SemViQA-BC is a core component of the SemViQA system, specifically designed for binary classification in Vietnamese fact-checking tasks. This model predicts whether a given claim is SUPPORTED or REFUTED based on retrieved evidence.

Model Information

Developed by: SemViQA Research Team
Fine-tuned model: Ernie-M
Supported Language: Vietnamese
Task: Binary Classification (Fact Verification)
Dataset: ISE-DSC01

SemViQA-BC is one of the key components of the two-step classification (TVC) approach in the SemViQA system. It focuses on binary classification, determining whether a claim is SUPPORTED or REFUTED. This step follows an initial three-class classification, where claims are first categorized as SUPPORTED, REFUTED, or NOT ENOUGH INFORMATION (NEI). By incorporating Cross-Entropy Loss and Focal Loss, SemViQA-BC enhances precision in claim verification, ensuring more accurate fact-checking results

Usage Example

Direct Model Usage

Python

# Install semviqa
!pip install semviqa

# Initalize a pipeline
import torch
import torch.nn.functional as F
from transformers import AutoTokenizer
from semviqa.tvc.model import ClaimModelForClassification

tokenizer = AutoTokenizer.from_pretrained("SemViQA/bc-erniem-isedsc01")
model = ClaimModelForClassification.from_pretrained("SemViQA/bc-erniem-isedsc01", num_labels=2)
claim = "Chiến tranh với Campuchia đã kết thúc trước khi Việt Nam thống nhất."
evidence = "Sau khi thống nhất, Việt Nam tiếp tục gặp khó khăn do sự sụp đổ và tan rã của đồng minh Liên Xô cùng Khối phía Đông, các lệnh cấm vận của Hoa Kỳ, chiến tranh với Campuchia, biên giới giáp Trung Quốc và hậu quả của chính sách bao cấp sau nhiều năm áp dụng."

inputs = tokenizer(
    claim,
    evidence,
    truncation="only_second",
    add_special_tokens=True,
    max_length=256,
    padding='max_length',
    return_attention_mask=True,
    return_token_type_ids=False,
    return_tensors='pt',
)

labels = ["SUPPORTED", "REFUTED"]

with torch.no_grad():
    outputs = model(**inputs)

logits = outputs["logits"]
probabilities = F.softmax(logits, dim=1).squeeze()

for i, (label, prob) in enumerate(zip(labels, probabilities.tolist()), start=1):
    print(f"{i}) {label} {prob:.4f}")
# 1) SUPPORTED 0.0007
# 2) REFUTED 0.9993

Evaluation Results

SemViQA-BC achieved impressive results on the test set, demonstrating accurate and efficient classification capabilities. The detailed evaluation of SemViQA-BC is presented in the table below.

Method	ISE-DSC01
ER	VC	Strict Acc	VC Acc	ER Acc	Time (s)
TF-IDF	InfoXLM_large	73.59	78.08	76.61	378
XLM-R_large	75.61	80.50	78.58	366
Ernie-M_large	78.19	81.69	80.65	403
BM25	InfoXLM_large	72.09	77.37	75.04	320
XLM-R_large	73.94	79.37	76.95	333
Ernie-M_large	76.58	80.76	79.02	381
SBert	InfoXLM_large	71.20	76.59	74.15	915
XLM-R_large	72.85	78.78	75.89	835
Ernie-M_large	75.46	79.89	77.91	920
QA-based approaches	VC
ViMRC_large	InfoXLM_large	54.36	64.14	56.84	9798
XLM-R_large	53.98	66.70	57.77	9809
Ernie-M_large	56.62	62.19	58.91	9833
InfoXLM_large	InfoXLM_large	53.50	63.83	56.17	10057
XLM-R_large	53.32	66.70	57.25	10066
Ernie-M_large	56.34	62.36	58.69	10078
LLM
Qwen2.5-1.5B-Instruct	59.23	66.68	65.51	19780
Qwen2.5-3B-Instruct	60.87	66.92	66.10	31284
LLM	VC
Qwen2.5-1.5B-Instruct	InfoXLM_large	64.40	68.37	66.49	19970
XLM-R_large	64.66	69.63	66.72	19976
Ernie-M_large	65.70	68.37	67.33	20003
Qwen2.5-3B-Instruct	InfoXLM_large	65.72	69.66	67.51	31477
XLM-R_large	66.12	70.44	67.83	31483
Ernie-M_large	67.48	70.77	68.75	31512
SER Faster (ours)	TVC (ours)
TF-IDF + ViMRC_large	Ernie-M_large	78.32	81.91	80.26	995
TF-IDF + InfoXLM_large	Ernie-M_large	78.37	81.91	80.32	925
SER (ours)	TVC (ours)
TF-IDF + ViMRC_large	InfoXLM_large	75.13	79.54	76.87	5191
XLM-R_large	76.71	81.65	78.91	5219
Ernie-M_large	78.97	82.54	80.91	5225
TF-IDF + InfoXLM_large	InfoXLM_large	75.13	79.60	76.87	5175
XLM-R_large	76.74	81.71	78.95	5200
Ernie-M_large	78.97	82.49	80.91	5297

Citation

If you use SemViQA-BC in your research, please cite:

bibtex

@misc{tran2025semviqasemanticquestionanswering,
      title={SemViQA: A Semantic Question Answering System for Vietnamese Information Fact-Checking}, 
      author={Dien X. Tran and Nam V. Nguyen and Thanh T. Tran and Anh T. Hoang and Tai V. Duong and Di T. Le and Phuc-Lu Le},
      year={2025},
      eprint={2503.00955},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.00955}, 
}

🔗 Paper Link: SemViQA on arXiv
🔗 Source Code: GitHub - SemViQA

About

Built by Dien X. Tran

For more details, visit the project repository.