On Unexpectedness in Recommender Systems: Or How toExpect the Unexpected

Novelty and Diversity in Recommender Systems 2011.

计算机科学;

Panagiotis Adamopoulos ; Alexander Tuzhilin

Others  :  http://ceur-ws.org/Vol-816/paper2.pdf
PID  :  42534

学科分类：计算机科学（综合）

来源: CEUR

【 摘 要 】

Although the broad social and business success of recommendersystems has been achieved across several domains, there is still along way to go in terms of user satisfaction. One of the keydimensions for improvement is the concept of unexpectedness. Inthis paper, we propose a model to improve user satisfaction bygenerating unexpected recommendations based on the utilitytheory of economics. In particular, we propose a new concept ofunexpectedness as recommending to users those items that departfrom what they expect from the system. We define and formalizethe concept of unexpectedness and discuss how it differs from therelated notions of novelty, serendipity and diversity. We alsomeasure the quality of recommendations using specific metricsunder certain utility functions. Finally, we provide unexpectedrecommendations of high quality and conduct several experimentson a “real-world” dataset to compare our recommendation resultswith some other standard baseline methods. Our proposedapproach outperforms these baseline methods in terms of