Store Recommendations in Offline Shopping : An Unbiased Ranking Approach Using Spatial Movement

In response to competition from online counterparts, offline shopping places attempt to enhance the personalized shopping experience of pedestrian shoppers by making recommendations. Given implicit feedback (e.g., store patronage behaviors), a crucial but underinvestigated issue is how to handle exposure bias for effective store recommendations. Exposure bias arises when customers are only exposed to a subset of stores, so that unvisited stores do not necessarily represent negative preferences. In offline settings, store exposure (i.e., how stores are presented to customers) is mainly driven by pedestrian customers’ movement processes in the space, in contrast to online platforms where system algorithms play a dominant role. To address this issue, we propose a novel recommendation method, unbiased movement-aware pairwise ranking (UMPR), which accounts for customers’ dynamic movement in the shopping space to achieve unbiased store recommendations. Specifically, we formulate an unbiased pairwise learning problem, propose a movement-aware recommendation model, and develop an alternating learning algorithm to optimize for the best model configurations. Using a real-world offline shopping dataset, we demonstrate that the proposed approach significantly outperforms classic and state-of-the-art recommendation methods. In addition to benchmarking, our evaluations also show that each proposed model component and each operationalized factor contributes clear added value to the recommendation performance of our method