ASAS: Enhancing Fanbase Engagement with Explainable Self-Learning Sentiment Analysis

Abstract

"This research addresses a critical challenge faced by individuals and brands with large fanbases: the difficulty in understanding fan sentiment and its potential impact on fan engagement and performance. This is particularly pertinent within fast-paced sports such as Formula 1, where fan opinions can significantly influence driver and team morale. Currently social media platforms lack the tools necessary for fanbase owners to efficiently monitor fan sentiment, requiring time-consuming manual analysis. To address this gap, this thesis proposes the use of sentiment analysis techniques. This approach offers fanbase owners an automated and streamlined method to assess overall fan sentiment, fostering improved communication and understanding within the fanbase. The author introduces a novel multi-class sentiment analysis system designed for self-improvement and enhanced interpretability to overcome the above-mentioned problem. The proposed system leverages a pre-trained DistilBERT model, fine-tuned on a comprehensive dataset encompassing 28 distinct emotional sentiments. Self-learning capabilities are enabled through self-supervised techniques, facilitating continuous model refinement based on its own predictions. To improve prediction transparency, the system incorporates an Explainable AI (XAI) method utilizing attention weights. This approach pinpoints the most influential words within a sentence, revealing the key factors driving the model's classification decisions. "