Keyword-based Automatic Research Structurization (KARS)

A tool that automatically extracts keywords from scientific literature based on bibliographic metadata, structures the research field, and analyzes research trends. It presents spatial structures and temporal flows of research.

Table of Contents

• Features
• Installation
• Quick Start
• Example Dataset: ReRAM Present and Future
• Results
• Citations
• Contributors

Features

• Automatic keyword extraction from bibliographic data
• Construction of modularized keyword networks (PageRank for importance, Louvain modularity for communities)
• Temporal trend analysis across years and communities
• Interactive HTML visualizations and Gephi export

Installation

1. Clone the repository

git clone https://github.com/khyeon-cnmd/KARS.git
cd KARS

2. Create and activate a Conda environment (Python 3.10)

conda create -n KARS python==3.10
conda activate KARS

3. Install Python dependencies

pip install jsonlines gradio==3.47.0 "networkx[default]" tqdm pandas scipy bokeh spacy

4. Download spaCy models

python -m spacy download en_core_web_sm
python -m spacy download en_core_web_trf

Quick Start

Run the GUI:

python KARS_GUI.py

1. DB construction

• Retrieve the bibliographic metadata using your preferred engines (e.g., Web of Science, Crossref)
• Make folder structure as below (refer to the image)

DB_name
└── 1
    └── KBSE.json
└── 2
    └── KBSE.json
└── ...

• Convert each metadata into the json format as in KBSE.json (see example_metadata.json for reference)
• Please refer to the example DB structure of 'ReRAM_DB.tar'

2. load_DB

• Enter the path to the database directory that contains bibliographic metadata, then click Submit to load.

3. keyword_extraction

• UPoS model:
  • en_core_web_sm — Faster, lighter.
  • en_core_web_trf — More accurate, slower.
• Text source:
  • title — Extract keywords from article titles.
  • abstract — Extract keywords from article abstracts.

4. network_construction

• Click Submit to build a keyword co-occurrence network across articles.

5. research_trend_analysis

• Keyword frequency cutoff (%) — Filter out infrequent keywords.
  • Higher → fewer, larger keywords; may miss relevant terms.
  • Lower → more, smaller keywords; includes rarer terms.
• Community min-frequency weight — Filter out less mature communities.
  • Higher → fewer communities.
  • Lower → more communities.
• Year range — Set minimum and maximum years to analyze.
• PLC maturity range — Select start and end stages:
  • Development, Introduction, Growth, Maturity, Decline

Results

1. KARS.gexf, KARS_community.gexf

   • Keyword importance (node size) via PageRank and communities (node color) via Louvain modularity, viewable in Gephi.

   

2. research_maturity.html

   • Yearly keyword counts for the entire community to assess research maturity based on the Product Life Cycle (PLC) model.

   

3. community_year_trend.html

   • Evolution of keyword distributions by community over years, useful for trend analysis.

   

4. keyword_evolution.html

   • Proportional changes of top keywords over time by community maturity level.

   

Example Dataset: ReRAM Present and Future

The provided ReRAM_DB.tar contains bibliographic metadata and structured research data used in the study “A keyword-based approach to analyzing scientific research trends: ReRAM present and future.”

Dataset contents:

• ReRAM_DB/KARS/metadata_source.csv — Original metadata
• ReRAM_DB/KARS/network_article.gephi — Original keyword network
• ReRAM_DB/database — Metadata converted for this code

Note: PageRank and modularity results may differ from those computed in Gephi.

How to test with the example data:

tar -xf ReRAM_DB.tar
python KARS_GUI.py
# In the GUI, set the extracted ReRAM_DB directory in "load_DB",
# then run: keyword_extraction → network_construction → research_trend_analysis

Citations

Kim, H., Kim, S.H., Kim, J. et al. A keyword-based approach to analyzing scientific research trends: ReRAM present and future. Sci Rep 15, 12011 (2025). https://doi.org/10.1038/s41598-025-93423-5

Contributors

• Conceptualization: Hyeon Kim, Donghwa Lee
• Data Curation: Hyeon Kim
• Formal Analysis: Hyeon Kim, Donghwa Lee
• Funding Acquisition: Donghwa Lee
• Investigation: Hyeon Kim, Eun Ho Kim, Jun Hyeong Gu, Donghwa Lee
• Methodology: Hyeon Kim, Seong Hun Kim, Jaeseon Kim, Donghwa Lee
• Project Administration: Hyeon Kim, Donghwa Lee
• Resources: Donghwa Lee
• Software: Donghwa Lee
• Supervision: Donghwa Lee
• Validation: Hyeon Kim, Donghwa Lee