Extractive text summarization of online scientific articles for digital library repository

Abstract

In a scientific community including researchers and students who are dedicated in reading, experimenting and writing ideas to the research world, they must refer to a significant number of articles on a daily basis. Digital libraries play a key role in supplementing scientific articles within online platforms. Due to the much abundance quantities of such articles presented in various platforms, the searching process tends toward time consuming, and identifying much related resources among them becomes again difficult. On the other hand, the majority of the scientific articles are available on a subscription-based and the online archives show only a document abstract but users necessitate extra material to determine if the article is extremely relevant or not, even if it merely provides a quick description. Therefore, this work aims to introduce an alternative approach to simplify the searching and sorting of the scientific articles for a digital library where a short subunit of sentences of the subscribed articles will be provided before purchasing it. To find a best algorithm for the process of summarisation of scientific articles within a short time and provide a good comprehension of the scientific document are to be investigated. Summaries from the publicly available SumPubMed dataset of scientific articles are evaluated using supervised and unsupervised approaches and manual summaries from them are compared. Text Rank algorithm performed better than the TF-IDF and K-Means algorithms, and the system achieved a better result when increasing the content size of the article.