Explicit feedback meet with implicit feedback in GPMF: a generalized probabilistic matrix factorization model for recommendation

Abstract

Recommender Systems focus on implicit and explicit feedback or parameters of users for better rating prediction. Most of the existing recommender systems use only one type of feedback ignoring the other one. Based on the availability of resources, we may consider more number of feedback of both the types to predict user’s rating for a particular item more accurately. However to the best of our knowledge, there is no generalized model that is fitted for multiple parameters or feedback. In this paper, we have proposed a Generalized Probabilistic Matrix Factorization (GPMF) model which uses multiple parameters of both the types for recommendation. To build GPMF, first we develop three models focusing on users’ crucial side information. First model PMFE (P robabilistic M atrix F actorization with E xplicit_Feedback) is proposed based on an explicit feedback of users and second one is PMFI (P robabilistic M atrix F actorization with I mplicit_Feedback), where an implicit feedback is considered. The last one is PMFEI (P robabilistic M atrix F actorization with E xplicit and I mplicit_Feedback), where both explicit and implicit feedback are considered. Extensive experiments on real world datasets show that PMFEI performs better compare to baselines. PMFEI model also performs better compare to baselines for cold-start users and cold-start items also. In our experimental section, it is shown that GPMF performs better when we consider both explicit and implicit feedback. The effectiveness of each parameter is not same for recommendation. Using GPMF we can estimate the effectiveness of a parameter. Based on this effectiveness, we can add or remove more parameters for better rating prediction.