Microblog Analyzer Aggregate Estimation Over A Micro Blog Platform

Abstract

Microblogging is a new mode of communication in which users can share their current status in brief and agile way in the form of text, image, video etc over smart phones, email or web. Recently, Microblogs such as Twitter, Tumblr, Google+ have experience phenomenal growth and are regularly used by millions of users. The data from microblogs is very useful for researchers to analyze various facets such as user behaviors, user intentions (like daily chatter, conversations, sharing information and reporting news), microblog social network structure etc. For example, a sociologist might want to use the microblog postings to analyze the popular opinion about a particular topic. However, existing approach to facilitate such analytics has certain limitations due to the various restrictions imposed by microblogs. Restristions include API rate limits(that restricts the amount of queries issued in a day) or other limits (Twitter search API only provides results for last few days and so on). In this thesis, we build an efficient microblog analytics platform - MICROBLOG-ANALYZER to enable the approximate estimation of aggregate queries over an online microblogging service. MICROBLOG-ANALYZER works by leveraging user timeline access offered by online microblogs. It dynamically constructs a level by level sub-graph of the microblog and performs sampling by a novel topology aware random walk. MICROBLOG-ANALYZER can handle a number of online microblogs such as Twitter, Google+, Weibo, Tumblr, Instagram etc.