ANALYSIS OF SCIENTIFIC TEXTS FOR IDENTIFYING POTENTIAL CO-AUTHORS

Abstract

The article analyzes modern approaches to finding scientific co-authors, particularly methods based on text data analysis. It discusses the key drawbacks of traditional methods, such as the use of scientific databases and academic social networks, which limit the accuracy and completeness of the search for potential partners. It is noted that these methods often focus only on explicit links between researchers, overlooking hidden communities or less obvious intersections of scientific interests. The potential of using text mining and large language models (LLM) to identify researchers with relevant expertise is emphasized. These methods can automatically extract key concepts from scientific papers, identify shared research interests, and uncover potentially productive scientific pairings. The advantages of these methods include the ability to identify hidden connections between researchers and provide deeper contextual analysis. Furthermore, the possibility of applying these methods to discover new interdisciplinary research directions is highlighted. The article also addresses the optimization of LLM to reduce computational resource costs and environmental impact. The necessity of using less resource-intensive architectures and algorithms with increased computational efficiency is emphasized. The importance of combining traditional statistical approaches with modern language models to enhance the effectiveness of recommendations is underlined. Methods of adapting LLM to the specifics of scientific texts to improve the quality of results are also discussed. The results suggest a significant potential of these methods in solving the problem of identifying scientific co-authors in modern open science systems. Additionally, the article outlines prospects for further research in combining various text analysis approaches to create comprehensive solutions in the field of scientometrics.