March 8, 2024

Identifying textual disinformation using Large Language Models

Key Points

Disinformation detection accuracy using large language models is comparable to that of expert debunking.
Analysis focuses on identifying emotions and motives within text to combat disinformation effectively.
Assessment includes multilingual detection to ensure adaptability across different languages and contexts, with varying training requirements for each language involved in the study context, specifically addressing challenges unique to multilingual scenarios. The implementation of a semi-automated disinformation labeling system may enhance current techniques significantly.

Abstract

The spread of disinformation is becoming a more acute challenge in modern society. The rise of AI technologies is providing it with an additional boost, making disinformation creation and propagation available to almost anyone. This change in the disinformation landscape needs to be responded to by the improvement of debunking and detection techniques. The plain fact-checking solutions might not be sufficient, since disinformative articles often consist of manipulated versions of correct information. Large Language Models (LLM) can be employed to identify emotions, stances, and motives behind the text. This research aims to find the way those abilities of LLM can be used for disinformation detection with an accuracy comparable to the debunking expert. Additionally, the question of multilingual detection with LLM models will be addressed, since different languages might require different approaches in LLM training and tuning. Based on these results, a semi-automated disinformation labeling system is to be built.

Ask AI

Helpful

Bookmark