원문정보
초록
영어
Recommender systems reduce the information overload by providing users’ “top-pick” recommendations. However, selecting the “best” out of good options can be more challenging than separating good and bad ones. Based on the of attraction effects frameworks, we examine the effectiveness of recommender systems including “worstpicks”. The addition of an unfavored item may alleviate consumers’ cognitive load and ease comparisons—all of which means the improved performance of recommendation structures. For empirical validation, we employed a randomized field experiment involving 475,339 unique users, 93,282 fashion products and 25,854,168 total instances of exposure to recommendations. The findings show that the attraction effect (AE) model outperformed the rational choice (RC) equivalent. AE model was more effective on PCs than over mobile phones. For male consumers, the AE model outcompeted the RC equivalent, but such difference was not detected among female shoppers. Based on these findings, we discuss the theoretical and practical implications.
목차
1. Introduction
2. Literature Review
2.1. Algorithmic advancements of recommender systems
2.2 Contextual and behavioral patterns reflected in the use of recommender systems
2.3. Effects of recommender systems on business performance
3. Choice Overload, Choice Set Quality, and Attraction Effects in Recommender Systems
4. Field Experiments
4.1. Experimental Design and Verification of Randomness
4.2. Recommendation Schemes using Item2Vec
5. Results of Field Experiments
5.1. Performance Comparison
5.2. Moderating Effects of Consumer Channel(Mobile/PC)
5.3. Moderating Effects of Consumer Demographics(Gender)
6. Discussion and Implications
Appendix (Training of Item2Vec)
References