Mitigating Information Overload in Scientific Text Analysis: Automated Extraction of Important Concepts Using Dependency Parsing and Semantic Similarity

Information overload poses a significant challenge in the scientific community, where researchers are often confronted with vast amounts of textual data. This overload can hinder effective information retrieval and decision-making, particularly when analyzing large volumes of scientific texts. To address this issue, a novel approach that combines dependency parsing with advanced semantic techniques to automate the extraction and evaluation of important concepts from scientific abstracts is proposed. By focusing on grammatical relationships within the words in each sentence, our method can identify key information accurately, which is then evaluated using Sentence-BERT to calculate semantic similarity between the extracted concepts and provided keywords. The evaluation is applied to a dataset of Scopus-published papers that are selected based on specific query string. The results demonstrate that while our approach effectively mitigates information overload by enhancing the relevance and accuracy of information retrieval, the average precision and F1-scores indicate room for improvement, particularly in complex domains. This paper contributes to the field of information sciences by offering a practical foundation for managing the cognitive loads associated with analyzing vast amount of scientific literature, with potential for further improvement and application across different research domains.