Examining influential factors for acknowledgements classification using supervised learning.

Area of Science:

• Bibliometrics
• Computational Linguistics
• Scientific Communication

Background:

• Acknowledgements are crucial for understanding contributions and intellectual debts in scientific publications.
• Previous analyses of acknowledgements were limited by manual examination and scope.
• Large-scale, automated analysis of acknowledgements is needed to accurately assess research contributions.

Purpose of the Study:

• To develop and evaluate an automated system for classifying scientific acknowledgements.
• To identify the best-performing supervised learning algorithms for acknowledgement classification.
• To investigate factors influencing classification performance, including algorithms, categories, and text representations.

Main Methods:

• Created a large-scale training dataset by sampling acknowledgements from PubMed Central.
• Employed and compared various supervised learning algorithms for classification tasks.
• Evaluated the impact of different classification algorithms, acknowledgement categories, and text representations on performance.

Main Results:

• The CNN+Doc2Vec algorithm demonstrated superior performance, achieving 93.58% accuracy on the original dataset and 87.93% on a converted dataset.
• Classification performance was significantly influenced by the characteristics of acknowledgement categories and sentence patterns.
• Classifiers performed best on categories such as financial support, peer interactive communication, and technical assistance.

Conclusions:

• Automated classification of scientific acknowledgements is feasible and effective on a large scale.
• The CNN+Doc2Vec model offers a robust solution for analyzing acknowledgement data.
• Understanding category characteristics and sentence patterns is key to optimizing acknowledgement classification systems.