HEP-Th Citation Network Analysis: Bridging vs. Siloing

Research Question

Do influential papers in High Energy Physics Theory (HEP-Th) bridge diverse research communities, or do they reinforce disciplinary silos?

Overview

This project analyses the citation network from arXiv's High Energy Physics Theory (HEP-Th) category to understand whether influential papers serve as bridges connecting different research communities or as pillars reinforcing disciplinary boundaries.

Key Findings

Main Conclusion

Influential papers predominantly reinforce silos rather than bridging diverse research communities.

Summary Statistics

• Network Size: 27,770 papers, 352,807 citations
• Community Structure: High modularity (0.51) with 3 dominant communities containing 89% of papers
• Cross-Community Citations:
  • Influential papers: ~15-18%
  • Non-influential papers: ~12-13%
  • Over 80% of citations stay within communities for all paper types

Key Insights

1. Strong Disciplinary Boundaries: Three major research communities with clear intellectual separation
2. Limited Bridging: Even the most influential papers derive over 80% of citations from within their own community
3. Types of Influence:
   • High betweenness papers: Best bridging (18%) - structural connectors
   • High PageRank papers: Elite influence within narrow circles
   • High in-degree papers: Most cited, reinforce community boundaries
4. Structural vs. Direct Bridging: Weak negative correlation between betweenness centrality and cross-community citations suggests structural bridge positions don't translate to direct cross-boundary citations

Methodology

Analysis Pipeline

1. Load Network & Basic Statistics

   • Load HEP-Th citation network as directed graph
   • Compute nodes, edges, density, degree distributions

2. Identify Influential Papers (Top 10%)

   • In-degree centrality (most cited)
   • PageRank (quality-weighted citations)
   • Betweenness centrality (structural bridges)

3. Community Detection

   • Modularity maximisation (greedy algorithm)
   • Assign papers to communities
   • Measure network modularity

4. Measure Bridging Behaviour

   • Cross-community citation ratio = (citations to different communities) / (total citations)
   • Calculate separately for in-citations and out-citations

5. Compare Groups

   • Influential vs. non-influential papers
   • Different types of influence (in-degree, PageRank, betweenness)

6. Visualise Results

   • Box plots: Cross-community ratios by paper type
   • Bar charts: Mean bridging behaviour
   • Scatter plot: Betweenness vs. bridging behaviour

Installation

Requirements

pip install -r requirements.txt

Required packages:

• networkx>=3.0
• numpy>=1.20
• matplotlib>=3.5
• scipy>=1.7

Quick Start

# Clone the repository
git clone https://github.com/dhruva-divate/hep-citation-analysis.git

cd hep-citation-analysis

# Install dependencies
pip install -r requirements.txt

# Run the analysis
python hep_citation_analysis.py --data cit-HepTh.txt --output-dir results

Usage

Basic Usage

python hep_citation_analysis.py

This assumes cit-HepTh.txt is in the current directory and will output results to the current directory.

Advanced Options

python hep_citation_analysis.py \
    --data path/to/cit-HepTh.txt \
    --output-dir results/ \
    --top-percent 10

Arguments:

• --data: Path to citation network edge list file (default: cit-HepTh.txt)
• --output-dir: Directory for output files (default: current directory)
• --top-percent: Percentage of top papers to consider influential (default: 10)

Output Files

The script generates:

• bridging_analysis.png: Comprehensive visualisation with 4 subplots
• Console output: Detailed statistics and interpretations for each analysis step

Data

Source

• Dataset: SNAP: cit-HepTh
• Description: Citation network from arXiv HEP-Th (High Energy Physics - Theory)
• Format: Edge list (tab or space separated)
• Size: 27,770 papers, 352,807 citation links
• Time Period: Papers from January 1993 to April 2003
• cit-HepTh.txt # Citation network data (IS INCLUDED IN THIS REPO)

Data Format

# Source  Target
9201234   9301456
9301456   9401123
...

Download

wget http://snap.stanford.edu/data/cit-HepTh.txt.gz
gunzip cit-HepTh.txt.gz

Results

Visualisations

The visualisation includes four key plots:

1. Top-left: In-citations cross-community ratios (who cites this paper)
2. Top-right: Out-citations cross-community ratios (what this paper cites)
3. Bottom-left: Mean bridging behaviour by paper type
4. Bottom-right: Betweenness centrality vs. cross-community citation ratio

Statistical Summary

Paper Type	In-Citations Mean	In-Citations Median	Out-Citations Mean	Out-Citations Median
High Betweenness	18.5%	11.1%	18.3%	11.1%
High PageRank	15.1%	9.4%	17.0%	3.5%
High In-Degree	14.7%	9.3%	17.3%	10.5%
Non-Influential	13.3%	0.0%	11.9%	0.0%

Key Observation: Median of 0% for non-influential papers means over half have NO cross-community citations.

Technical Details

Logical Flow Review

The code follows a sound logical progression:

1. Network Loading: Proper directed graph construction from edge list
2. Centrality Measures: Three complementary measures capture different types of influence
3. Community Detection: Modularity maximisation on undirected graph
4. Bridging Calculation: Separate in/out calculations with proper handling of edge cases
5. Statistical Comparison: Comprehensive comparison across multiple paper groups
6. Visualisation: Clear multi-panel figure

Performance Notes

• Betweenness Centrality: Computationally expensive O(n³), script uses sampling (k=5000 nodes)
• Memory Usage: ~1-2 GB for full network
• Runtime: ~2-5 minutes on modern hardware
• Scalability: Can handle networks with 100K+ nodes with sampling

Citation

If you use this analysis in your work, please cite:

@misc{hep_citation_analysis_2025,
  author = {Dhruva Divate},
  title = {HEP-Th Citation Network Analysis: Bridging vs. Siloing},
  year = {2025},
  publisher = {GitHub},
  url = {https://github.com/dhruva-divate/hep-citation-analysis}
}

Original dataset:

@misc{snapnets,
  author = {Jure Leskovec and Andrej Krevl},
  title = {{SNAP Datasets}: {Stanford} Large Network Dataset Collection},
  howpublished = {\url{http://snap.stanford.edu/data}},
  month = jun,
  year = 2014
}

Licence

MIT Licence - see LICENCE file for details

Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

Future Work

Potential extensions of this analysis:

1. Temporal Analysis: How has bridging behaviour evolved over time?
2. Content Analysis: Do paper abstracts/titles reveal topical differences between communities?
3. Author Networks: Do author collaboration patterns mirror citation patterns?
4. Comparative Study: Compare HEP-Th with other arXiv categories
5. Intervention Studies: What factors promote cross-community citation?

Acknowledgements

• Stanford Network Analysis Project (SNAP) for providing the dataset
• arXiv for making scientific papers freely available
• NetworkX development team for excellent graph analysis tools