On Ego Network Modeling And Analysis

The advancing of natural science greatly broaden our knowledge about the environment we are living in, while the social science grows much slower. Big data era provides unprecedent opportunities for the interdisciplinary research, thus give birth to the computational social science. Ego Network (EN) is crucial in anthropology. It is composed of a centered ego, direct contacts namely alters, and the iteractions among them.The widespread of mobile communication makes it cheaper and easier for people to communicate with each other. In fact, communication data is valuble and is becoming crucial in studying human behaviors. In this thesis,all the studies are conducted on the communication data, which is provided by one of the largest operators in China. This data covers a city and lasts from Jan. 2014 to Jun. 2014.With the development of Information and Communication Technologies (ICT), people are able to maintain social relationships with plentful ways. Can this trend always increase people's ego network size?Does the layered ego network structure still exist? At the same time, how to interactively explore and visualize large volumes of data is also the key challenge for visual analytics.The answers to the above questions are the 3 main sub-topics of this thesis and they are as follows:Ego network modeling and its critical size. Most related studies have explored the upper limits of the EN sizes based on Facebook and Twitter datasets. Such data only records the "forward" and "comment"behaviors, but not the communication behaviors. Furthermore, most of the studies only pay attention to the specific kind of interations, but none of them systematically introduce an ego network model, let alone the information embedded witin the communication directions. This thesis model the ego network as directed weighted graph, and investigate the correlations between the ego network properties and its size. Considering the social implicastions of communication directions, the attractiveness balance and tie balance are introduced to describe the balance between an ego and the whole network as well as the balance between strong ties and weak ties within an ego network. The results show that, ego network size has significant impacts on ego network properties, and the critical size of ego network is close to 150 (±20%), the Dunbar's Number. Dunbar's Number states that the upper limit of social relationships one person can maintain is 150. Once the ego network size goes beyond the critical size,the average communication strength will fall dramatically and the ego network balance collapses (indicated by the sudden drops of attractiveness balance and tie balance). This result can serve as a cross-culture supportive evidence for the SBH.Layerd structure of ego networks. Most related works have discovered the layered structure of ENs based on Online Social Network data, however their clustering algorithms require specifying the number of layers manually. Differernt from the above researches, we introduce two parameter-fr-ee algorithms to explore the ENs: parameter-free Jenks natural breaks algorithm, which is provided by this thesis, and the Head/Tail breaks algorithm. This study discovered the layerd structure of ego networks from both communication frequency and duration perspectives. Results show that, about 74%-98.92% stable ego networks are organized as 3-5 layered structures, among thet, 30%?66% users have 4-layer ego networks. The algorithms confirm the support clique and discover a new inner layer within it, which is composed of 1-2 people. Moreover, the ratio of number of people within the concecutive two layers is around 3 (±30%). However,the layerd structures discovered in this study are 1.73-4.68-15.45-66.58 and 1.09-2.38-7.56-21.52-78.84, which do not support the classic 5-15-50-150 layered structure.Visual analytics system based on ego network perspective. In order to interacticely explore and visualize huge volumns of communication data,the researchers brought about the multi-scale interface, which is consistent with the idea of this paper, however this design did not support visualizing ENs. Moreover, the EN visualization systems did not show the detailed interactions between an ego and the alters. To interactively explore the communication data from ego network perspective, this thesis brings about a visual analytics system named egoPortray. The main contributions of this system are the interactive 3-scale views, namely, the macroscopic statistical view, which applies scatter layout and gives the information of data distribution and correlation; the mesoscopic group view, which uses matrix layout of glyph-based designs and enables comparing different egos;and the microscopic signature view, which utilizes the stacked layout and chord diagram design for visualizing the detailed interations between ego and alters. The anomalous user detection task is taken as the case study,and the results illustrate the servicebility of this system.In summary, this research starts with proposing the directed weighted graph model of ego network, then it studies the critical size and the layered structure of ego networks progressively. At last, it fuses the research perspective of ego network into visual analytics technologies, and puts forward a visual analytics system named egoPortray. The case study demonstrates its usefuness and potentials of applying such system into the scenes where perosonalized services and data driven operations are in need.Ego network perspective pays more attention on ego itself rather than the whole network, this is consistent with the trend of personalized services,and it can be applied in other scenes and applications where network structure exists.