Adaptive Text_Summarization: A Meaningful Summarization of Reviews

Abstract

Adaptive text_summarization is a model which takes elaborated data set of reviews containing more than two attributes as input and further breaks down those large chunks of text into summarized - form by eliminating attributes which are of less important for final evaluation. We keep only those attributes which contribute for better adaptation of meaning of the dataset. We have executed the techniques of Recurrent Neural-Network and Seq2Seq models to accomplish the aim with better efficiency. The selected attributes are passed through the encoder and decoder. The encoder converts the text into vectorised form of data which further is passed through the unidirectional LSTMs of the decoder providing us with naturally acceptable modified version of the source text. We have also explored both the techniques of extractive and abstractive text_summarization. The extractive text summarization evaluates the source text, analyses the focused words and present them as output without any language formation and disobeying grammatical rules. On the contrary, abstractive text summarization accepts the source dataset and evaluates the meaning out of it producing naturally acceptable summarized form of text. This model can be acquired to overcome the issue of long summary generation in various product reviews published by the users of the e-commerce applications.

Keywords: Recurrent neural network, Seq2Seq, DUC (Document Understanding Conferences), abstractive and extractive text summarization, NLP (Natural Language Processing), bidirectional LSTM (Long-short Term Memory) Encoder, unidirectional LSTM (Long-short Term Memory) decoder.

Cite this Article: Twinkle Kapoor, Tarushi Chaudhary, Umang Gupta, Sparsh Agarwal. Adaptive Text_Summarization: A Meaningful Summarization of Reviews. International Journal of Software Computing and Testing. 2020; 6(1): 1–7p.