The Power of Pre-trained Transformers for Extractive Text Summarization: An Innovative Approach

Abstract

In this study, we suggest a unique method for text summarization that combines the TextRank algorithm, Kmeans clustering, and neural network classification. To determine which phrases in a given text are most crucial, the basic model uses TextRank, a graph-based algorithm. In order to group together comparable sentences, these sentences are subsequently clustered using K-means. The best representative statement from each cluster is chosen as the final summary in the last phase of our method using neural network classification. In order to enhance TextRank’s functionality, we also suggest an optimization strategy called cosine similarity with TextRank (Cosim-TextRank). In order to further improve the model’s accuracy, we also suggest using weighted cosine similarity. Overall, our method successfully creates a summary of the text by choosing significant and illustrative phrases while maintaining the context and content of the original text. The experimental findings demonstrate that, in terms of ROUGE scores and human evaluation, our suggested strategy performs better than the current state-of-the-art methods.