The explosive growth of the Internet and social networks has gained more and more popular content, but at the same time, fake news and informal news have spread.
This false information not only causes confusion to people in sensitive areas such as health, natural disasters, and public policy, but also resulted in serious consequences such as economic losses, organizational reputation damage and national security threats.
Vegraph is a research and development of the engineer team of the Viettel Data Services and Artificial Intelligence Center (Viettel AI) and developed and developed the information verification method (photo: Viettel AI).
In the past, this information only affected certain time and scope, but now it is becoming more and more dangerous as it becomes part of the input data of the AI system. Systems such as chatbots or virtual assists are likely to use, create or disseminate false information if they do not have their own verification ability in the process of searching and collecting information to respond to users.
The fact confirmation is the process of verifying the truth or false information of information. For example, the purpose of confirmation of the statement that “NASA found life on Mars” is to check whether it is based on the facts published through reliable sources such as newspapers, scientific data or knowledge.
The AI system must be able to understand the sentences, find relevant information, and draw logical conclusions based on objective data.
VEGRAPH (Verify-in-The-Graph) is an information verification method published by NAACL 2025, one of the three international conferences in the field of natural language processing, researched and developed by the engineer team of the Viettel Data Service and Viettel AI.
According to experimental results for HOVER and FEVEROUS, the two popular verification data sets, VEGRAPH improves accuracy by 2-5% compared to conventional methods.
Currently, most of the facts confirmation methods are still struggling to handle ambiguous, metaphorical or multilayer statements. This statement is often used to “disguise” false information, and it is difficult for users to trust the results because they cannot clearly explain why the content is considered false.
In addition, many methods rely greatly on large -scale language models (LLM), which lacks authoritative knowledge sources and ability to cross -check, and there is a risk of drawing wrong conclusions, also known as “hallucinations.”
Unlike tools that only inferring internal reasoning within the model, VEGRAPH is compared with reliable reference materials such as legal documents, government databases, professional documents, etc., separating the request for information verification into a small proposition in advance. The entire verification process is organized into a clear stage, making it easy for users to track and check and increase accuracy. This is more and more important in AI applications. This feature can also quickly adjust and improve your system when your organization is needed.
VEGRAPH, which aims for a transparent and safe digital environment, is highly regarded for its practical application. In addition to the ability to integrate into the AI system to increase accuracy, this technology can be developed as an information verification system in various fields such as medicine, journalism, law or national management. This allows people to verify important information that requires high authenticity, such as the latest news, drugs, vaccines, pharmaceuticals, or legal regulations.
Systems such as chatbots or virtual assists are very likely to use false information to create or distribute (Viettel AI).
In the future, VEGRAPH will expand to process various data formats such as images, videos, and audio, recognize complex language forms such as metaphors, and meaning, and integrate knowledge graphs to improve reasoning ability.
The NAACL 2025 (Annual Meeting of the American Society for Modern Linguistics) is a science forum leading the world in natural language processing and calculation linguistics.
More than 3,000 papers were registered at this year’s conference, and the adoption rate of the mainstream papers (about 22%) was very competitive, and the most groundbreaking studies gathered in one place. NAACL 2025 focuses on the development of large -scale language models, multilateral and cultural natural language processing, new reasoning functions and breakthrough development in the field of AI.
Viettel AI is a business unit under the Military Industry-Telecommunications Group (Viettel) and is leading the development and acquiring products and services of AI, big data, robotics, and digital twin fields.
Currently, Viettel AI ecosystems include the highest quality product lines in Vietnam, and are trusted and used by many large domestic and foreign institutions and companies.
source:
