PaperTrack-Agent

A Multi-Agent System for Automated Paper Tracking and Research Assistance

Features • Architecture • Installation • Quick Start • Usage • License

---

Demo

▶️ Click the image above to watch the demo video

---

Overview

PaperTrack-Agent is an AI-powered research assistant that leverages a multi-agent architecture to automate the entire research paper analysis workflow. From discovering the latest papers on arXiv to generating research ideas and experiment plans, this system provides comprehensive support for researchers.

Key Capabilities

• Automatic Paper Discovery: Track latest papers from arXiv based on your research fields
• Deep Paper Analysis: Extract structure, methodology, and key findings from PDFs
• Comparative Analysis: Compare multiple papers and identify common approaches
• Trend Analysis: Identify emerging research trends and hot topics
• Research Gap Finder: Discover unexplored research opportunities
• Writing Assistance: Generate paper outlines and related work sections
• Experiment Planning: Create detailed experiment plans with baseline recommendations

---

Features

10 Specialized Agents

Phase	Agent	Description
Discovery	Field Tracker	Search arXiv, expand queries with LLM, score relevance
Discovery	Paper Retrieval	Download PDFs, store metadata in SQLite
Analysis	Paper Reader	Parse PDFs, extract structure, analyze figures
Analysis	Code Extractor	Extract algorithms, formulas, find GitHub repos
Comparison	Comparative Analysis	Compare methods, results, identify themes
Comparison	Trend Analyzer	Analyze temporal trends, keyword evolution
Comparison	Citation Network	Build citation graphs, analyze impact
Innovation	Research Gap Finder	Identify gaps, generate research ideas
Writing	Writing Advisor	Generate outlines, related work sections
Writing	Experiment Planner	Create experiment plans, suggest baselines

Web Interface

A beautiful Streamlit-based UI with:

• Real-time pipeline progress tracking
• Interactive paper exploration with detailed analysis
• Trend visualization charts
• Research insight dashboard

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     PaperTrack-Agent System                      │
├─────────────────────────────────────────────────────────────────┤
│  Phase 1: Discovery        Phase 2: Analysis                    │
│  ┌─────────────────┐       ┌─────────────────┐                  │
│  │ Field Tracker   │──────▶│ Paper Reader    │                  │
│  │ Paper Retrieval │       │ Code Extractor  │                  │
│  └─────────────────┘       └────────┬────────┘                  │
│                                     │                            │
│  Phase 3: Comparison               ▼                            │
│  ┌─────────────────────────────────────────────┐                │
│  │ Comparative Analysis │ Trend Analyzer │ Citation Network    │
│  └─────────────────────────────────────────────┘                │
│                            │                                     │
│  Phase 4: Innovation      ▼           Phase 5: Writing          │
│  ┌─────────────────┐       ┌─────────────────┐                  │
│  │ Research Gap    │──────▶│ Writing Advisor │                  │
│  │ Finder          │       │ Experiment Plan │                  │
│  └─────────────────┘       └─────────────────┘                  │
└─────────────────────────────────────────────────────────────────┘

---

Installation

Prerequisites

• Python: 3.8 or higher
• Operating System: Linux, macOS, or Windows
• API Keys: DeepSeek API key (required), DashScope API key (optional, for vision analysis)

Step 1: Clone the Repository

git clone https://github.com/Pandakingxbc/PaperTrack-Agent.git
cd PaperTrack-Agent

Step 2: Create Virtual Environment (Recommended)

# Using venv
python -m venv venv
source venv/bin/activate  # Linux/macOS
# or
venv\Scripts\activate  # Windows

# Or using conda
conda create -n papertrack python=3.10
conda activate papertrack

Step 3: Install Dependencies

pip install -r requirements.txt

# Additional dependencies for PDF processing
pip install PyMuPDF dashscope

Step 4: Configure API Keys

Create a .env file in the project root:

cp .env.example .env

Edit .env and add your API keys:

# Required: DeepSeek API for text analysis
DEEPSEEK_API_KEY=your-deepseek-api-key-here
DEEPSEEK_BASE_URL=https://api.deepseek.com/v1

# Optional: DashScope API for vision analysis (figure understanding)
DASHSCOPE_API_KEY=your-dashscope-api-key-here

Getting API Keys

Provider	Purpose	How to Get
DeepSeek	Text analysis (DeepSeek-V3)	DeepSeek Platform
DashScope	Vision analysis (Qwen-VL)	Aliyun DashScope

---

Quick Start

Option 1: Web Interface (Recommended)

streamlit run app.py

Then open your browser at http://localhost:8501

Option 2: Command Line

# Demo mode - run full pipeline with default settings
python main.py --mode demo

# Quick mode - only run Phase 1 (paper discovery)
python main.py --mode quick

# Interactive mode - step-by-step execution
python main.py --mode interactive

# Custom run with specific parameters
python main.py \
    --fields "Multi-Agent Systems" \
    --time-range last_month \
    --max-papers 10 \
    --phases 1 2 3 4 5

---

Usage

Web Interface

1. Configure Search: Enter your research field in the sidebar
2. Set Time Range: Choose preset (7 days, 1 month, etc.) or custom range
3. Start Analysis: Click "Start Analysis" to run the pipeline
4. Explore Results:
   • Papers Tab: Browse discovered papers, click to expand details
   • Analysis Tab: View comparative analysis and method comparisons
   • Trends Tab: Explore keyword trends and technology evolution
   • Insights Tab: Discover research gaps and generated ideas

Programmatic Usage

from main import run_full_pipeline

# Run the complete pipeline
state = run_full_pipeline(
    research_fields=["Multi-Agent Systems", "LLM Agents"],
    time_range="last_month",
    max_papers_per_field=10,
    phases=[1, 2, 3, 4, 5]
)

# Access results
print(f"Found {len(state['tracked_papers'])} papers")
print(f"Generated {len(state['research_gaps']['research_ideas'])} research ideas")

Using Individual Agents

from src.agents import FieldTrackerAgent, ResearchGapFinderAgent

# Track papers in a field
tracker = FieldTrackerAgent(max_papers=10)
papers = tracker.track_papers(
    field="Multi-Agent Systems",
    time_range="last_month"
)

# Find research gaps
gap_finder = ResearchGapFinderAgent()
gaps = gap_finder.analyze(papers["papers"], field="Multi-Agent Systems")

# Print generated research ideas
for idea in gaps["research_ideas"]:
    print(f"- {idea['title']}")
    print(f"  Feasibility: {idea['feasibility']}")

---

Configuration

Main Configuration (configs/config.yaml)

# LLM Models
models:
  text_analysis:
    provider: deepseek
    model: deepseek-chat
    temperature: 0.1
    max_tokens: 8192

  vision_analysis:
    provider: qwen
    model: qwen3-vl-flash

# arXiv Settings
arxiv:
  max_results_per_field: 20
  sort_by: submittedDate

# Agent Settings
agents:
  field_tracker:
    enabled: true
    expand_query: true

  paper_reader:
    extract_figures: true
    max_figures_per_paper: 10

  research_gap_finder:
    generate_ideas: true
    max_ideas: 8

---

Output Structure

PaperTrack-Agent/
├── papers/                      # Downloaded PDFs and metadata
│   └── {arxiv_id}/
│       ├── paper.pdf
│       └── metadata.json
├── paper_analysis/              # Deep analysis results
│   └── {arxiv_id}/
│       └── analysis.json
├── code_extracts/               # Extracted algorithms and code
├── comparative_analysis/        # Method and result comparisons
├── trend_analysis/              # Keyword and technology trends
├── citation_network/            # Citation graph data
├── research_gaps/               # Identified gaps and ideas
├── writing_outputs/             # Generated outlines and drafts
├── experiment_plans/            # Experiment configurations
├── reports/                     # Field tracking reports
└── outputs/                     # Pipeline state snapshots

---

Tech Stack

Component	Technology
Multi-Agent Framework	LangGraph
Text Analysis LLM	DeepSeek-V3
Vision Analysis LLM	Qwen-VL-Flash
PDF Processing	PyMuPDF
Metadata Storage	SQLite
Vector Database	ChromaDB
Citation Data	Semantic Scholar API
Web Interface	Streamlit
Visualization	Plotly

---

Roadmap

• Phase 1: Paper Discovery (Field Tracker + Paper Retrieval)
• Phase 2: Deep Analysis (Paper Reader + Code Extractor)
• Phase 3: Comparative Analysis (Comparison + Trends + Citations)
• Phase 4: Research Innovation (Gap Finder)
• Phase 5: Writing Assistance (Writing Advisor + Experiment Planner)
• Web Interface with Streamlit
• Multi-language support
• PDF export for reports
• Integration with reference managers (Zotero, Mendeley)
• Real-time paper alerts

---

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Author

Yang Zhi - @Pandakingxbc

---

Acknowledgments

• DeepSeek for the powerful LLM API
• arXiv for the open access to research papers
• Semantic Scholar for citation data
• All the open-source libraries that made this project possible

---

Made with ❤️ for the research community