| --- |
| library_name: transformers |
| license: apache-2.0 |
| base_model: distilbert-base-uncased |
| tags: |
| - generated_from_trainer |
| model-index: |
| - name: help-classifier-v2 |
| results: [] |
| datasets: |
| - King-8/help-request-messages-v2 |
| --- |
| # 🤖 Help Classifier Model (v2) |
|
|
| ## 🧠 Overview |
|
|
| The **Help Classifier Model (v2)** is a fine-tuned NLP model designed to classify student help requests into meaningful categories within a collaborative learning environment. |
|
|
| This model is part of a larger AI system built for the **Coding in Color (CIC)** ecosystem, supporting students working across domains such as AI development, game development, 2D/3D art, and robotics. |
|
|
| Its primary purpose is to: |
|
|
| * Interpret real student messages |
| * Identify intent behind help requests |
| * Route inputs to appropriate downstream systems (e.g., generators, agents) |
|
|
| --- |
|
|
| ## 🚀 Version Update (v1 → v2) |
|
|
| ### 🔹 v1 |
|
|
| * Trained on ~100 examples |
| * Limited generalization |
| * Struggled with messy or informal input |
|
|
| ### 🔹 v2 (Current) |
|
|
| * Trained on **1,000 examples** |
| * Balanced dataset across all categories |
| * Strong performance on: |
|
|
| * informal/slang input |
| * mixed tone messages |
| * ambiguous phrasing |
| * real CIC-style check-ins |
|
|
| 👉 v2 significantly improves **accuracy, stability, and real-world usability** |
|
|
| --- |
|
|
| ## 🧩 Task Definition |
|
|
| **Task Type:** Text Classification |
|
|
| **Input:** Student message |
| **Output:** One of 5 help categories |
|
|
| --- |
|
|
| ## 🏷️ Labels |
|
|
| | Label | Description | |
| | ------------------ | --------------------------------------------------- | |
| | `learning_help` | User is trying to understand a concept or skill | |
| | `project_help` | User needs direction or next steps in a project | |
| | `technical_issue` | Something is broken or not working | |
| | `attendance_issue` | User missed a meeting or needs to catch up | |
| | `general_guidance` | User expresses uncertainty, stress, or needs advice | |
|
|
| --- |
|
|
| ## 🏗️ Model Architecture |
|
|
| * Base Model: distilbert-base-uncased |
| * Fine-tuned for sequence classification |
| * Number of labels: 5 |
|
|
| --- |
|
|
| ## ⚙️ Training Configuration |
|
|
| * Epochs: 4 |
| * Learning Rate: 2e-5 |
| * Batch Size: 8 |
| * Weight Decay: 0.01 |
| * Train/Validation Split: 80/10/10 |
|
|
| --- |
|
|
| ## 📊 Training Results |
|
|
| | Epoch | Training Loss | Validation Loss | |
| | ----- | ------------- | --------------- | |
| | 1 | 0.552 | 0.512 | |
| | 2 | 0.111 | 0.122 | |
| | 3 | 0.032 | 0.077 | |
| | 4 | 0.025 | 0.064 | |
|
|
| --- |
|
|
| ## 📈 Performance Summary |
|
|
| * **Low validation loss (~0.06)** |
| * Strong generalization across unseen inputs |
| * Stable convergence during training |
| * Handles: |
|
|
| * messy/slang text |
| * indirect requests |
| * multi-layered inputs |
|
|
| --- |
|
|
| ## 🧪 Example Predictions |
|
|
| **Input:** |
|
|
| ``` |
| i missed the meeting and now idk what we’re doing |
| ``` |
|
|
| **Output:** |
|
|
| ``` |
| attendance_issue |
| ``` |
|
|
| --- |
|
|
| **Input:** |
|
|
| ``` |
| my model works but the predictions are weird and I don’t know why |
| ``` |
|
|
| **Output:** |
|
|
| ``` |
| technical_issue |
| ``` |
|
|
| --- |
|
|
| **Input:** |
|
|
| ``` |
| I feel like I’m behind and don’t know what to focus on |
| ``` |
|
|
| **Output:** |
|
|
| ``` |
| general_guidance |
| ``` |
|
|
| --- |
|
|
| ## 🔗 System Integration |
|
|
| This model is integrated into an MCP (Model Context Protocol) system where it acts as: |
|
|
| > **Entry-point classifier for routing student inputs** |
|
|
| Pipeline example: |
|
|
| ``` |
| User Input → Help Classifier → (Future: Generator / Summarizer) |
| ``` |
|
|
| --- |
|
|
| ## 🎯 Use Cases |
|
|
| * Help request classification |
| * Slack/Discord message routing |
| * Educational AI assistants |
| * CIC ecosystem tools |
| * AI agent pipelines |
|
|
| --- |
|
|
| ## ⚠️ Limitations |
|
|
| * Single-label classification (some messages may contain multiple intents) |
| * Edge cases may still overlap between categories |
| * Domain-specific (focused on student tech environments) |
|
|
| --- |
|
|
| ## 🔮 Future Improvements |
|
|
| * Multi-label classification |
| * Larger dataset (2,000+ examples) |
| * Confidence scoring |
| * Integration with response generation models |
| * Continuous retraining with real user data |
|
|
| --- |
|
|
| ## 👤 Author |
|
|
| Created by Kingston Lewis as part of the Coding in Color program for the AI Dev team. |
|
|
| --- |
|
|
| # help-classifier-v2 |
|
|
| This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on the King-8/help-request-messages-v2 dataset. |
| It achieves the following results on the evaluation set: |
| - Loss: 0.0643 |
|
|
|
|
| ### Training hyperparameters |
|
|
| The following hyperparameters were used during training: |
| - learning_rate: 2e-05 |
| - train_batch_size: 8 |
| - eval_batch_size: 8 |
| - seed: 42 |
| - optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments |
| - lr_scheduler_type: linear |
| - num_epochs: 4 |
| - mixed_precision_training: Native AMP |
| |
| |
| ### Training results |
| |
| | Training Loss | Epoch | Step | Validation Loss | |
| |:-------------:|:-----:|:----:|:---------------:| |
| | 0.5524 | 1.0 | 88 | 0.5124 | |
| | 0.1114 | 2.0 | 176 | 0.1221 | |
| | 0.0324 | 3.0 | 264 | 0.0771 | |
| | 0.0249 | 4.0 | 352 | 0.0643 | |
| |
| |
| ### Framework versions |
| |
| - Transformers 5.0.0 |
| - Pytorch 2.10.0+cpu |
| - Datasets 4.0.0 |
| - Tokenizers 0.22.2 |
| |
