Abstractive Text Summarization Approaches with Analysis of Evaluation Techniques.

In today’s world where all the information is available at our fingertips, it is becoming more and more difficult to retrieve vital information from large documents without reading the whole text. Large textual documents require a great deal of time and energy to understand and extract the key components from the text. In such a case, summarized versions of these documents provide a great deal of flexibility in understanding the context and important points of the text. In our work, we have attempted to prepare a baseline machine learning model to summarize textual documents, and have worked with various methodologies. The summarization system takes the raw text as an input and produces the predicted summary as an output. We have also used various evaluation metrics for the analysis of the predicted summary. Both extractive and abstractive based text summarization have been described and experimented with. We have also verified this baseline system on three different evaluation metrics i.e. BLEU, ROUGE, and a textual entailment method. We have also done an in-depth discussion of the three evaluation techniques used, and have systematically proved the advantages of using a semantic-based evaluation technique to calculate the overall summarization score of a text document.