Digital platforms have used recommender systems to recommend relevant products to their users based on their historical interactions. Recently, neural network-based recommender systems that generate embedding vectors have gained popularity in both research and practice and show improved performance over traditional methods. However, it is often difficult to explain why and how the recommended items are provided to specific users by these black box systems. In this study, we propose a novel user-centric approach to recommending retail items by exploiting the latent intent of the users from transaction histories. The latent theme is learned using a Latent Dirichlet Allocation topic modeling method. The proposed method can explain the intent of the focal user and other similar users. A preliminary evaluation study shows our method outperform the baseline methods in both the accuracy and the interpretability of the recommended items.

Abstract Only