README.md

metadata

title: Review Intelligence Platform
colorFrom: blue
colorTo: yellow
sdk: docker
app_port: 7860
pinned: false
preload_from_hub:
  - cardiffnlp/twitter-xlm-roberta-base-sentiment
  - MoritzLaurer/multilingual-MiniLMv2-L6-mnli-xnli
  - sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2

Review Intelligence Platform

Review Intelligence is a desktop-first review analysis workspace that turns noisy public feedback into a readable report. It accepts a single source or a two-source comparison, runs a multilingual analysis pipeline, and returns a structured dashboard with evidence, summaries, and exports.

What Was Built

• Single-source analysis flow for YouTube, Play Store, Amazon, CSV, and raw review text
• Compare flow for side-by-side source evaluation
• Live progress timeline with background job tracking
• Session-based report history in the browser
• JSON export and server-side PDF export
• Desktop-focused report UI with source profiles, charts, evidence blocks, and comparison views

Supported Sources

• YouTube
  • real video metadata
  • real top-level comments through the YouTube Data API
• Play Store
  • app metadata
  • live review fetch through google-play-scraper
• Amazon
  • product metadata
  • live public review snippets from product pages
• CSV
  • secure CSV parsing with path, size, and row limits
• Raw text
  • direct review blocks or pasted feedback

Analysis Layers

• Multilingual sentiment analysis
• Emotion classification
• Comment-type separation
  • opinion
  • reaction
  • lyric quote
  • personal story
  • spam/noise
• Weighted sentiment scoring using review engagement signals
• Aspect insights with evidence examples
• Semantic conversation clusters
• Executive summary
• Trust and confidence notes
• Trend timeline when dated review data exists
• Language mix detection across multiple scripts

Report Output

Each report can include:

• source profile and metadata
• executive summary
• report card and score dimensions
• sentiment and emotion visuals
• semantic topic clusters
• aspect insights
• representative comments
• separated comment-type samples
• trust overview
• analyst brief
• downloadable JSON and PDF reports

Compare Output

Compare mode builds two full analyses and then derives:

• overall winner
• weighted sentiment edge
• score rows across key dimensions
• overlap vs unique themes
• overlap vs unique aspects
• one-line comparison verdict
• direct links back to the full left and right reports

Technical Shape

• Frontend
  • React
  • Vite
  • Tailwind CSS
  • Recharts
• Backend
  • FastAPI
  • async background job flow
  • SSE progress streaming
  • in-memory or Redis-backed session storage
• NLP / ML
  • cardiffnlp/twitter-xlm-roberta-base-sentiment
  • MoritzLaurer/multilingual-MiniLMv2-L6-mnli-xnli
  • sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
• Export
  • server-generated PDF via ReportLab

Project Characteristics

• Desktop-first interface
• Multilingual review handling
• Background analysis with progress updates
• Structured logging
• Targeted backend test suite
• Hugging Face Docker Space compatible

Current Constraints

• Amazon support depends on public page structure and can be less stable than API-backed sources
• Sessions are temporary by design
• First model load can be slower on a fresh environment
• The interface is intentionally desktop-first, not mobile-optimized

License

MIT