Latent Graph Predictor Factorization Machine (LGPFM) for modeling feature interactions weight

Regression is a machine learning model that predicts the target based on input data. Factorization Machines (FMs) are new class models that in addition to regression, present factorized interactions between a pair of features. FMs have been proven to accomplish good accuracy for the rating prediction tasks such as recommender systems. However, FM models all the interactions with the same weight which can be ineffective, since useless interactions may cause noisy results. In this paper, we propose a new model named: Latent Graph Predictor Factorization Machine (LGPFM) that capture the interaction weight of each pair of features using Convolutional Neural Network (CNN). LGPFM combines FM model with the benefits of the CNN that works efficiently in grid-type topology, which would improve significantly the accuracy of results.