Scholar Team Finder - link prediction model for identifying scholars in academic social networks
Abstract
[EMBARGOED UNTIL 5/1/2024] Collaborations between scholars from multiple fields are becoming more common in research tasks. However, identifying suitable co-workers can be a challenging and time-consuming process. In response to this, the authors propose a novel model called ScholarTeamFinder, which utilizes a knowledge graph to identify collaborators within an academic social network (ASN) for multi-disciplinary research tasks. The model uses graph-based deep learning to learn node embeddings from the knowledge graph and recommends a scholar team. The approach uses semantic text features to improve the link prediction for identifying a suitable team. The authors evaluate the ScholarTeamFinder using large ASN datasets, including the NSF award dataset of federal grant awards, scholars' publication data, and two other widely used datasets. They also propose a beam-search algorithm for scholar team prediction based on the model. The results show that the ScholarTeamFinder outperforms state-of-the-art baseline models by approximately 15 percent across different datasets. The ScholarTeamFinder project aims to improve the collaboration process for researchers by providing recommendations for potential collaborations with related scholars. However, the model has not been trained with a knowledge graph, which limits its functionality, usability, and accuracy. Future work includes rebuilding the model with a knowledge graph to better represent relationships between scholars, venues, and publications, expanding the recommendations to include publications, venues, and other useful points, combining user queries with data models, and integrating expanding data models with a science gateway for ease of use. To ensure the accuracy of the model, data collection would involve gathering scholarly publications, conference proceedings, and related data sources to create a comprehensive knowledge graph that can be integrated with the existing model. Additionally, collecting data on user queries and interactions with the model will help ensure that it is effectively meeting the needs of its users. In summary, the ScholarTeamFinder model addresses the challenge of identifying suitable collaborators for multi-disciplinary research tasks. It utilizes a knowledge graph and graph-based deep learning to recommend a scholar team. The model outperforms baseline models by approximately 15 percent across different datasets, and future work includes expanding the recommendations and integrating data models with a science gateway for ease of use.
Degree
M.S.