Improving Recommender Systems by Using Time-Weighted Sentiment Analysis

For users, keeping track of the film industry's ever-growing range of movies and TV series is increasingly difficult and exacerbates the paradox of choice: that too many choices hinder decision-making. As a countermeasure, recommender systems can personalize offers and limit the variety of media to what users find relevant. In this paper, we present an approach to improving the accuracy of a collaborative filtering system that applies auction theory by differentiating data into private versus common value information. Incorporating objective, time-weighted information with reference to discounted utility theory, we tested the approach in a prototype using the MovieLens dataset expanded with movie reviews published in The New York Times. By optimizing the algorithm used, we improved recommendation quality by 0.78% compared with the unmodified collaborative filtering approach. With that novel combination of dataset expansion, auction theory, and discounted utility theory, we have narrowed a particular gap in research on recommender systems