Automatic Coherent and Concise Text Summarization using Natural Language Processing

Over the past two decades, with the advancements in the World Wide Web and Internet, there has been an exponential increase in the amount of online information causing difficulties in retrieving the precise and necessary information quickly. The solution to this problem is the Automatic Text Summarization, one of the important domains of Natural Language Processing (NLP) which is being extensively focused by the research community. Text Summarization helps to shrink the size of the source document and presents only the key features without compromising the overall context of the input document. Summarization is broadly classified into two types - extractive and abstractive summarization depending on how the original content is structured in the final summary. In the past, the researchers concentrated widely on extractive approaches and now there has been a gradual shift in the research trend towards abstractive methods and fusion of both extractive and abstractive ones. Consequently, in this paper, we propose a novel Combined Extractive Abstractive Text Summarization (CEATS) model which integrates the benefits of both extractive and abstractive approaches to achieve more concise, logical and human readable summaries of the online product reviews collected over a period of time. The extractive stage includes word frequency based sentence feature extraction, graph based sentence ranking algorithm whereas the abstractive phase involves deep artificial neural network approach. It consists of sequence to sequence encoder decoder model made of RNN LSTM networks