Exploring Financial Trends through Topic Modeling and Time-Series Analysis: A Clustering Approach Using Latent Dirichlet Allocation (LDA) on Twitter Data

Abstract

Social media platforms, particularly Twitter, have emerged as influential arenas for financial discourse, shaping and reflecting market sentiment in real time. This study explores the thematic structure of financial discussions on Twitter, employing Latent Dirichlet Allocation (LDA) to identify key topics and their temporal dynamics. A dataset of 11,932 finance-related tweets was analyzed, revealing five distinct topics encompassing corporate earnings, macroeconomic policies, geopolitical trade issues, and market trends. By correlating tweet volumes and topic prevalence with significant financial events, the study demonstrates the utility of social media as a barometer for market activity. Unlike traditional sentiment analysis, which predominantly classifies tweets into sentiment categories such as bullish, bearish, or neutral, the application of LDA enabled the extraction of latent themes that underpin these sentiments. This nuanced approach provided deeper insights into the narratives driving market discussions, offering a more comprehensive understanding of how thematic shifts in financial discourse align with market movements. Visualization techniques, including topic-term matrices and word clouds, further elucidated the structure of these conversations, enhancing interpretability and accessibility. The findings contribute to the growing body of research on social media analytics in finance, highlighting the potential of unsupervised learning techniques for financial trend analysis. By bridging the gap between thematic exploration and temporal analysis, this study offers a methodological framework for leveraging social media data to uncover actionable insights. The implications extend beyond academic research, providing practical tools for investors, financial analysts, and policymakers to navigate the dynamic relationship between social media narratives and market behavior. Future research could expand on these insights by integrating more advanced modeling techniques, such as transformer-based models, and exploring domain-specific patterns across asset classes like stocks, commodities, and cryptocurrencies. By examining the intersection of social media, financial events, and market dynamics, this study lays the groundwork for a deeper understanding of digital narratives in financial ecosystems.