Estimating Optimal Recommendation Set Sizes for Individual Consumers

Online consumers must burrow through vast piles of product information to find the best match to their preferences. This has boosted the popularity of recommendation agents promising to decrease consumers' search costs. Most recent work has focused on refining methods to find the best products for a consumer. The question of how many of these products the consumer actually wants to see, however, is largely unanswered.This paper develops a novel procedure based on signal detection theory to estimate the number of recommendable products. We compare it to a utility exchange approach that has not been empirically examined so far. The signal detection approach showed very good predictive validity in two laboratory experiments, clearly outperforming the utility exchange approach. Our theoretical findings, supported by the experimental evidence, indicate conceptual inconsistencies in the utility exchange approach. Our research offers significant implications for both theory and practice of modeling consumer choice behavior.