metadata
license: mit
tags:
- sentiment-analysis
- aspect-based-sentiment-analysis
- absa
- product-reviews
- scikit-learn
language:
- en
metrics:
- accuracy
- f1
library_name: sklearn
pipeline_tag: text-classification
AspectLens - Aspect-Based Sentiment Analysis
A production-ready Aspect-Based Sentiment Analysis (ABSA) model that extracts granular sentiment for specific product features from customer reviews.
Model Description
This model identifies sentiment for 6 product aspects:
- Battery, Camera, Display, Performance, Build, Price
Model Details
- Architecture: TF-IDF Vectorization + Logistic Regression
- Training Data: 200K Amazon mobile phone reviews
- Accuracy: 90% on test set
- Inference Time: <100ms per review
- Language: English
Performance
| Metric | Score |
|---|---|
| Overall Accuracy | 90.0% |
| Positive Class | 100% |
| Negative Class | 80% |
| Neutral Class | 90% |
| Weighted F1 | 0.90 |
Usage
from inference import predict
# Analyze a review
review = "Great camera but terrible battery life. The display is amazing though!"
result = predict(review)
print(result)
# Output:
# {
# 'aspects': [
# {'aspect': 'camera', 'sentiment': 'positive', 'confidence': 0.982},
# {'aspect': 'battery', 'sentiment': 'negative', 'confidence': 0.947},
# {'aspect': 'display', 'sentiment': 'positive', 'confidence': 0.991}
# ],
# 'overall_sentiment': 'positive'
# }
API Endpoint
Once deployed on Hugging Face Inference API:
import requests
API_URL = "https://api-inference.huggingface.co/models/YOUR_USERNAME/AspectLens-Aspectwise-sentiment-breakdown"
headers = {"Authorization": "Bearer YOUR_HF_TOKEN"}
def query(payload):
response = requests.post(API_URL, headers=headers, json=payload)
return response.json()
output = query({
"inputs": "Great phone but battery dies quickly"
})
Training Details
- Dataset: Amazon Product Reviews (Mobile Phones)
- Preprocessing: Text cleaning, sentence segmentation, aspect detection
- Features: TF-IDF with max 5000 features
- Classifier: Logistic Regression with balanced class weights
- Validation: 80/20 train-test split
Limitations
- English language only
- Focused on mobile phone reviews (generalizes to other electronics)
- Rule-based aspect detection (keyword matching)
- May struggle with sarcasm or complex negations
Intended Use
- Product development teams identifying pain points
- Customer support prioritizing issues
- E-commerce platforms analyzing feedback
- Marketing teams understanding preferences
Citation
@misc{aspectlens2026,
author = {Aryan Patel},
title = {AspectLens: Aspect-Based Sentiment Analysis for Product Reviews},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/YOUR_USERNAME/AspectLens}}
}
License
MIT License - Free for commercial and research use
Contact
- GitHub: @Aryan1patel
- Repository: AspectLens
Built with ❤️ using scikit-learn