Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SemViQA: A Semantic Question Answering System for Vietnamese Information Fact-Checking

Dien X. Tran, Nam V. Nguyen, Thanh T. Tran, Anh T. Hoang, Tai V. Duong, Di T. Le, Phuc-Lu Le

TL;DR

This work tackles misinformation in Vietnamese by introducing SemViQA, a hybrid fact-checking framework that merges Semantic-based Evidence Retrieval (SER) with a Two-step Verdict Classification (TVC). It balances semantic evidence extraction with efficient, multi-stage verdicts, enabling robust handling of long-context claims. Empirical results show state-of-the-art strict accuracy on ISE-DSC01 (78.97%) and ViWikiFC (80.82%), with SemViQA Faster achieving up to 7× speedups. The approach advances practical, scalable fact verification for low-resource languages and offers a publicly available implementation for broader adoption.

Abstract

The rise of misinformation, exacerbated by Large Language Models (LLMs) like GPT and Gemini, demands robust fact-checking solutions, especially for low-resource languages like Vietnamese. Existing methods struggle with semantic ambiguity, homonyms, and complex linguistic structures, often trading accuracy for efficiency. We introduce SemViQA, a novel Vietnamese fact-checking framework integrating Semantic-based Evidence Retrieval (SER) and Two-step Verdict Classification (TVC). Our approach balances precision and speed, achieving state-of-the-art results with 78.97\% strict accuracy on ISE-DSC01 and 80.82\% on ViWikiFC, securing 1st place in the UIT Data Science Challenge. Additionally, SemViQA Faster improves inference speed 7x while maintaining competitive accuracy. SemViQA sets a new benchmark for Vietnamese fact verification, advancing the fight against misinformation. The source code is available at: https://github.com/DAVID-NGUYEN-S16/SemViQA.

SemViQA: A Semantic Question Answering System for Vietnamese Information Fact-Checking

TL;DR

This work tackles misinformation in Vietnamese by introducing SemViQA, a hybrid fact-checking framework that merges Semantic-based Evidence Retrieval (SER) with a Two-step Verdict Classification (TVC). It balances semantic evidence extraction with efficient, multi-stage verdicts, enabling robust handling of long-context claims. Empirical results show state-of-the-art strict accuracy on ISE-DSC01 (78.97%) and ViWikiFC (80.82%), with SemViQA Faster achieving up to 7× speedups. The approach advances practical, scalable fact verification for low-resource languages and offers a publicly available implementation for broader adoption.

Abstract

The rise of misinformation, exacerbated by Large Language Models (LLMs) like GPT and Gemini, demands robust fact-checking solutions, especially for low-resource languages like Vietnamese. Existing methods struggle with semantic ambiguity, homonyms, and complex linguistic structures, often trading accuracy for efficiency. We introduce SemViQA, a novel Vietnamese fact-checking framework integrating Semantic-based Evidence Retrieval (SER) and Two-step Verdict Classification (TVC). Our approach balances precision and speed, achieving state-of-the-art results with 78.97\% strict accuracy on ISE-DSC01 and 80.82\% on ViWikiFC, securing 1st place in the UIT Data Science Challenge. Additionally, SemViQA Faster improves inference speed 7x while maintaining competitive accuracy. SemViQA sets a new benchmark for Vietnamese fact verification, advancing the fight against misinformation. The source code is available at: https://github.com/DAVID-NGUYEN-S16/SemViQA.

Paper Structure

This paper contains 32 sections, 8 equations, 9 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. SemViQA - Semantic Vietnamese Question Answering
  4. Evidence Extraction via Question Answering with Token Classification
  5. Data Processing
  6. Question Answering with Token Classification (QATC)
  7. Semantic-based Evidence Retrieval (SER)
  8. Two-step Verdict Classification (TVC)
  9. Stage 1: Three-Class Classification.
  10. Stage 2: Binary Classification.
  11. Final Prediction Rule.
  12. SemViQA Pipeline System
  13. Experiments
  14. Dataset
  15. Experimental Setup
  16. ...and 17 more sections

Figures (9)

  • Figure 1: Overview of a Sample Information Fact-Checking Task
  • Figure 2: SemViQA: A Two-Stage Method for Semantic-based Evidence Retrieval (SER) and Two-step Verdict Classification (TVC), where $P_2$ and $P_3$ represent the probabilities of the two-class and three-class classifications, respectively, and $\hat{y}_{\text{2}}$ and $\hat{y}_{\text{3}}$ denote their corresponding predictions.
  • Figure 3: Long context processing solution.
  • Figure 4: Comparison of methods in terms of peak performance and total inference time across datasets. Each retrieval approach is evaluated by its best score, while overall efficiency is reflected through cumulative inference time. See Table \ref{['tab:main_results']} for details.
  • Figure 5: Graph representing the lengths of contexts.
  • ...and 4 more figures