Multi-label Convolution Neural Network for Personalized News Recommendation based on Social Media Mining

Abstract

Prediction of user’s multi label interests and recommending the users interest based popular news articles through mining the social media are difficult task in Hybrid News Recommendation System (HYPNRS). To overcome this issue, this study proposes a deep learning approach - Multi-label Convolution Neural Network for predicting users' diversified interest in 15 labels using the binary relevance method. Based on labels of user’s interest, the most popular news articles are determined and their labels were clustered by mining social media feeds Facebook and Twitter along with current trends. The reliability of retrieved popular news articles also verified for recommendation. Eventually, the latest news articles catered from news feeds integrated along popular news articles and current trends together provide a recommendation list with respect to user interest. Experimental results show the proposed method diversified users interest labels prediction performance improved 5.87%, 12.09%, and 18.49% with the following state of art Support Vector Machine (SVM), Decision Tree and Naïve Bayes. The recommendation performance concerning users’ interest achieved 90%, 93.3%, 90% with social media feeds Facebook, Twitter and News Feeds accordingly.