ChatShieldNLP (PyQt6 + OCR)

A desktop app to detect inappropriate or unsafe messages in text or screenshots.
Built with PyQt6 for the GUI and Tesseract OCR for extracting text from images.
Designed to be lightweight, explainable, and offline (no cloud/ML dependency).

---

ChatShieldNLP

Rule-based desktop application for detecting inappropriate or unsafe messages using NLP preprocessing + regex scoring.
Built with PyQt6 GUI, pytesseract OCR, and lightweight rule weights.

---

Demo

---

Features

• Input options
  • Paste or type text
  • Upload a screenshot → OCR → analyze
• Scoring system with transparent phrase matches (e.g., ask_send_pic)
• Filter levels
  • Easy (standard, ≥ 0.55)
  • Medium (strict, ≥ 0.30)
• Categories covered
  • Requests for images or sensitive content
  • NSFW/explicit vocabulary
  • Coercion & manipulative persistence (“don’t ignore me”)
  • Unsolicited advances/inappropriate openers
  • Harassment (slurs, insults, threats, doxxing, stalking)
• Conversation tracking – recent scores are accumulated in AppState
• Runs fully offline once Tesseract is installed
• Text or image input (OCR via Tesseract).
• Rule‑based safety scoring with intensity filter (Easy=0.55, Medium=0.30).
• Non‑blocking UI (QThread workers) with analyzing spinner + results dialog.
• Customizable background color (persisted with QSettings).
• Windows‑first setup with auto Tesseract path detection (fallback manual path).

---

Quickstart

# 1) Clone & enter
git clone https://github.com/sp2023lab/ChatShieldNLP.git
cd ChatShieldNLP

# 2) Create venv
python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
# source venv/bin/activate

# 3) Install deps
pip install -r requirements.txt

# 4) (Windows) Install Tesseract OCR:
#   https://github.com/UB-Mannheim/tesseract/wiki

# 5) Run
python main.py

---

How it works (tl;dr)

• Normalize text (unicode fold, leetspeak cleanup), then run regex rules.
• Each rule contributes a weight; total score ∈ [0,1].
• Intensity filter gates results: Easy≥0.55, Medium≥0.30.
• OCR runs in a worker thread; UI stays responsive.

---

📦 Requirements

• Python 3.11+
• Tesseract OCR (for screenshots, optional if you only analyze text)
  • Windows typical paths:
    • C:\Users\<you>\AppData\Local\Programs\Tesseract-OCR\tesseract.exe
    • C:\Program Files\Tesseract-OCR\tesseract.exe
• Python packages:
  • PyQt6
  • pytesseract
  • Pillow

---

Troubleshooting

pytesseract not found
Activate your venv and run:

pip install pytesseract Pillow

---