Automated recommendation model using ordinal probit regression factorization machines

In the recent world with the increasing trend of online activities, there is a rapid growth of online users and online services resulting in high-dimensional sparse user–item interaction matrices. An efficient recommendation system can help an e-commerce industry to grow rapidly and ease the communication process for the users but the sparseness of data provides huge difficulty in the path of creation of an efficient recommendation model. In this paper, we have discussed and analyzed a methodology that works very well in reducing the high-dimensional user–item interaction matrices to a low-dimensional one considering a few important features. Our model uses factorization machines to extract the feature vector considering the explicit and the side information extracted from the data with the Categorical Probit Regression Model to further optimize the recommendation system. Ordinal probit regression factorization machines (OPRFMs) handle the sparsity of the data and low correlation problems and work well with multiclass ordinal data. OPRFM not only uses the information provided by the ratings for the movies but also uses the intercepts between the ratings for efficient recommendations. OPRFM proves to be an efficient algorithm and outperforms the state-of-the-art approaches on the benchmark datasets Movielens, i.e., ML-100K, ML 1 M, and ML-10 M.