The Value of Nothing: Multimodal Extraction of Human Values Expressed by TikTok Influencers

A team of researchers from Hebrew University of Jerusalem and Reichman University has published a groundbreaking study titled "The Value of Nothing: Multimodal Extraction of Human Values Expressed by TikTok Influencers." The research addresses how societal and personal values are transmitted to younger generations through social media, specifically focusing on TikTok influencers who target children and adolescents. The team curated and annotated a dataset of hundreds of TikTok videos according to the established Schwartz Theory of Personal Values, which categorizes universal motivations like security, tradition, and stimulation.

The researchers experimented with multiple AI pipelines to identify values expressed in these short-form videos. They tested two main approaches: directly extracting values from the multimodal video content, and a novel two-step method where videos are first converted into detailed textual scripts (using automatic speech recognition and scene description), then analyzed for values using large language models. Their findings revealed that the two-step approach using few-shot prompting with LLMs like GPT-4 significantly outperformed direct extraction methods and even fine-tuned masked language models.

This work represents the first systematic attempt to extract human values specifically from TikTok and visual social media content at scale. The team has publicly shared their values-annotated TikTok dataset, creating a valuable resource for future research in computational social science. The study demonstrates how advanced AI models can be leveraged to understand complex, implicit communication in entertainment-focused platforms, paving the way for larger-scale analysis of value transmission in digital environments.