Automatic Generation of Abstracts for Research Papers

Automatic Generation of Abstracts for Research Papers Dushan Kumarasinghe Department of Computer Science and Engineering University of Moratuwa dushan.21@cse.mrt.ac.lk Nisansa de Silva Department of Computer Science and Engineering University of Moratuwa nisansadds@cse.mrt.ac.lk Abstract Summarizing has always been an important utility for reading long documents. Research papers are unique in this regard, as they have a compulsory summary in the form of the ab- stract in the beginning of the document which gives the gist of the entire study often within a set upper limit for the word count. Writing the abstract to be sufficiently succinct while being descriptive enough is a hard task even for native English speakers. This study is the first step in generating abstracts for research papers in the computational linguistics domain auto- matically using the domain-specific abstractive summarization power of the GPT-Neo model. Keywords: NLP, Summarization, GPT-Neo 1 Introduction The abstract of a research paper provides a quick summery of the entire paper: from the problem to the proposed solution to the result. Thus by definition, this section is expected to be concise and informative (de Silva et al., 2017). Text sum- marization is one of the main domains in Natural Language Processing (NLP) which has numerous use cases. There are two broad categories for this: extraction and abstraction. In extractive methods it uses existing words, phrases or sentences to form a summary. In contrast, abstractive methods follow a more complex mechanisms. First, a semanatic representation of the content is built. Then natural language generation mechanisms are used to create the summary using the aforementioned representa- tion. This research proposes a hybrid mechanism of text summarization to generate the abstract sci- entific papers with evaluating several paths for the proposed solution. The objective of this research is to reduce the burden on researchers by automatically generating the abstract section by using the sections of the paper that follows it. The researchers then may do minor adjustments to the generated section and publish. Considering existing summarization techniques, abstractive solutions have domain specific limi- tations. On the other hand, domain specific im- plementations perform better in the perspective of precise representation of the subject matter. Ab- stractive solutions gain domain specificity from the process of models being built upon and information extracted from the training documents. Despite the loss of generalization, this improves the accuracy of the solution within the selected domain. Thus, we propose to build and test our solution for re- search paper abstract generation with the scope limited to the domain of Computational Linguis- tics. As future work, it may then be extended to other research domains. 2 Related Work El-Kassas et al. (2021) emphasize the importance of developing abstractive automatic text summa- rization methods. The paper describes the different approaches, methods, building blocks, techniques, datasets, evaluation methods, and future research directions of summarization methods. Referring Dutta et al. (2019), El-Kassas et al. (2021) claim that different algorithms produce different sum- maries from the same input texts and it is very promising to combine outputs …

Content truncated to 500 words