recommender systems SP - 153 KW - evaluation KW - explanation KW - recommender KW - user-centered KW - user-model N1 - The recommender systems community is reaching a consensus that accuracy metrics such as mean average error (MAE), precision and recall, can only partially evaluate a recommender system [9]. User satisfaction and derivatives thereof such as serendipity [8], diversity [12] and trust [3] are increasingly seen as important. Explanations of recommendations can play an important role in improving the user experience. However, the definition of a good explanation is still largely open and depends on the general aim of the recommender system. Good explanations could help inspire user trust and loyalty, increase satisfaction, make it quicker and easier for users to find what they want, and persuade them to try or purchase a recommended item. A second limitation of the experiment of Herlocker et al. is that the explanations based on movie properties such as favorite actor/actress were not personalized for the participants, but rather for the main author of the paper [6]. This may have resulted in the relatively poor, yet significant, acceptance for explanations using this type of information. It would seem plausible that a movie feature such as favorite actor/actress is more important to some users than others, and that it would depend on each user's disposition toward the particular actor/actress. The features mentioned varied largely between subjects, the most commonly mentioned feature was good in its genre (e.g. this movie is funny, when talking about a comedy) followed by script complexity and mood. We confirmed that participants differ in the features they use when describing their favorite movies. Participants believed that their mood is likely to influence the genre they choose to see, and as a secondary effect, what features they consider important. Currently, we are developing a prototype which generates explanations for movie recommendations. The user model used in this system weighs the movies features elicited by our exploratory corpus analysis and focus groups according to specified user utility. Relevant meta-data is extracted from the Amazon e- Commerce Service (ECS) for this purpose. Textual recommendations are generated by a flexible natural language generation system. This flexibility allows us to modify parameters such as which features to mention, and how to describe them. AU - Tintarev, Nava AU - Masthoff, Judith PY - 2007/// UR - http://dx.doi.org/10.1145/1297231.1297259 ER -