Neuro fuzzy inference model based on adaptability in egocentric emailing system environment

Abstract

Social networks are in the midst of a boom in terms of connecting and linking people. People are more active in this medium as compared to other mediums of communication, such as telephonic conversations or personal meetings. Conferring an estimate, Facebook holds 900,000,000 active users every month, while Twitter has 310,000,000 monthly unique visitors; meanwhile, about 90% of internet users are involved in the activity of sending and receiving emails via email communication. Hence, on social networks, people are more active in linking, updating their profiles and sharing their interests because these sets of actions take a few minutes and most are free of charge. This research has focused one of the domains of social network that is the egocentric email communication. An egocentric email communication is an online repository where people keep the very rich profile of their own. Profiles are a contextual outline of a person that include who you are, where you are, what your personal and professional interests are, and with whom you are associated or connected. Such contextual information helps to investigate more about a person. This research aims to prove that “A change in human context is detectable by analyzing in real-time his/her egocentric email communication”. It identifies the features of an egocentric email communication that interpret the contextual outline of an individual. Also, it tried to visualize the change with different level of intensity in a clear and precise way. In order to verify the objectives two egocentric email networks are considered as the primary datasets. The main modules of the proposed research are longitudinal analysis and visualization of networks, extraction of suitable features to detect a possible change in the network and finally classification to identify the intensity of change. In the first phase, email data is considered and is transformed to get contextual outlines of ego. The second phase consists of extraction of novel features for analysis of network, i.e. frequency of communication between alters, active and inactive alters in each month’s communication, upturn and reduction in monthly communication, addition and removal of alters in the network and above all to gauge the degree of prominence of each alter using 9 network measures. In the final stage, a hierarchal classification scheme has been applied using neuro-fuzzy classifier named Adaptive Neuro-Fuzzy Inference systems (ANFIS). The classification stage first detects network changes as significant or non-significant and then at the second level it further detects the intensity of change as no change, mild, moderate or severe change. A feature matrix is generated by using all the selected features and a rate of change is measured to conclude the level of change. The evaluation of the proposed system has been carried out using two real email datasets. ENRON data set which is publicly available and contains emails data of different members of the organization for a period of 24 months; the second dataset-Is composed on a private organization over a period of 58 months. The results showed that the proposed system detected instances with 90% accuracy where major changes happened in an egocentric network as mentioned in the timeline of both sets. The proposed model can be used in real time to detect and grade changes in an egocentric network.