A Dual-Framework Approach for Interpreting and Predicting Social Media Discourse through AI and Data Science Techniques

Abstract

In the age of digital consumerism, electronic Word of Mouth (eWOM) significantly influences customer opinions and decision-making processes. Social media platforms, such as Twitter, Facebook, and Instagram, have become vital channels for public discourse, generating vast amounts of unstructured data. This thesis addresses the need for innovative analytical approaches to harness the value of social media data, proposing a comprehensive dual framework that integrates sentiment analysis (SA) and network analysis. The research aims to fill gaps in existing methodologies by developing a framework that identifies key influencers, patterns of information spread, and major themes, while also analysing public sentiment and its changes over time. The methodology involves advanced natural language processing (NLP) techniques, including a sentiment classifier that incorporates emoji features to enhance classification accuracy. Additionally, network analysis techniques, such as link analysis and community detection, are employed and applied at the user level and tweet level, respectively, to evaluate online user engagement. Case studies focusing on the United Kingdom’s (UK's) cost of living crisis and movie box office performance were conducted to validate the framework's effectiveness in terms of comprehensiveness, accuracy, robustness, and practical utility. The results demonstrate the framework's ability to provide actionable insights, revealing a range of public responses and identifying key topics and influencers. Notably, the study revealed a significant impact of online user engagement on business performance. The findings indicate the utility of the dual-framework in various domains, from governmental policy analysis to business strategy and crisis management. This research addresses several gaps in existing Social Media Analytics (SMA), studies, particularly the integration of sentiment and network analysis and the handling of informal social media language. By enhancing methodological transparency and developing standardised guidelines, this thesis contributes to knowledge about the reliability and validity of social media research, offering a valuable tool for understanding the complexities of online discourse.