Code Plagiarism Checker

FastAPI-based backend for code similarity/plagiarism scoring using preprocessing + TF-IDF + cosine similarity.

Features

• Accepts code and language id through an HTTP API.
• Removes comments based on language style.
• Creates a temporary source file in temp/.
• Tokenizes code and vectorizes with TfidfVectorizer.
• Computes similarity score using cosine similarity (returned internally as percentage).

Tech Stack

• Python 3.12+
• FastAPI
• scikit-learn

Project Structure

.
├── main.py                    # FastAPI app and routes
├── base/model.py              # TF-IDF vectorization + similarity search
├── model/code.py              # Request body schema (Pydantic)
├── modules/preprocess.py      # Language-aware comment removal + token prep
├── modules/genrate_code_file.py # Writes submitted code to temp file
├── temp/                      # Generated temporary source files
├── setup.bash                 # Create virtual environment (uv)
└── activate.bash              # Activate virtual environment

Supported Languages

Language ids used by the API:

• 0 → C
• 1 → C++
• 2 → Java
• 3 → JavaScript (node)
• 4 → Rust
• 5 → Python

Setup

1) Create virtual environment

./setup.bash

2) Activate environment

source .venv/bin/activate
# or
./activate.bash

3) Install dependencies

pip install -e .

Run Server

fastapi dev main.py

App runs at http://127.0.0.1:8000 by default.

API

GET /

Health check/sample route.

Response:

{
	"Hello": "World"
}

POST /

Request body:

{
	"language": 0,
	"code": "#include <stdio.h>\nint main(){return 0;}"
}

Current response:

{
	"message": "Done"
}

How Scoring Works (Current Implementation)

1. Input code is preprocessed (comment removal by language).
2. Cleaned code is saved in temp/ with language extension.
3. File content is split into tokens and used as corpus.
4. TF-IDF matrix is built from corpus.
5. Cosine similarity score is computed against a query snippet.

Current Limitation

The current POST / route computes a score internally but uses a hardcoded query snippet and does not return the score in the API response yet.

License

This project is licensed under the MIT License.