Feature Based Opinion Summarization

Abstract

Web technologies in association with social media have seen immense growth in the last newlinefew years. This growth has, in turn changed the human behavior over the net. The users newlinewho were just the information consumers earlier have now become the data generators. newlineThis change in the role of the web users has managed to create a massive volume of newlineopinionated text. The Opinionated content so generated is of little use unless we have a newlinemechanism to interpret and analyze this data automatically. Also, due to the remarkable newlinebenefit that the social network and media analysis have on the business world, politics, newlinefinance market, etc., the field of Sentiment Analysis is growing at an exponential rate. newlineThe main task of Sentiment Analysis is to extract the sentiments or opinions expressed in newlinetexts. Many researchers have focused on techniques to extract meaningful opinions from newlinethe review data sets. Though these opinions extracted are useful, their real potential can newlineonly be understood when these opinions are summarized. A subtask of Sentiment Analysis newlinecalled Opinion Summarization is the research area which focuses on techniques to generate newlinedifferent types of Opinion Summaries. The thesis addresses this problem of generating newlinerelevant Opinion Summary by proposing various techniques that are domain- independent. newlineThe research is carried out by dividing the entire task of Opinion Summarization into four newlinedifferent phases of Preprocessing, Feature Extraction, Feature Opinion Pair generation, and newlinelastly the Summary Generation. newlineThe input text for the Summarization task is from various social networking sites such as newlineBlogs, Twitter and Product review sites. As the nature of text in these sources is found to newlinebe highly informal and often unstructured, a Preprocessing model with various techniques newlineto transform and filter unwanted tokens was proposed. In the next phase of Feature Extraction, a framework model and an Automated Rule-based algorithm were proposed to extract relevant features.