Reddit’s subreddit recommendation engine based on deep neural network & sentiment analysis

Abstract

This paper attempts to develop a model-based subreddit recommendation engine that includes the sentiment of the user interaction when providing recommendations. Sentiment was integrated into an LSTM-based collaborative filtering recommendation engine with the aid of Dirichlet probability distribution to better identify user preferences such that recommendation quality can be improve. A Recurrent Neural Network approach to sentiment analysis was explored and a GRU-based sentiment analyser was implemented to analyse the sentiment of user historical interactions.