A Prediction Scheme For Movie Preference Rating Based on DeepFM Model

Most recommendation systems work based on a user-item matrix with users on the rows and items on the columns. Each cell of the user-item matrix contains a rating value that indicates the user's preference for an item. However, in reality, there can be many missing rating values in the user-item matrix. So, the rating matrix is not completely filled and becomes very sparse. Such a problem is called a "data sparsity problem". To alleviate the data sparsity problem, predicting the ratings of unrated items with high accuracy is very important. The previously proposed prediction model is DeepFM, which reflects both low and high-order interaction of input features. But DeepFM is a general prediction model that does not depend on a specific problem domain. In this paper, we customized the DeepFM model to reflect better the low and high-order features interaction of the movie recommendation dataset than the original DeepFM. The evaluation results show that our proposed scheme predicts the rating with higher accuracy than the original DeepFM and other existing methods.