Sentiment Rating Prediction using Neural Collaborative Filtering
Pretty Paul, Rimjhim Padam Singh
Abstract
The main purpose of this study is to examine the working of neural collaborative filtering by utilizing the user reviews. We have done this research using the Amazon Reviews dataset. E-commerce is the most well-known application of recommendation algorithms. Websites that use information about a customer's tastes to provide a list of recommended things. Sentiment analysis can help you better understand the emotions, attitudes, and views of your users. Collaborative-filtering in a recommender system is beneficial to improve suggestion reliability as it focuses on the interaction between items and users, and the similarity of two items is determined by the similarity of their ratings by users. In the paper, sentiment rating for each user review has been calculated using the Vader lexicon approach to get the polarity score and a general framework of Neural Collaborative Filtering with a neural architecture has been studied that can learn any function from the input by replacing the inner product. NCF (Neural Collaborative Filtering) is a generic framework that can express and generalize matrix factorization, train user–item interaction function and introduce a multi-layer perceptron to boost NCF modeling with non-linearity. The Mean absolute error results obtained for different models claim for the NCF to be the best approach.