A Recommender System Integrating Long Short-Term Memory and Latent Factor

Recommender system is designed to model user’s interests based on their historical behavior. Recommended information meets the user’s interests and needs by itself without users providing explicit requirements. It can solve the problem of ”information overload” caused by the information explosion and the popularity of large data effectively. This paper put forward an LSTM-LFM recommender system that fused long short-term memory and latent factor models. This system can divide into two parts: the first part is a prediction of the user’s interestingness in the item’s name. First use the word2ver method to vectorize the item’s name, then establish positive and negative sample training data, and train the LSTM model, thus implementing the prediction of the user’s interestingness in the item’s name. The second part is the extraction and characterization of the item’s latent factor vector. In this paper, the user’s interestingness in the item’s name is introduced into the LFM model to train and obtain the latent factor matrix which integrates the user’s interestingness in the item’s name. Each column vector of the latent factor matrix represents the unique latent factor vector corresponding to each item. At last, by predicting the user’s interestingness to items, the most interesting items are recommended to the user as a recommendation list. In this paper, the comparison experiment on the MovieLens dataset verifies that this recommendation system that integrates long and short-term memory and latent factor is superior to the traditional UserCF, ItemCF, and LFM algorithms in precision and recall rate.