Reconciling the accuracy-diversity trade-off in recommendations

In recommendation settings, there is an apparent trade-off between the goals of accuracy (to recommend items a user is most likely to want) and diversity (to recommend items representing a range of categories). As such, real-world recommender systems often explicitly incorporate diversity separately from accuracy. This approach, however, leaves a basic question unanswered: Why is there a trade-off in the first place? We show how the trade-off can be explained via a user’s consumption constraints—users typically only consume a few of the items they are recommended. In a stylized model we introduce, objectives that account for this constraint induce diverse recommendations, while objectives that do not account for this constraint induce homogeneous recommendations. This suggests that accuracy and diversity appear misaligned because standard accuracy metrics do not consider consumption constraints. Our model yields precise and interpretable characterizations of diversity in different settings, giving practical insights into the design of diverse recommendations.

Introduction. A large body of work in recommendations has developed methods to navigate an apparent trade-off between the goals of accuracy (to recommend items a user is most likely to want) and diversity (to recommend items from a range of categories) [1–23]. Real-world recommender systems use heuristics to directly incorporate diversity into recommendations [24, 25], and empirical evidence demonstrates that users prefer diverse recommendations [26–29]. A fundamental question remains: Why is there a trade-off in the first place? More specifically, why is “accuracy” unaligned with a user’s true preference for diversity? Without a principled understanding of the accuracy-diversity trade-off, attempts to diversify recommendations have difficulty moving beyond a heuristic basis—and difficulty articulating what they are accomplishing at a deeper level. In this work, we introduce and analyze a stylized model of recommendations that helps explain and reconcile the apparent accuracy-diversity trade-off.

Discussion / Conclusion. We introduce and analyze a stylized model that reconciles the apparent accuracy-diversity trade-off in recommendations. We characterize the diversity of optimal sets both when “optimality” captures and does not capture user consumption constraints. Broadly speaking, we show that the former naturally induces diversity while the latter results in homogeneity. Therefore, the apparent accuracy-diversity trade-off is partially due to traditional accuracy metrics not accounting for consumption constraints. Limitations and future work. A particular strength of our model is that we were able to derive precise and interpretable characterizations of diversity in many settings. One limitation of our work is that many of our results are asymptotic (i.e., n →∞). We expect that it is possible to obtain further results in our model for finite n. We gave Proposition 2 as one such example, and outline additional possible directions in Appendix B.