Title: Bayesian Mallows model for clicking data
Authors: Qinghua Liu - University of Oslo (Norway) [presenting]
Abstract: Learning individual preferences from clicking data is an important step in order to make personal recommendations. One of the most popular approaches to personal recommendation is Collaborative Filtering (CF), which is based on a low rank matrix factorization technique. One important challenge of CF is the lack of reliable uncertainty quantification. We developed a Bayesian Mallows Model (BMM) to make inference from clicking data to make personal recommendations for each user. We treated clicking data as pairwise comparisons, and the method includes clustering of users and relies on Bayesian data augmentation. Recommendations are made based on posterior probabilities. We compare the accuracy of BMM with that of CF's.