Some studies on abstractive multi-document summarization

Abstract

The main objective is to propose a framework for abstractive summarization of multi-documents, which aims to select contents of summary from the semantic representation of the documents but not from the source document sentences itself. Therefore, the goal is to extract the gist of the content from the multi-documents in easily readable, grammatically correct format. In this framework, contents of the source documents are represented by predicate argument structures after implementation of semantic role labelling. Then source text is analysed semantically based on the predicate argument structure. Post this step, the cluster of semantic similar predicate argument structures across the text are formed. The proposed framework differs from other abstractive summarization models. First, it employs semantic role labelling for semantic representation of text. Secondly, it analyses the source text semantically by utilizing semantic similarity measure to cluster semantically similar predicate argument structures across the text; and finally, it ranks the predicate argument structures based on features weighted by genetic algorithm (GA) and generates the abstract summary of the folder.