This study introduces a novel approach to measuring interdisciplinarity using keyword extraction and graph-based analysis. Keywords, extracted via KeyBERT, are mapped onto a Discipline Knowledge Graph as nodes, where interdisciplinarity scores are computed by using the BFS algorithm. Evaluation with a classification task using the SPECTER model reveals that zero-score papers encounter fewer fields, while interdisciplinary papers show lower classifier confidence, highlighting their complex, boundary-spanning nature. This method offers a content-driven perspective on interdisciplinarity, addressing limitations of journal-based analyses.
The title and abstract texts of papers from the OpenAlex data
- building_graph.ipynb : Building a knowledge graph
- en_only_df.csv : the dataset used to build a knowledge graph -> too big to upload in github link
- en_inter_df.csv : the test dataset for the knowledge graph, and the score of this dataset is measured through the BFS algorithm. -> too big to upload in github link
- data_process.py : Data preprocessing to remove short texts and non-english texts
- final_test_data.csv : The dataset to predict the classification and model confidence
- final_train_df.csv : The dataset used to fine-tune the SPECTER model
- specter_dataset.py : The Dataset class to fine-tune the SPECTER model
- specter_main.py : The main functions to fine-tune the SPECTER model
- specter_model.py : The Classifier class to fine-tune the SPECTER model