Asian Research Thesis Index

Abstract

The World Wide Web and its millions of internet applications have revolutionized the way people to communicate, socialize, stay healthy, educate, conduct business, do politics, and do ?any-thing-at-all?. The huge amount of data generated by these applications is also proving quite valuable in the context of analyses, evaluations and predictions. These predictions help in organizational planning & management, to a very acceptable degree of accuracy. This research thesis is a study of several current researches on the extraction and analysis of short text messages or micro-blogs, from a social-science perspective. Specifically, this study investigates linguistic-frequency and sentiment analysis of micro-blogs in general, by identifying, collating and visualizing conversational and behavioral biases of people, from a very large twitter dataset of 6+ million tweets. This study explores short text analysis using freely available text processing and visual analysis tools. A novel contribution of this work is to investigate the effectiveness of automated labeling of tweets instead of manual labeling. Comparison of automatically labelled tweets with baseline STS-gold set. The baseline dataset of manually labelled twitter dataset show a very small deficiency of 6-8%, making our method viable for huge/big datasets. Main result of this study is a framework that encapsulates three current techniques for analyzing and visualizing microblogs. The frequency measurements and classification have been performed using the NLTK text processing tool. Sentiment Analysis has been carried out using NLTK and WEKA tools. Network Graphs made in Gephi software have been used to visualize user trends and behaviors, utilizing metrics from Networks Theory.

Add/Update Thesis