An Efficient Recommendation Framework on Social Media Platforms Based on Deep Learning

Abstract

Accurate user behavior modeling is the key to recommendation system. Traditional methods of friend recommendation extract the part of the information of users which results in an incomplete description of the user behavior. The efficiency and accuracy are far from expectation. We propose a Deep Graph-Based Neural Network (DGBNN) framework for friend recommendation on social media platforms. We obtain a comprehensive representation of user features to generate a more reasonable recommendation list taking into account the user's relationship and individual attributes. In order to get an accurate prediction, Back-Propagation Neural Network is used to predict the social links in social graphs. Recommendation list is constructed based on the predicted social links between users. Experiment with weibo datasets shows performance improvement on friend recommendation.