FOREST | Team Website

About the Project

Transforming unstructured video opinions into navigable, verifiable argument graphs.

FOREST is a research-focused argument analysis system for political opinions and current affairs. Users post short face-camera opinion videos on trending topics. Other users reply with their own videos, creating a branching debate tree. Our AI pipeline transcribes each video, classifies its stance as FOR, AGAINST, or UNRESOLVED, and links every classification to an exact transcript excerpt and timestamp — making the entire discussion structure traceable and auditable.

Project Roadmap

Our phased approach to building FOREST.

Phase 1 — Define & Annotate

Focus on political-opinion videos, collect 20 videos, and create human-annotated ground truth using FOR / AGAINST / UNRESOLVED labels.

Phase 2 — AI Argument Recovery

Transcribe videos using OpenAI Whisper and identify stances with exact timestamps and evidence using Gemini/GPT-4o API.

Phase 3 — Build Debate Graph

Convert relationships into a Root → Reply → Reply argument tree and visualize the discussion as an interactive graph.

Phase 4 — Evaluate the System

Compare FOREST with transcript/summary-based baselines using Precision, Recall, F1, stance accuracy, and Fleiss' Kappa.

Phase 5 — Full Platform

Add community verification, credibility scoring, moderation, recommendations, and private collaborative trees.

Our Team

Group 1 • UCS503 Software Engineering Lab

Sahibnoor Singh

Lead Backend & AI Engineer

Role: Spring Boot Server & Gemini AI Master. Sets up the backend, writes prompts for the Gemini API, extracts JSON metadata, and creates the REST APIs.
Qualifications: Strong backend engineering skills and system architecture experience.

Yaksh

Graph Architect & Frontend Developer

Role: Neo4j Database & Frontend UI. Handles the graph database architecture (Cypher queries) and builds the client-facing UI to visualize the debate tree.
Qualifications: Strong domain knowledge in conceptual product design and logic structuring.

Ridhima

Python Data Scripter

Role: Raw Data Extraction. Writes Python scripts to strip audio from videos and leverages the OpenAI Whisper API for transcript generation.
Qualifications: Python scripting and basic API integration skills.

Jasmine Tiwana

Data Operations & QA

Role: Ground Truth Engineering. Manually collects dataset videos, creates ground-truth excel sheets, and conducts QA testing.
Qualifications: Detail-oriented data entry and manual testing.

Current Status

Active Planning Phase — Preparing Planning Presentation v1