Identifying Sensitive Events using Re-Tweet Graph Analysis

Abstract

Online social networking sites have become an indispensable part of human life. Numerous studies have enumerated the fact that people are spending more time in these websites. Human civilization is shifting from real life to a virtual world where interactions are predominantly using these online social networking websites. This has made people available to each other irrespective of them being long distances apart or being total strangers. Twitter is one of the micro-blogging social networking sites which has exhibited its potential for spreading maximum information content. Twitter is based on small messages called tweets, which are typically lesser than 140 characters long. When someone tweets about an event, these pop up on home pages of others who follow them or those who are following the event. If someone wants to spread the tweet to a wider audience, it can be re-tweeted by him/her. Thus we can form a graph structure where nodes are the person active in twitter and edges are formed when someone re-tweets the other person. These graphs are called re-tweet graph. In this thesis we start with collection of twitter data for different case studies identified to be relevant in Indian and world context using API's exposed by twitter and simulation of retweet graph modeling by Thij. et. al on them. In addition to observations by Thij, it is observed from the data that the average size of other components, apart from largest connected component is always less than three in all case studies. The average clustering coefficient of the graph gradually increases to reach a saturation value. There are a few nodes of moderately high betweenness value in all the case studies, having higher number of neighbors. This work proposes that all re-tweet graphs have a central subset of all nodes, which plays as a crucial factor in opinion making of the general public. Thij's model was extended to include an additional parameter ? which represents the section of people from the community who are part of this central subset of influential nodes. Highly Interacting Kernel and Critical K-Core value are introduced in this work which are determined using the K-Core concept inherent in graph theory. The highly interacting kernel was found to be dis-assortative in each case which implies a shell kind of structure within it. This highly interacting kernel which is isolated in each case study represents the set of influential people in the community who have play a crucial role in public view formation in the society. We propose a new algorithm which calculates the Critical K-Core value that can be used as a new metric to identify sensitive events. Further we show that the Critical K-Core value has a correlation to the sensitivity of the event. The higher is the Critical K-Core value, the more is the sensitivity of the event. The estimates developed in this work help us to identify the sensitive events in initial few tweets (approximately 25000) and thus can act as an alarm to the stakeholders. Also it can be incorporated as an additional parameter to identify the trends in twitter. Security agencies can also use it to identify the potential threats to the nation so that timely measures can be taken.