README.md

---
title: Topic Analysis
emoji: 📊
colorFrom: green
colorTo: blue
sdk: docker
pinned: false
app_port: 7860
---

# 📊 Sentiment & Topic Analysis Dashboard

Upload CSV, JSON, or Excel files containing customer feedback, support tickets, or reviews — get instant multilingual sentiment analysis, topic clustering, anomaly detection, and interactive visualizations.

## Features
- **Multilingual sentiment analysis** using `cardiffnlp/twitter-xlm-roberta-base-sentiment`
- **Dynamic topic clustering** with BERTopic (HDBSCAN + UMAP)
- **Interactive force-directed** topic cluster graph
- **Sentiment trend charts** with confidence intervals
- **Data quality dashboard** flagging low-confidence predictions, mixed languages, duplicates
- **Comparison mode** to contrast time periods or segments
- **Export** to CSV, JSON, or PDF
- **Dark mode** support

## Usage
1. Upload a file with text data (CSV, JSON, Excel)
2. Wait for analysis to complete (~30s for 50 entries)
3. Explore the dashboard tabs: Overview, Data Quality, Compare

**API Key**: Use `dev-key-1` (pre-configured in the UI)

## Tech Stack
- **Backend**: FastAPI, PyTorch, Transformers, BERTopic
- **Frontend**: React, TypeScript, Recharts, D3.js