Shall I Work with Them? A Knowledge Graph-Based Approach for Predicting Future Research Collaborations

Area of Science:

• Computer Science
• Information Science
• Bibliometrics

Background:

• Predicting future research collaborations is crucial for scientific advancement.
• Existing methods often focus solely on structural information within scientific knowledge graphs.
• Integrating diverse data sources can enhance the accuracy of collaboration prediction.

Purpose of the Study:

• To investigate the impact of integrating unstructured textual data into a scientific knowledge graph for link prediction.
• To evaluate the effectiveness of graph kernels and machine learning models for collaboration prediction.
• To propose a novel pipeline for leveraging structural, textual, and pre-trained word embeddings.

Main Methods:

• Developed a three-phase pipeline combining graph algorithms and natural language processing techniques.
• Integrated structural information with textual data from a scientific knowledge graph.
• Utilized pre-trained word embeddings and evaluated performance using accuracy, recall, and precision.

Main Results:

• The integration of textual data significantly improved link prediction performance.
• The proposed pipeline demonstrated superior results compared to classical link prediction algorithms.
• Experiments on the COVID-19 Open Research Dataset confirmed the approach's effectiveness.

Conclusions:

• Combining structural and textual information in scientific knowledge graphs enhances collaboration prediction.
• The proposed pipeline offers a robust method for predicting future research partnerships.
• This approach has significant implications for understanding and fostering scientific collaboration.