Abstractive Text Summarization using Deep Learning

Abstract

In current era, organizing the data has been a very tedious process as there is a lot of data growing every day and storing and managing them is getting tougher, to solve this problem one of the ways is to summarizing the data which in turn will save lot of storage and makes managing data easier. In this paper we have looked into various techniques and methods to summarise the text data using deep learning techniques and come to the conclusion that using Encoder-Decoder Architecture which consists of LSTMs, which is efficient in producing optimum results. This is because LSTMs are good at remembering the long dependencies by overcoming the problem of remembering the context of the previous line in its memory to predict the next sequence. We will first train the model after setting up the model to predict the target sequence. In this training process, the encoder takes an input and processes it and stores the context of that input whereas the Decoder predicts the next word with respect to the previous input. Later a new test input is uploaded to the model to test for which the target is unknown. Attention mechanism is used in this model as it gives attention to the important word in the sentence to understand the context of that sentence and to generate new words from it.