Single Document Summarization Using Natural Language Processing

The need for text summarization is crucial as we enter the era of information overload. However, the current implementations are specific to a domain or a genre of the source document. In this paper, we discuss an algorithm for text summarization which is independent of domain and document source. This algorithm creates text summaries by analyzing the logical structure of the sentences. Sentences are parsed and important relationships are identified, stored in the form of a graph, thus graph corresponding to each sentence in the document is generated and merged to form graph of the document, now this graph is clustered into sub-graphs which represent the different topics in the document. Then a graph scoring algorithm scores the graph, and helps in extracting the important sentences towards summary. To increase the coherence of the summary, the pool of extracted sentences undergoes some transformation in a specified order, resulting in final sentences that form the summary of the document.