Scientific Articles Summarization using Deep Learning Techniques
Scientific Articles Summarization using Deep Learning Techniques
This was my course project for the subject "Text Analytics" and I focused on automatically summarizing scientific research papers to help researchers quickly understand large volumes of literature. Using deep learning based encoder–decoder models with attention mechanisms, the system was able to generate concise summaries while preserving the key ideas of the original text. I have used two architectures
LSTM-Attention
GRU-Attention
I observed that the LSTM-based model achieved stronger performance.
Pytorch Framework
In this work the SCI-TLDR dataset has been used for implementing scietific article summarization. It is an open sourced dataset provided in a CSV format. The dataset consists of 5400 TLDRS on the basis of 3200 research articles. Here only the dataset involving abstract and its summarizations are used. A total of 1992 train data samples, 619 test data sample and 620 validation data have been used. The input to the model are the abstract of the scientific research paper and the output is the summary of the abstract.
View Code: Code