dat201204/vietnamese-caucu-comments

Vietnamese Cau Cuu Facebook Comments

Dataset Summary

This dataset contains Vietnamese Facebook comments collected from a natural-disaster discussion thread and auto-labeled for binary emergency detection.

The target task is to detect whether a comment is a real-time rescue request (cau_cuu) versus a non-emergency comment (khong_phai_cau_cuu).

This release is intended as a bootstrap dataset for triage modeling and should be treated as a weakly supervised resource. Human review is strongly recommended before production use.

Task Definition

• —0: khong_phai_cau_cuu Non-emergency content such as sympathy, reposts, hotline aggregation, updates that the family is already safe, or unrelated discussion.
• —1: cau_cuu Active rescue requests where people are trapped, in immediate danger, isolated, or explicitly requesting emergency evacuation/support.

Priority metric for downstream models: recall on label 1.

Data Source

• —Source type: Vietnamese Facebook comments from a disaster-related post/thread.
• —Data was flattened from both top-level comments and nested replies.
• —Original extraction and weak labeling were produced locally for research and experimentation.

Data Fields

• —id: Stable synthetic identifier derived from comment tree position.
• —text: Raw Vietnamese comment text.
• —label: Binary weak label (0 or 1).
• —confidence: Heuristic confidence score from the auto-labeling pipeline.

Class Distribution

Current version statistics:

• —Total rows: 1492
• —Label 0: 1050
• —Label 1: 442

Labeling Method

Labels were assigned with a rule-based weak supervision pipeline using:

• —urgent rescue keywords such as cứu, mắc kẹt, ngập tới mái, khẩn cấp, SOS
• —structural signals such as phone numbers, map links, GPS-like coordinates, and location mentions
• —negative filters for resolved cases (đã được cứu, đã an toàn) and hotline/broadcast style comments

Because labels are weakly supervised, false positives and false negatives remain possible.

Recommended Use

• —Training or bootstrapping a Vietnamese emergency comment classifier
• —Error analysis and heuristic refinement
• —Human-in-the-loop triage experiments

Limitations and Ethics

• —Comments may contain sensitive situational information such as phone numbers and addresses.
• —Labels are machine-generated and not fully human-verified.
• —Do not use this dataset for surveillance or any harmful downstream purpose.
• —Review privacy, legal, and platform-policy constraints before redistribution or deployment.

Citation

If you use this dataset, please cite the project/repository that publishes this dataset card and describe it as a weakly supervised Vietnamese rescue-request classification dataset.