New TCRG research explores how to identify tobacco industry narratives in social media

at

A new paper from the Tobacco Control Research Group explores how to fine-tune Large Language Models  (LLMs) to identify the intent of social media messaging around tobacco and e-cigarettes. Traditional social media analysis has often just looked at the sentiment of messages (positive/neutral/negative), whereas this research goes one step further to look at the intent of the messaging from a tobacco control viewpoint: whether the messages are pro-tobacco, anti-tobacco, pro e-cigarette or anti e-cigarette.

The researchers developed a mixed human/generative AI methodology for fine-tuning LLMs to classify tobacco and e-cigarette related content, and in particular showed how a relatively resource-light LLM (Flan-T5) can be fine-tuned to achieve impressive results.

This work builds on other work from the team on using LLMs to classify sentiment, and topic modelling of the twitter conversations around the WHO FCTC COP9.

Co-author John Mehegan explains:

The accurate determination of the intent of tobacco-related messaging on social media is important for being able to monitor where and when tobacco industry narratives are being pushed on social media and by whom.

Read the paper:

Enhancing sentiment and intent analysis in public health via fine-tuned Large Language Models on tobacco and e-cigarette-related tweets, S. Elmitwalli, J. Mehegan, A. Gallagher, R. Alebshehy, Frontiers in Big Data, 28 November 2024, doi: 10.3389/fdata.2024.1501154

Tobacco Tactics Resources:

Print Friendly, PDF & Email