A New Method for Predicting the Importance of Scientific Articles on Topics of Interest Using Natural Language Processing and Recurrent Neural Networks

The realization of the state of the art presents significant challenges for researchers, as it involves addressing the extensive and dynamic amount of existing literature in a specific area. The explosion of information and the constant evolution of knowledge make identifying and synthesizing the most relevant contributions a complex task. Variability in source quality, diversity of methodological approaches, and lack of standardization in results presentation also hinder information systematization. Additionally, the speed at which new research emerges adds an additional layer of challenge to keeping the state of the art up to date. In this context, researchers must develop critical search skills, efficient information management, and discernment to provide a comprehensive and accurate view of existing research in a specific field. Initially, we employed the term frequency-inverse document frequency (TF-IDF) method along with total citation influence (CIT) calculations to determine the main topics and influence of articles within the scientific community. The primary innovation of this research is the development of an RNN model. This model has proven to be equally effective as the traditional TF-IDF method, complemented by CIT, in identifying the relevance and importance of articles.