Vietnamese AI: Knows how to check information, knows how to tell the truth

At NAACL 2025 - one of the world's top 3 conferences on natural language processing held in the US, Viettel AI introduced VeGraph. This is a new information verification method that helps improve accuracy by 2% - 5% compared to current methods on two popular verification datasets, HoVer and FEVEROUS.

The proliferation of the Internet and social media has made fake news and misinformation increasingly common, especially in sensitive areas such as health , natural disasters, and public policy. This problem becomes even more serious when this misinformation becomes input data for AI systems such as chatbots or virtual assistants. If not verified, AI can unintentionally spread or even create misinformation.

VeGraph is an information verification method researched, developed and presented by Viettel AI at the NAACL 2025 conference.

Fact-checking is the process of verifying the truth or falsity of a statement based on reliable sources such as newspapers, scientific literature, or official databases. To do this, AI systems need the ability to understand language, retrieve data, and reason logically.

Born from the engineering team of Viettel's Data and Artificial Intelligence Service Center, VeGraph (Verify-in-the-Graph) decomposes verification requests into small clauses and compares them with authoritative sources such as legal documents, government data, and specialized documents. Unlike many current methods that rely solely on large language models (LLM), VeGraph organizes the verification process into clear steps, helping to increase transparency and reduce the phenomenon of "illusion" - when AI automatically creates false information.

In addition to its effectiveness in verifying information, VeGraph is also highly applicable. This technology can be integrated into AI systems to enhance reliability or deployed independently in areas such as healthcare, journalism, law or state management. People can completely use it to self-verify important information that requires high authenticity such as current news, drug information, vaccines or legal regulations.

In the future, Viettel AI aims to develop VeGraph in the direction of processing multiple data formats such as images, videos, and audio; at the same time, increasing the ability to understand complex languages ​​such as metaphors, implications, and integrating knowledge graphs to improve reasoning capabilities.

NAACL 2025 (Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics) is the world's leading prestigious scientific forum in the field of natural language processing and computational linguistics.

This year, the conference attracted a record number of papers, more than 3,000, with a very competitive acceptance rate for key papers (around 22%), bringing together the most groundbreaking research.

NAACL 2025 specifically focuses on breakthrough advances in large language model development, cross-cultural and multilingual natural language processing, emerging inference capabilities, and responsible AI.

Source: https://nld.com.vn/ai-cua-nguoi-viet-biet-kiem-tra-thong-tin-biet-noi-su-that-196250517125935127.htm