Konstantin Mertsalov

Finding Overlapping Communities in Social Networks

Increasingly, methods to identify community structure in networks have been proposed which allow groups to overlap. These methods have taken a variety of forms, resulting in a lack of consensus as to what characteristics overlapping communities should have. Furthermore, overlapping community detection algorithms have been justified using intuitive arguments, rather than quantitative observations. This lack of consensus and empirical justification has limited the adoption of methods which identify overlapping communities. In this text, we distil from previous literature a minimal set of axioms which overlapping communities should satisfy. Additionally, we modify a previously published algorithm, Iterative Scan, to ensure that these properties are met. By analyzing the community structure of a large blog network, we present both structural and attribute based verification that overlapping communities naturally and frequently occur.

Defining and Discovering Communities in Social Networks

The categorization of vertices in a network is a common task across a multitude of domains. Specifically, identifying structural divisions into internally well connected sets have been shown to be useful in computer science, social science, and biology. In each of these areas, grouping vertices using structural boundaries helps one to understand the underlying processes of a network. Identifying such groupings is a non-trivial task and has been a subject of intense research in recent years.

Discovery, analysis and monitoring of hidden social networks and their evolution

Social networks that arise spontaneously and evolve over time have become an important component of ever growing global societies used for spreading ideas and indoctrinating people. Their loose membership and dynamics make them difficult to observe and monitor. We present a set of tools for discovery, analysis and monitoring evolution of hidden social groups on the internet and in cyberspace. Two complementary kinds of tools form a core of our approach. One is based on statistical analysis of communication network without considering communication content. The other focuses on communication content and analyzes recursive patterns arising in it. First, we present a software system SIGHTS (Statistical Identification of Groups Hidden in Time and Space), designed for the discovery, analysis, and knowledge visualization of social coalition in communication networks by analyzing communication patterns. We discuss how our algorithms extract groups and track their evolution in Enron-email dataset and in Blog data. The goal of SIGHTS is to assist an analyst in identifying relevant information. A complementary set of tools uses Recursive Data Mining (RDM) to identify frequent patterns in communication content such as email, blog or chat-room sessions. Our approach enables discovery of patterns at varying degrees of abstraction, in a hierarchical fashion, and in language independent way. We use RDM to distinguish among different roles played by communicators in social networks (e.g., distinguishing between leaders and members). Experiments on the Enron dataset, which categorize members into organizational roles demonstrate that use of the RDM dominant patterns improves role detection.

Overlapping communities in social networks

Traditionally, methods to identify community structure in networks have focused on partitioning the vertex set into a number of disjoint groups. However, recently proposed methods have included mechanisms to account for possible overlap between communities. These approaches have taken a wide variety of forms, resulting in a lack of consensus as to what characteristics overlapping communities should have. Additionally, the application of algorithms which account for community overlap are often justified via intuitive rather than empirical arguments. In this text, each of the issues mentioned above is examined. From previous literature, a minimal set of axioms which overlapping communities should satisfy is presented. Additionally, a modification of a previously published algorithm, iterative scan, is introduced to ensure that these properties are met. Finally, the overlap between communities discovered in a large, real world communication network is examined. The analysis offers empirical justification tha...

Models of Communication Dynamics for Simulation of Information Diffusion

We study information diffusion in real-life and synthetic dynamic networks, using well known threshold and cascade models of diffusion. Our test-bed is the communication network of the LiveJournal Blogosphere. We observe that the dynamic and static versions of the Blogograph, yield very different behaviors of the diffusion. It was earlier discovered that the communication dynamics of the Blogograph is quite high - over 60% of the links each week were not present in the previous week, though the size of the node set is relatively stable. Our models of the Blogograph evolution reproduce general stable statistics of the real-life Blogograph. We discover that the diffusion footprint on our models closely approximate the diffusion footprint of the real-life dynamic network.

A permutation approach to validation

We give a permutation approach to validation (estimation of out‐sample error). One typical use of validation is model selection. We establish the legitimacy of the proposed permutation complexity by proving a uniform bound on the out‐sample error, similar to a Vapnik‐Chervonenkis (VC)‐style bound. We extensively demonstrate this approach experimentally on synthetic data, standard data sets from the UCI‐repository, and a novel diffusion data set. The out‐of‐sample error estimates are comparable to cross validation (CV); yet, the method is more efficient and robust, being less susceptible to overfitting during model selection. Copyright

graphOnt: An ontology based library for conversion from semantic graphs to JUNG

In this work, we present the software library graphOnt. The purpose of this library is to automate the process of dynamically extracting “interesting” graphs from semantic networks. Instructions on the extraction are fed into the library via an ontological language specification custom built for this application. A set of SPARQL queries are used to define vertices and edges in the constructed graph. Extracted graphs are returned using the JUNG framework, which offers many algorithmic and visualization options. This work allows a set of individuals analyzing the same semantic network to extract and analyze dynamically created graphs using sophisticated, specific algorithmic tools without needing to manually construct classical graphs from the data.

Stable Statistics of the Blogograph

The primary focus of this paper is to describe stable statistics of the blogospheres evolution which convey information on the social networks dynamics. In this paper, we present a number of non-trivial statistics that are surprisingly stable and thus can be used as benchmarks to diagnose phase-transitions in the network. We believe that stable statistics can be used to identify anomalous behavior at all levels: that of a node, of a local community, or of the entire network itself. Any substantial change in those stable statistics must alert the researchers and analysts to the need for further investigation. Furthermore, the usage of these or similar statistics that are based solely on the communication dynamics and not on the communication content, allows one to diagnose anomalous behavior with minimal intrusion of privacy.

SIGHTS: A Software System for Finding Coalitions and Leaders in a Social Network

We present an extended version of a software system SIGHTS (statistical identification of groups hidden in time and space), which can be used for the discovery, analysis, and knowledge visualization of social coalitions in communication networks such as Blog-networks. The evolution of social groups reflects information flow and social dynamics in social networks. Our system discovers such groups by analyzing communication patterns. The goal of SIGHTS is to be an assistant to an analyst in identifying relevant information. The functionality of SIGHTS includes: discovery of coalitions (clusters) and their leaders; finding hidden groups using communication persistence techniques; discovering hidden groups in communication streams; matching topics of the blogs and detecting sentiments; tracking the evolution of clusters; visualization of collections of individual clusters.

A locality model of the evolution of blog networks

In this paper, we present a novel model for the evolution dynamics of social networks which supports public communication (communication which is visible to all members of the network). Though our model is general, it is particularly applicable to blog-networks, which is the domain we use for testing. We use a directed graph, the blogograph, to represent the communication activity in the Blogosphere. Our model is based on three fundamental principles for describing evolution dynamics in social networks.