Automated reference validation for scholarly publications using NLP

Abstract

Accurate references in scholarly publications are a crucial aspect of scientific writing. The manual validation of references can be a time-consuming and error-prone process. This research introduces an updated version of the automated referencing validation model that makes the peer review process efficient. The proposed model utilizes the capabilities of Natural Language Processing generating sentence embeddings which uses an efficient algorithm. Our model first breaks down the scholarly article into sections and uses topic modeling to group every section according to their context properly. After that, It generates sentence embeddings for each section. By making sets of embeddings, they are used to calculate the semantic similarity between the query and the referred article. Additionally, this methodology addresses the valid references for non-contextual scenarios such as having common name entities. Lastly, strategic feature engineering is also being used for better performance. We have created a dataset of scholarly papers with manually verified references to evaluate the efficiency and accuracy of our model. This improved version of the referencing validation model aims to outperform traditional models such as Document-BERT, BERT, and SBERT regarding efficiency and accuracy. The model can be used in interactive real-time systems, providing quick and reliable feedback to peer reviewers. This study aims to make a contribution to the field of automated referencing validation in scholarly publications. The model offers a solution to the limitations of manual validation which makes it a valuable tool for peer reviewers and researchers.