Exploring Customer Preferences with Probabilistic Topics Models

Customer preference learning and recommendation for (e-)commerce is a widely researched problem where a number of different solutions have been proposed. In this study we propose and implement a novel approach to the problem of extracting and modelling user preferences in commerce using latent topic models. We explore the use of probabilistic topic models on transaction itemsets considering both single one-time actions and customers’ shopping history. We conclude that the extracted latent models not only provide insight to the consumer behaviour but also can effectively support an item recommender system.