Certain investigations on machine learning techniques for text summarization

Abstract

Creating text summaries from large volume of unstructured text like customer reviews web log posts and social media posts is an important task in text mining applications These summaries reveal the useful information which portrays the entire document or reviews Summarization task could be performed using two major approaches Extractive and abstractive approach Extractive approach identifies the significant portion of the document that exposes the entire content and extracts it to form summary Abstractive approach creates summary based on keywords and semantics from the document which makes it difficult when compared to the other approach The extractive summarization task is complex due to redundancy large volume of text variability and semantics of natural language in the text The wide applicability as well as challenging nature of the task has inspired active research in the domain by both academic and industry experts This research focuses on the design of machine learning based systems for two core areas related to text mining namely Feature based text summarization and text similarity detection Existing feature ranking and summarization systems employ a variety of methods including latent semantic indexing Naïve Bayes and other semantics based approaches Due to the complexity of the task there is a need for developing efficient systems In this thesis a feature ranking system based on customer preferences have been developed Three different machine learning approaches have been adopted for feature based micro level extractive text summary formation In the feature ranking system features are extracted using standard dependency parsing algorithm newline