Analysis of Emerging Technology Growth Uncertainty through AI-empowered Multimodal Sentiment Incongruence Analysis: A Case Study of Large Language Models

This study proposes a novel methodology and a quantitative metric, the Sentiment Incongruity Index (SII), for analyzing multimodal signals and their incongruence in video-based evidence on emerging technologies, particularly large language models (LLMs). Our approach integrates AI-based facial expression analysis and text mining into a unified multimodal framework to quantify discrepancies between textual sentiment and facially expressed sentiment. Applying this framework to LLMs as a case study reveals a consistent positivity bias in expert verbal expressions, and pronounced sentiment incongruence across both LLMs in general and their key sub-technologies. The findings further underscore the importance of experts' nonverbal cues for mitigating biases inherent in expert judgment and traditional text analytics, and lay the groundwork for incorporating nonverbal signals into emerging technology analysis.