Integrating Sentiment Analysis and Community Detection in Long Short-Term Memory Time Series Predictions: The WGA Strike Case

Amidst the 2023 labor strikes involving screenwriters and actors, the entertainment industry encountered substantial turbulence affecting both its creative endeavors and financial trajectories. This study aims to elucidate the intricate interplay among media discourse, disruptive events, and stock prices within this dynamic sector. Utilizing sophisticated natural language processing (NLP) techniques, including sentiment analysis with the Vader model and entity extraction, alongside network graph analysis, we scrutinize a rich dataset comprising over 13,000 news headlines sourced from specialized entertainment industry outlets. These headlines encapsulate a spectrum of themes, ranging from studio operations to strike resolutions, crucial during the labor strikes. To deepen our insights, we leverage the Louvain community detection algorithm to cluster headlines, revealing thematic coherence and prevailing topics within the news corpus. We develop three Long Short-Term Memory (LSTM) forecasting models to integrate these insights. The baseline model incorporates technical indicators from relevant tickers for the industry studied across the stock market, while the second augments these with compound sentiment scores and moving averages. Notably, the third model enriches prediction accuracy by incorporating headline counts from the top three topic communities alongside the parameters of the second model. Our findings underscore the significant advantage gained by integrating network analysis and sentiment score of the news corpus into the LSTM model, exhibiting substantial performance improvements. Specifically, the third model demonstrates the most remarkable enhancement, reducing mean average error by up to 17.99% and enhancing the R2 test by up to 13.52%. Our study underscores the pivotal role of network analysis in augmenting predictive modeling within a case study of an industry having a disruptive event.