About ReviewIQ

NLP-powered review intelligence for the AI x Data 2.0 Hackathon.

How It Works

1

Upload Reviews

Paste reviews or upload a CSV file with customer feedback.

2

AI Analysis

Hugging Face transformers classify sentiment and extract features, complaints, and trends.

3

View Insights

Explore your dashboard with charts, ranked lists, and exportable reports.

What We Analyze

  • Sentiment (positive, negative, neutral) via zero-shot classification
  • Product features customers love (fabric, fit, color, quality)
  • Common complaints (shrinking, sizing, quality issues)
  • Trending keywords with sentiment-colored word cloud

Trained on patterns from 23,000 women's clothing reviews. Uses facebook/bart-large-mnli for sentiment via Hugging Face Inference API.

Limitations

  • Sarcasm and irony may be misclassified as positive or negative
  • Slang and domain-specific language can reduce accuracy
  • Review bias: extreme reviews are overrepresented online
  • API rate limits on free Hugging Face tier for large batches
  • Feature extraction uses pattern matching — may miss novel phrases

Responsible AI

ReviewIQ surfaces aggregate patterns, not individual judgments. Sentiment scores should inform decisions, not replace human review. Models may reflect biases present in training data and review platforms. We recommend merchants validate insights against their own quality data before making business decisions. Users should not use outputs for automated rejection of products or harassment of reviewers.

Questions or feedback? Open an issue on GitHub or email your team lead.

Try ReviewIQ