A HYBRID DEEP LEARNING FRAMEWORK FOR MOVIE RECOMMENDATION USING LATENT FEATURE REPRESENTATION

Abstract

Abstract: The ensuing paper presents a recommendation hybrid deep learning model of movies based on latent feature representation, intended to enhance the personal movie discovery process with the help of heterogeneous data sources integration. The model is a combination of structured information, e.g. movie metadata and user ratings, and unstructured text information, obtained by plot summary. Genre-sensitive latent embeddings are trained to learn both semantic and contextual links between movies and patterns of user preferences based on previous interactions. The deep learning system uses convolutional and recurrent neural networks to predict the sequence and semantic attributes of plot descriptions and categorical and numerical attributes are processed in fully connected layers. These modalities are merged into a single latent feature space which allows learning to represent effectively. The large scale of experiments run on benchmark movie datasets show that the proposed framework remains always better than the state-of-the-art recommendation models in terms of precision, recall, and normalized discounted cumulative gain (nDCG). In addition, its hybrid design based on the latent feature promotes the use of text-based and category data considerably to enhance the performance of cold-start recommendation to produce more correct, varied, and contextual recommendations of movies to different user profiles.