ClaimCheck

Does this seller deliver what they promise?

ClaimCheck compares what sellers claim in their product descriptions against what customers actually experience in their reviews, to help buyers verify their next purchase before placing an order.

Beyond helping buyers decide, ClaimCheck pushes sellers toward transparency and accountability.

Built for the Nigerian e-commerce market where "what I ordered vs what I got" is still one of the most common frustrations on platforms like Jumia.

---

The Problem

Nigeria's e-commerce market is projected to reach $22.9B by 2030. But behind that growth, the most common and persistent challenge remains:

"What I ordered vs what I got"

Customers receive products that don't match their expectations. Over time this affects something deeper than just a bad purchase —

trust.

And once trust drops, so does purchase intention.

ClaimCheck bridges that gap.

---

How It Works

Paste any Jumia product URL
              ↓
Scrape product name, description, rating and reviews
              ↓
Extract every claim the seller makes
              ↓
Run aspect-based sentiment analysis on customer reviews
              ↓
Compare seller claims against real customer experiences
              ↓
Generate a clear breakdown with trust scores and verdicts

---

Features

Aspect Verdict Cards Every product aspect the seller describes gets its own card showing their exact claims, real customer sentiments, and a verdict on whether it holds up.

Four verdicts:

• ✓ CONFIRMED — customers back the claim
• ~ MIXED — customers are split
• ? UNVERIFIED — not enough review data to judge
• ✕ CONTRADICTED — customers say otherwise

Trust Score A 0–100 score computed from all verified aspect verdicts, weighted by how strongly each verdict is supported.

Unclaimed Aspects The most unique feature. ClaimCheck surfaces aspects sellers never mention in their descriptions but customers repeatedly bring up in reviews, whether it's a flaw they're hiding or a strength they forgot to mention.

Confidence-weighted verdicts A single negative review can't contradict a seller claim. ClaimCheck requires a minimum of 2 reviews per aspect and a 75% sentiment majority before committing to a verdict, making the output trustworthy rather than reactive.

---

Tech Stack

Layer	Tool
Scraping	Playwright + BeautifulSoup
NLP preprocessing	spaCy
LLM inference	Groq (llama-3.3-70b-versatile)
Frontend	Streamlit
Language	Python 3.11+

---

Project Structure

claim_check/
├── .streamlit/ 
│      └── config.toml                      # Streamlit theme and configuration
│
├── prompts/
│   ├── __init__.py
│   ├── extract_sellers_claim_prompt.py     # Prompt for extracting seller claims from descriptions
│   ├── grammar_fix_prompt.py               # Prompt for cleaning and fixing description grammar
│   └── review_analysis_prompt.py           # Prompt for running aspect-based sentiment analysis
│
├── src/
│      ├── app.py                           # Streamlit UI and application entry point
│      ├── comparator.py                    # Compares seller claims against customer sentiments
│      ├── extractor.py                     # Extracts seller claims and runs sentiment analysis on reviews
│      └── scraper.py                       # Scrapes product details and reviews from Jumia
│
├── .dockerignore                           # Files and folders excluded from the Docker build context
├── .gitignore                              # Files and folders excluded from version control
├── .python-version                         # Pinned Python version managed by uv
├── Dockerfile                              # Container configuration for production deployment
├── LICENSE                                 # MIT License
├── README.md                               # Project documentation
├── pyproject.toml                          # Project metadata and dependencies
└── uv.lock                                 # Dependency lockfile

---

Setup

1. Clone the repo

git clone https://github.com/s4nni_yusuf/claimcheck.git
cd claimcheck

2. Install dependencies

uv sync

3. Install Playwright browser

uv run playwright install chromium

4. Install spaCy language model

uv run python -m spacy download en_core_web_sm

5. Create a .env file in the root directory and add:

GROQ_API_KEY_1=your_groq_api_key_here

You can get a free Groq API key at console.groq.com.

6. Run the app

uv run streamlit run src/app.py

---

Demo

Demo Video on Youtube

---

Feedback

Especially from anyone who's been humbled by a bad Jumia order.

Open an issue, submit a PR, or reach out on X @s4nni_yusuf.

If you find ClaimCheck useful, consider giving it a star ⭐. It helps more people find the project.

---

License

MIT - See LICENSE