AventIQ-AI/bert-employee-behaviour-analysis

Model Details

Model Name: Employee behaviour Analysis Model
Base Model: distilbert-base-uncased
Dataset: yelp_review_full

Training Device: CUDA (GPU)

---

Dataset Information

**Dataset Structure:**
DatasetDict({
train: Dataset({
features: ['employee_feedback', 'behavior_category'],
num_rows: 50,000
})
validation: Dataset({
features: ['employee_feedback', 'behavior_category'],
num_rows: 20,000
})
})

Available Splits:

• Train: 15,000 examples
• Validation: 2,000 examples

Feature Representation:

• employee_feedback: Textual feedback from employees (e.g., "The team is highly collaborative and supportive.")
• behavior_category: Classified behavior type (e.g., "Positive Collaboration")

---

Training Details

Training Process:

• Fine-tuned for 3 epochs
• Loss reduced progressively across epochs

Hyperparameters:

• Epochs: 3
• Learning Rate: 3e-5
• Batch Size: 8
• Weight Decay: 0.01
• Mixed Precision: FP16

Performance Metrics:

• Accuracy: 92.3%

---

Inference Example

import torch
from transformers import DistilBertTokenizer, DistilBertForSequenceClassification

def load_model(model_path):
    tokenizer = DistilBertTokenizer.from_pretrained(model_path)
    model = DistilBertForSequenceClassification.from_pretrained(model_path).half()
    model.eval()
    return model, tokenizer

def classify_behavior(feedback, model, tokenizer, device="cuda"):
    inputs = tokenizer(
        feedback,
        max_length=256,
        padding="max_length",
        truncation=True,
        return_tensors="pt"
    ).to(device)
    outputs = model(**inputs)
    predicted_class = torch.argmax(outputs.logits, dim=1).item()
    return predicted_class

# Example usage
if __name__ == "__main__":
    model_path = "your-username/employee-behavior-analysis"  # Replace with your HF repo
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model, tokenizer = load_model(model_path)
    model.to(device)

    feedback = "The team is highly collaborative and supportive."
    category = classify_behavior(feedback, model, tokenizer, device)
    print(f"Feedback: {feedback}")
    print(f"Predicted Behavior Category: {category}")

Expected Output:

Feedback: The team is highly collaborative and supportive.
Predicted Behavior Category: Positive Collaboration

---

Use Case: Employee Behavior Analysis Model

Overview

The Employee Behavior Analysis Model, built on DistilBERT-base-uncased, is designed to classify employee feedback into predefined behavior categories. This helps HR and management teams analyze workforce sentiment and improve workplace culture.

Key Applications

• Sentiment & Engagement Analysis: Identify trends in employee feedback to assess workplace satisfaction.
• Performance Review Assistance: Automate categorization of peer reviews to streamline HR evaluation.
• Conflict Resolution: Detect negative patterns in feedback to address workplace conflicts proactively.
• Leadership Assessment: Analyze feedback about managers and team leaders to enhance leadership training.

Benefits

• Scalability: Can process thousands of employee responses in minutes.
• Objective Analysis: Reduces bias by using AI-driven classification.
• Actionable Insights: Helps HR teams make data-driven decisions.

Future Improvements

• Expand dataset with more diverse employee feedback sources.
• Fine-tune with additional behavioral categories for nuanced classification.
• Integrate with company HR software for real-time feedback analysis.

---