Fusion of Word Embedding and Encoder Decoder Model for Text Summarization

Abstract

Extractive text summarization is a core activity in the field of natural language processing, aiming to condense the most important information from a given text while preserving its core meaning. This study investigates a novel methodology that integrates word embeddings using GLOVE and LSTM based encoder decoder model, two widely recognized methodologies, to enhance the effectiveness of extractive summarization. This novel approach incorporates the advantages of both word embeddings and LSTM to enhance the summarization process. Word embeddings can capture semantic links between words and phrases, so facilitating a more profound comprehension of textual context. Encoder decoder-based LSTM model identifies predicted summary based on the original summary. BLEU and cosine similarity are metrics that evaluates the importance of terms in a collection of documents, so guaranteeing that crucial phrases are given suitable weighting. An algorithm is provided for the task of extracting text summarizing, wherein sentence embedding is achieved by following word embedding, and the score of the summary is obtained. The model is tested on a dataset obtained from Kaggle and consists of news summaries. The model that was suggested obtained a BLEU score of 59.4% and a cosine similarity of 50.2 %; when these findings were compared with the state of the art work, it was found that the proposed model produced superior results.