Text Summarization Using Lexical Chaining and Concept Generalization

Abstract

Abstract Text summarization is the process of generating a condensed text form from single or multiple text documents. Extractive summaries are generated by stringing together selected sentences taken from the input text document without any modification, whereas abstractive summaries convey the salient ideas in the input, may reuse phrases or clauses from the input text, and may add new phrases/sentences which are not part of the input text. Extractive summarization guarantees that the generated sentences will be grammatically correct, whereas abstractive methods can generate summaries that are coherent, at the expense of grammatical errors. To this end, we propose to integrate an extractive method based on lexical chains and an abstractive method that uses concept generalization and fusion. The former method tries to identify the most important concepts in the input document with the help of lexical chains and then extract the sentences that contain those concepts. The latter method identifies generalizable concepts in the input and fuses them to generate a shorter version of the input. We evaluated our method using ROUGE and the results show that the integrated approach was successful in generating summaries that are more close to human generated versions.KeywordsLexical chainsAbstractive and extractive summariesGeneralizable conceptsDependancy parsingGeneralizable versions