Impact Rating Scales on Recommender System and Using Deep Learning and Neural Network Models to Improve Rating Prediction

Abstract

A lot of websites enables users to have the ability to rank products and share their rankings with other users for personalization or social resolutions. In specific the recommender systems, personalization recommendations are produced by forecasting ratings for products which users are insensitive of, grounded on the scores users offered for various commodities. Explicit consumer ratings are composed by way of graphical widgets known as a rating scale. Every website or system usually applies a certain rating scale in several instances opposing from scales applied by other networks in their numbering, granularity, existing of neutral position or visual metaphor. Several tasks in the area of survey design testified on the impact of ranking scales on consumer ratings, these, though, are usually considered impartial tools when in recommender systems. This paper offers novel empirical data on the impact of rating scales on consumer ratings. We describe how the rating scale can affect the recommendation system user performance. In addition, we propose a novel approach based on deep learning-based augmentation of the collaborative filtering approach with deep neural networks for discovering the complex and deep interactions in the shared space between users, and ratings/reviews, and provide significant improvement for predicting user ratings.