Origify

Overview

Origify is a plagiarism detection tool built to maintain academic integrity in state education. The goal is to create a reliable system that identifies plagiarism in exam papers, ensuring every assessment remains authentic, transparent, and of the highest quality. By addressing academic dishonesty, Origify fosters a learning environment rooted in honesty, openness, and excellence.

Motives

• Authenticity, transparency, and quality in education
• Personal experience with the education system

Objective

• Promoting fairness and integrity in state education

Architecture

Data Processing

1. Document Text Preprocessing:

   • Tokenization: Breaking text into individual tokens (words, punctuation, etc.) using nltk.word_tokenize.
   • Stop Word Removal: Filtering out common words (e.g., "the", "a", "is") using NLTK’s stopwords corpus.
   • Stemming/Lemmatization: Reducing words to their base form using PorterStemmer from NLTK.

2. Document Vectorization:

   • TF-IDF (Term Frequency-Inverse Document Frequency): Converts text into numerical vectors that represent word importance across documents using sklearn.feature_extraction.text.TfidfVectorizer.

   Formula:

   • Term Frequency:
     

   • Inverse Document Frequency:
     
     Where:

     • ( N ) is the total number of documents in the corpus.
     • ( df(t,D) ) is the number of documents containing term ( t ) in the corpus ( D ).

   • TF-IDF Score:
     
     A higher TF-IDF score indicates the term is more relevant to the document compared to the entire corpus.

3. Similarity Calculation:

   • Cosine Similarity: Measures the similarity between TF-IDF vectors of the query document and existing documents using sklearn.metrics.pairwise.cosine_similarity.

TODOs

• Logo
• UI & UX design
• Read about TF-IDF
• Save & Load Model
• Convert to Text
  • Image to text (OCR)
  • PDF to text
• Data scraping
  • Ency
  • Degrees
    • CEM
    • BEM
    • BAC
  • Dzexams (this was easy)
  • Setup a sub-process to clean duplicated data from Ency & Dzexams
• Plain-Text Extraction
  • Extract and save texts
  • Stream text extraction to preprocessing & vectorization
• Preprocess data
  • Save preprocessed text
• Vectorize data using TF-IDF
• Setup database
• Setup FastAPI /compare endpoint authentication
• Implement manual exam data submission via /upload (maybe)
  • Check if uploaded exams already exist
• Setup an HTTP endpoint for running the data mining process (admin)
• Wait-listing system
• Payment Gateway Integration
  • Stripe
  • CIB/EDAHABIA

Features

• Users select the subject of the exam to optimize resource usage
• Shareable interactive links/documents/badges (inspired by Spotify's shareable images)
• Teacher scoring system (users enter the exam provider’s information)

Developer Tech Stack

Frontend

• React Native
• Logsnag - Event tracking
• Sentry - Performance monitoring & error tracking
• React Native Document Scanner - Camera vision
• Lucide - Icons & fonts

Backend

• Python
• HTTPX
• selectolax
• Playwright
• FastAPI
• sklearn
• nltk
• Betterstack

Data Sources

• Dzexams
• Ency Education

Design Prototypes

Wireframe

Mobile User Flow

Marketing

• Social Media - Origify Beta Short Form Ad
  • Reddit
  • TikTok
  • Instagram
  • Facebook Groups (education-related)
• Public Posters
• Direct messages to students & teachers
• Email outreach (inspired by Arc Browser’s copywriting)

Monetization

• Limited requests per week, with a pay-to-use credits system

Papers

• A research paper documenting the problem, solution, and methodology (inspired research paper)
• An analysis of how the availability of educational exam papers online reflects the decline of the Tamazight language (TODO: replace "decline" with a more appropriate term)