Project Overview

The Moderated Discussion Board (MDB) in VU's LMS is frequently cluttered with non-academic responses such as "good," "done," "present," or phone numbers for WhatsApp groups. These messages reduce efficiency for faculty who must manually review hundreds of posts. This project proposes developing a browser-based tool integrated with the LMS front-end to automatically detect, filter, and optionally reply to non-academic messages. The solution will improve faculty productivity and maintain the MDB's academic integrity.

• Simulate MDB data using mock HTML pages or exported static content.
• Collect and label a dataset of sample MDB messages (academic vs. non-academic).
• Implement two filtering approaches: a keyword-based system and an AI-powered classifier then compare results.

Users and Roles

Admin:

• Manage global keyword list.
• Update AI models.
• Set default filtering policies.

Faculty Members:

• Use the tool for filtering MDB messages.
• Manage keyword list locally.
• Review and evaluate AI-classified results.

Functional Requirements

Dataset Creation Module:

• Collect at least 500–1,000 sample messages (synthetic or crowdsourced), labeled as academic or non-academic.
• Store datasets in CSV or JSON format.

Keyword-Based Filtering:

• Implement regex-based filtering for known patterns (good, done, present, sir, phone numbers).
• Provide a toggle to enable/disable keyword filtering.

AI/NLP-Based Classification:

• Use TF-IDF + Logistic Regression or Naïve Bayes for classification (Python + scikit-learn).
• Optionally experiment with BERT or DistilBERT for advanced filtering.
• Display model accuracy (precision, recall, F1).

Comparison Dashboard:

• Provide metrics comparing keyword filtering and AI classification accuracy.
• Allow faculty to review misclassified examples.

Mock LMS Integration:

• Build a static MDB interface (HTML/JS/CSS) to simulate the LMS environment.
• Inject filtering functionality via a browser extension or userscript.

Export Feature:

• Export filtered academic queries into CSV or text.

Tools / Languages / Frameworks

Expected Outcomes

1. A dual-filtering system: keyword-based for simplicity and ML-based for adaptability.
2. Demonstrates AI and front-end skills without requiring LMS backend access.
3. A reusable dataset of MDB-like messages for future research or improvements.