Enhancing Movie Recommendation Systems Using Hybrid Filtering and Twitter-Based Sentiment Analysis

Abstract

Recommendation Systems (RS) have become an integral component of the online world, assisting users in navigating through the extensive information space of online e-commerce, as well as online entertainment systems.Although methods such as Collaborative Filtering (CF) and ContentBased Filtering (CBF) have been shown to be efficient and effective, they suffer from the problem of requiring a large historical user preference profile, thus often facing challenges such as the Cold Start Problem and Sparsity. To address such limitations, a multimodal movie recommendation framework that combines CF, CBF, and Sentiment Analysis of microblogging characteristics, namely, tweets, is presented in this paper. The addition of the Sentiment component ensures that the system encodes real-time public opinion, trending, and immediate reactions of the audience, thus enhancing the recommendation space beyond just rating matrices.This proposed framework uses the MovieTweetings dataset, which contains user ratings and various pieces of movie information, making it an ideal choice for sentiment-based analysis. This is because all tweets are processed and purified from noise consisting of slang terms, hyperlinks, and special characters before being scored using the VADER analyzer tool. Compound scores are then converted into a numerical rating system and combined using weighted scores of similarity indices consisting of various pieces of movie information. Experimental analysis proves the correlation between the ratings and IMDb ratings, and this proves the efficiency and effectiveness of combining the social signals. Correlation factors like PLCC, SROCC, and KRCC show the alignment of the sentiments extracted from the tweets and the preferences of the users. In addition to this, the hybrid model performs better on the Top-N recommendation procedure than the standard CF and CBF approaches. This approach generally establishes the significance of combining the structured ratings and the unstructured data from the social media site.