AI-Driven Sentiment Analysis for Automotive Market Intelligence

With over 4.5 billion internet users worldwide, online video engagement and entertainment remain the most popular activities among users. In this context, social media platforms such as YouTube have emerged as a growing source of data, the efficient analysis of which has become even more critical. This thesis, based on the practical experience at Jato, exploits the potential of this sheer volume of web-sourced data, to perform sentiment and competitive analysis of automotive related content and draw market intelligence. The work leverages a framework of Large Language Models (LLMs) to process large amounts of textual data obtained from YouTube comments and video transcripts pertaining to vehicles in the British C-segment market. The study involves the design of a data retrieval pipeline capable of identifying and prioritizing the most relevant videos for a specific model and make, through a customized scoring system that favors user interaction above views and likes. Using both the YouTube API and web-scraping tools, we collect textual data that is subsequently fed to LLM-based analyzers for extracting sentiment, praised and criticized features, and mentioned competitors. This extracted information undergoes a normalization step where semantically similar features are clustered and competitor names are standardized, to allow aggregation. Afterwards, the system is extended into an autonomous agent capable of combining both sources and producing human-readable reports. Besides performing a full analysis for a given vehicle, the agentic system analyzes trends across several years and differences with competing models. The proposed framework demonstrates that custom LLM solutions outperform both traditional Aspect-Based Sentiment Analysis and general-purpose AI prompting, delivering higher-level and more controlled insights. The final prototype offers a scalable market intelligence tool, built for cross-analysis of various models and timelines and ready for integration with future data sources and different analysis.