Textual Context Aware Factorization Machines
Fatima Zahra Lahlou, Houda Benbrahim, Ismail Kassou · 2018
Context Aware Recommender Systems (CARS) are Recommender Systems (RS) that consider, in addition to users and items, other contextual information to provide more accurate predictions. However, in the real-life applications, obtaining contextual information in order to build such systems is not an obvious task. In this paper, and for the first time, we use the whole reviews written by users as contextual information. Furthermore, we propose a new CARS algorithm, based on the generic recommendation algorithm Factorization Machines, called: Textual Context Aware Factorization Machines (TCAFM). TCAFM take as input contextual data where the textual reviews are considered as context and compute context aware rating predictions. Experiments show that using the whole reviews as contexts significantly improves recommendation quality. Furthermore, using TCAFM leads to additional improvements. Our implementation of TCAFM is publicly available at [14].