Abstractive Text Summarization using Recurrent Neural Network with Attention Based Mechanism

Files
Full Thesis.pdf (1.66 MB)
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Department of Computer Science & Information Technology
Abstract
To facilitate the task of reading and searching information, it became necessary to find a way to reduce the size of documents without affecting the content of original document. The solution to this problem is abstractive text summarization system. Abstractive means that they are not restricted to simply selecting and rearranging passages from the original text. It is an automated system which takes an input text to produce another shorter version of given source documents without losing relevant data and meaning conveyed by the original text maintaining its semantic and grammatical correctness. This research aims to propose the RNN Encoder Decoder model with LSTM using attention mechanism for abstractive summarization to generate the summary carrying meaningful representation of the source document. ROUGE metric has been calculated to evaluate the summaries generated by the model run in 64 batch size with 256 hidden unit forming one layer.ROUGE1 and ROUGE2 score has been calculated among different metrics. The average ROUGE1 score obtained for validation data is 17.60 and ROUGE2 is 1.65.
Description
Keywords
RNN (Recurrent neural network) , LSTM (Long short term memory) , ROUGE (Recall-Oriented Understudy for Gisting Evaluation)
Citation
URI
https://hdl.handle.net/20.500.14540/9960
Collections
Computer Science & Information Technology