Russian Neural Network to Find the Truth on Social Media
Scientists have refined AI technologies to verify the authenticity of posts on social media.
Russian scientists are working on creating responsible artificial intelligence. At the Kola Scientific Center of the Russian Academy of Sciences (RAS), they have improved an algorithm that analyzes posts on social media. Such information serves as an important data source for gauging public sentiment.
Often, neural networks collect and initially analyze this data. However, unverified information may be misinterpreted as reliable. To address this issue, the Russian Academy of Sciences has integrated large language models into open data monitoring systems.
Clustered Approach
LLMs (large language models) can interact with data in three ways: a simple query (fast but less reliable), a pipeline with preliminary keyword extraction to form topics, and the most accurate method, where texts are converted into vectors using large language models and grouped by similar features. Clusters are then summarized separately.
The development was tested in neighborhood chat groups. The clustered approach showed 94% traceability and 100% stability.
