A Deep Learning Architecture for Profile Enrichment and Content Recommendation

Abstract

The objective of our work is to propose a personalized recommender system based on user–resource interactions and user/resource characteristics in order to suggest the most appropriate resources to users. Recommender systems are commonly used in connection with artificial intelligence because machine learning techniques are frequently used to create recommendation algorithms. We suggest using neural networks to improve resource suggestions. Neural networks can quickly perform complex tasks and easily handle massive data. Our new approach is based on an auto-encoder used as a data extractor to improve user performance and a deep neural network (DNN) for item recommendation. For this, we use an information extraction technique based on the reduction of dimensionality by the use of an auto-encoder to compress the features (the proposed tags). We want to create some semantics between the data in order to enrich the profile of the user. Then based on the history of user ratings as well as data on users and items (useful information such as genre, year, tag and rating), we develop a framework for deep learning to learn a similarity function between users and predict item ratings. We evaluate the effectiveness of the proposed approach through an experimental study on a real-world dataset such as the MovieLens film recommender system, according to a number of properties. Experimental validation combines both the accuracy of the recommendation system and a set of quality metrics.