RecNN: A Deep Neural Network based Recommendation System

Deep learning’s breakthrough in speech recognition, image analysis and natural language processing has helped it gain a considerable amount of recognition in today’s highly modernized world. As it is known, collaborative filtering and content-based filtering are two incredibly desired memory-based methods used for recommending new products to the targeted users, but it does happen to have certain restrictions and it thus fails to provide the intended user with effectual recommendations as primarily required. In this paper, we evaluate the performance of a revised version of a deep learning-based recommender system for movies and books using multiple fully connected dense layered neural net embeddings-based structures that primarily ensembles deep neural networks integrated alongside embedding layers and their dot product values. We develop and test the recommendation systems using the data provided by Wikipedia book dataset and MovieLens 100k dataset for books and movies respectively. To recommend books or movies to a particular user, the inputs are converted to an embedding layer and then passed through dense layers for obtaining recommendations. The same process was applied to 2,3 and 4 layer architecture for obtaining recommendations for both books and movies. The results that we’ve obtained shows that our approach is a promising solution when compared with the independent memory-based collaborative filtering methods or content-based methods. It leads us to conclude that our comparative study of multiple layered architectures provides probable research directions for deep learning-based recommender systems in the near future.