When is Concentration a Good Thing? The Dynamic Effect of Recommendation Concentration on Consumer Purchase

The effect of recommendation concentration is critical to profit maximization of e-commerce platforms. However, recommendation with high concentration level may not always be a good thing to purchase behavior which researchers know little about. To fill this gap, we develop a stochastic model to capture the state-dependent effect of recommendation concentration on purchase behavior. Specifically, we introduce a hidden Markov model (HMM) to investigate the following research questions: (1) is there any hidden state that potentially governs the dynamic purchase journey? (2) If so, how will concentration-level factors affect purchase behavior given a consumer's hidden state? (3) How does recommendation concentration play a role in transferring consumers among different hidden states? Would concentration levels in different recommendation systems (popularity recommender, content-based recommender and collaborative filtering recommender) make a difference? Based on Bayesian estimation of the model with a granular click-stream dataset from a vertical e-commerce platform of liquors, we identify three consumer states governing the purchase process, which are termed as "awareness'', "interest'' and "desire'' respectively. Our results show that the effects of recommendation concentration on purchase are state-dependent. Concentrated recommendation with focal product is effective to boost the focal product's conversion rate in awareness state, whereas it would depress the conversion rate of the focal product in interest state. In addition, recommendation concentration has a significant impact on moving consumer into a deeper state in which the intrinsic purchase intention is higher. Furthermore, through counterfactual simulations of different concentration strategies, our findings offer implications for e-commerce platforms on when and how recommendation concentration should be introduced dynamically, which stimulates more purchase actions eventually