- Objective: This project aims to develop a model that classifies threatening conversations (threats, extortion, workplace bullying, and other types of harassment) and general conversations in Korean in 3days. By training an NLU (Natural Language Understanding)-based model, we seek to analyze text data effectively and automatically determine whether a conversation is threatening.
Data tunib-ai/DKTC: TUNiB's Self-Produced Dataset to Participate in the Voice Recognition Track of the 4th AI Grand Challenge 2021.
- The training dataset consists of approximately 1,000 samples per class for four categories: threats, extortion, workplace bullying, and other harassment.
- The test dataset consists of approximately 100 samples per class for five categories: threats, extortion, workplace bullying, other harassment, and general conversation.
| Dataset | Classes | Samples |
|---|---|---|
| Train | Threats, Extortion, Workplace Bullying, Other Harassment | around 4,000 |
| Test | Threats, Extortion, Workplace Bullying, Other Harassment | around 500 |
| Additional Data | General Conversation (Synthetic Data, AI-hub Dataset) | (+)samples |
Evaluation: Measure by the f1 score between the results classified by the model and the correct answer.
┣ assets/ # About Dataset, results(photos) and report(.md)
┣ models/ # Trained models and checkpoints
┣ src/ # Source code, data analysis, and experiment notebooks
┃ ┗ notebooks/ # Data analysis and experiment notebooks (.ipynb)
┃ ┗ utils/ # Preprocess and other required modules (.py)
┗ 📜 README.md # Project description file
| Version | Date | Contentes | Update Notes | Number of Members |
|---|---|---|---|---|
| v1.0.0 (done) | Feb.21.2025 | Draft | Group project | 4 |