A Comprehensive Study on Fact-checking Social Media Content using RAG and LLM
Abstract
The fast development of digital media and online news websites has made it easier and
quicker for people to get information. However, the spread of incorrect, misleading, and
possibly false information has become a major problem for individuals looking for accurate
and trustworthy sources. In response to this situation, this research introduces a dynamic
agentic fact-checking system. This system combines Natural Language Processing (NLP),
Deep Learning, Large Language Models (LLMs), and Retrieval-Augmented Generation
(RAG) to verify news claims in real time. The system uses the “openai/gpt-oss-120b”,
“openai/gpt-oss-20b”, “qwen/qwen3.6-27b” Large Language Model via the Groq API to
analyze user input in Bangla, English, and Banglish. The process begins with NLP to detect
important words and names. Then, a web search using DuckDuckGo fetches up-to-date and
relevant information from online news sources. The system analyzes the collected
information alongside the original claim to determine whether the claim is accurate.
Finally, it provides a well-structured, impartial verdict along with a detailed explanation
based on the evidence found. Unlike traditional fact-checking methods that rely on fixed
datasets or pre-existing knowledge, this system retrieves current online evidence before
forming a conclusion. This design allows the system to check information in real time and
explain its findings with supporting data. The proposed system shows how combining NLP,
LLMs, Agentic AI, and RAG can create an effective and flexible fact-checking tool for
verifying information across multiple languages.
Collections
- 2026-2030 [51]