ViClickbait-2025: A comprehensive dataset for Vietnamese clickbait detection

Area of Science:

• Natural Language Processing
• Machine Learning
• Information Retrieval

Background:

• Clickbait headlines pose challenges in online information consumption.
• Developing robust automatic clickbait detection systems is crucial for content credibility.
• A need exists for specialized datasets in non-English languages, such as Vietnamese.

Purpose of the Study:

• To introduce ViClickbait-2025, a novel Vietnamese-language dataset for clickbait detection research.
• To provide a comprehensive resource for training and evaluating machine learning models for clickbait identification.
• To facilitate advancements in understanding and combating deceptive online content.

Main Methods:

• Web scraping of 3414 headlines from eight Vietnamese news platforms (2023-2025).
• Annotation of headlines as clickbait or non-clickbait by three independent reviewers (Cohen's Kappa: 0.822).
• Data preprocessing including HTML removal, deduplication, normalization, and inclusion of nine key attributes.

Main Results:

• The ViClickbait-2025 dataset contains 31.2% clickbait headlines across 13 news categories.
• High inter-annotator agreement (Cohen's Kappa = 0.822) ensures annotation quality.
• The dataset is available in JSONL and CSV formats under a CC BY 4.0 license.

Conclusions:

• ViClickbait-2025 is a valuable, high-quality resource for Vietnamese clickbait detection research.
• The dataset supports the development of more accurate and reliable automatic clickbait detection models.
• This work contributes to improving the trustworthiness of online news consumption in Vietnam.