Tripartite Collaborative Filtering with Observability and Selection for Debiasing Rating Estimation on Missing-Not-at-Random Data

Most collaborative filtering (CF) models estimate missing ratings with an implicit assumption that the are missing-at-random, which may cause biased rating estimation and degraded performance since recent deep exploration shows likely be missing-not-at-random (MNAR). To debias MNAR estimation, we introduce item observability user selection to depict generation of propose a tripartite CF (TCF) framework jointly model triple aspects generation: observability, selection, ratings, ratings. An variable is introduced complete infer whether observable user. TCF also conducts for utilizes dependent on values items. We further elaborately instantiate as Tripartite Probabilistic Matrix Factorization (TPMF) by leveraging probabilistic matrix factorization. Besides, TPMF introduces multifaceted dependency between influence Extensive experiments synthetic real-world datasets show modeling effectively outperforms state-of-the-art methods in estimating