The problem Abstractio solves

It will generate an abstract or a simplified summary of the paper for easier understanding of the paper. The summary's structure would look similar to the original paper like Abstract, Methodolody, Result, Conclusion, etc.

A Summary can be of 2 types - abstractive summary (simplify summary in your own words) or extractive summary (use the most important lines in the text to restructure the input text in the form of a summary).

Since the abstract and the conclusion don't contain much technical details, we will use abstractive summarization using BART model. Both the abstract and conclusion will be fed separately to the BART model.

The methodology part of the paper generally contains some important details that needs to be retained. A researcher needs to know the technical details of the paper as well. Hence extractive summarization should be used (we will use sentence-BERT model) to extract the important steps in the methodology or the important results mentioned in the paper.

Challenges we ran into

The inaccuracy in the pdf to text conversion using OCR can further mess up the summarization model to generate meaningful text out of it.
Implementation of BART and sentence-BERT models needs to be done such that it can summarize texts effectively.
The BART model for abstractive summarization can have a tendency to oversimplify the sentences in abstract and conclusion and thus could make the summary meaningless w.r.t. the paper.
The sentence-BERT model for extractive summarization can sometimes lose out on the essential details like steps in the methodology or important results due to its inherent nature. of discarding information.