Extractive Text Summarization Using Topological Features

Abstract

The amount of text data generated these days is increasing exponentially, and it is becoming a very tedious process to extract meaningful information from the huge amounts of text data. In this work, we propose two methods to summarize the texts using topological features that capture the information over the topological structure, such as connected components and holes in the text data. The first method uses the concept of minimum dominating set to group the sentences into multiple clusters and to find the similarities between clusters using topological features (connected components and tunnels). Sentence scoring and extraction of key sentences from each cluster are done by the existing method of TextRank. The second method uses the pretrained GloVe (global vectors to represent the words) and to find the similarities between sentences using topological features. A classical set cover based algorithm has been used to extract the key sentences for the summary. Both methods are compared on the basis of rouge scores with the existing method, i.e., TextRank, and the results are satisfactory.