Automated Academic Peer Review Assistant

An AI-powered system for automated academic peer review.

This tool analyzes research papers to provide structured, reviewer-style feedback. It performs citation quality checks, novelty search, plagiarism detection, factual consistency analysis, and claim mapping, then synthesizes the results into a professional review report. Designed for researchers, educators, and institutions to accelerate the peer review process.

---

Key Features

• Automated Analysis Suite

  • Novelty Search: Retrieve and compare papers using FAISS + semantic embeddings.
  • Plagiarism Detection: Detect exact and paraphrase overlaps via semantic similarity + string matching.
  • Factual Checks: Validate numerical values and units for consistency and plausibility.
  • Claim Mapping: Extract and match scientific claims against prior publications.
  • Citation Alert (via GROBID): Parse references, check citation quality, and flag missing/incorrect citations.

• Enhanced Retrieval Features

  • Deep Search Mode: On demand, fetch up to N new papers (via ArXiv, Semantic Scholar, CrossRef), store locally, and rebuild FAISS index for fresh comparisons.

• LLM-Powered Review Synthesis

  • Structured Review Generation: Summarize findings into section-wise scores, strengths/weaknesses, claim novelty, and final recommendation.

• User-Friendly Interface

  • Web App (Flask): Upload a PDF, optionally enable deep search, and receive a detailed review report.
  • Exportable Results: Outputs JSON artifacts (novelty, plagiarism, claim mapping, factual checks, citations) and a consolidated review file.

---

Technology Stack

• Core Libraries: PyPDF2, requests, argparse, json, re
• NLP & Embeddings: sentence_transformers (all-MiniLM-L6-v2), faiss, scikit-learn
• Citation Parsing & Alerts: grobid (for PDF parsing + reference extraction)
• Claim & Factual Analysis: pint (unit normalization), regex-based claim extraction
• Web & UI: Flask, Jinja2
• LLM Integration (optional): google.generativeai (Gemini), groq, Hugging Face Inference API

---

Website Overview

---

Quick Start

1. Clone the repository and install dependencies:

   git clone https://github.com/BhaveshBhakta/Automated-Academic-Peer-Review-Assistant
   cd Automated-Academic-Peer-Review-Assistant
   pip install -r requirements.txt

2. Set up environment variables (for LLM integration): Create a .env file in the project root with:

   GEMINI_API_KEY=your_key_here
   GROQ_API_KEY=your_key_here
   HF_API_KEY=your_key_here

   Note: These are optional, but required for LLM-based review synthesis.

3. Run PDF parsing (extract text + citations using GROBID):

docker run -t --rm -p 8070:8070 lfoppiano/grobid:0.7.2

(Keep grobid running in a separate terminal.)

python utils/pdf_parse.py

4. Build FAISS index for similarity search:

   python utils/faiss_index.py \
       --pdf_dir data/pdfs \
       --index_path data/faiss_indexes/global_index.bin \
       --mapping_path data/faiss_indexes/global_mapping.json \
       --metadata_path data/metadata.json

5. Run the application:

   python app.py

6. Access the UI: Open http://localhost:5000 in your browser.

---

High-Level Architecture

User (Browser, PDF Upload)
        ↓
     Flask App
        ↓
 ┌─────────────── Pipeline ────────────────┐
 │  1. PDF Parsing (GROBID: text + citations) │
 │  2. Citation Alert (check missing refs)    │
 │  3. Novelty Check (FAISS + Embeddings)     │
 │  4. Plagiarism Detection                   │
 │  5. Factual Consistency Check              │
 │  6. Claim Extraction & Mapping             │
 │  7. Review Synthesis (LLM/Heuristics)      │
 └───────────────────────────────────────────┘
        ↓
   Structured Review Report + JSON Outputs

Deep Search Flow:

• If the user enables Deep Search:

  1. Fetch up to N new papers (ArXiv, Semantic Scholar, CrossRef).
  2. Save PDFs + metadata locally.
  3. Rebuild FAISS index with new data.
  4. Run the pipeline again with updated knowledge base.

---

Roadmap & Future Work

• Scalability: Containerize with Docker and add background workers for large-scale reviews.
• Improved Claim Extraction: Use advanced NLP/LLM models for precise claim detection.
• Richer Novelty Detection: Combine dense embeddings (FAISS) with sparse retrieval (BM25) for hybrid search.