With advances in Internet technology and prominence of mobile and smart devices in our lives, opportunistic and pervasive networks are now ubiquitous in solving many existing service limitations. The challenge lies in the underlying time-varying graph of the network due to mobility and intermittent connectivity. This introduces technical limitations in successful realization of services and applications e.g., efficient routing, maximal coverage with minimal latency, data offloading, and effective dissemination over mobile networks. Efficient solution to these inter-related problems lies in the novel prediction strategies for most accurate future contacts (i.e., links or interactions). In contrast to the existing strategies that consider either network structure or regular pattern and periodic nature of contacts, we propose novel use of seasonal autoregressive integrated moving average model and recurrent neural network model that are capable of capturing multi-periodic, dependent contact patterns. We predict the number of contacts relative to a node and over all nodes in any future interval over a given user and a pair of users. Finally, we validate our models with three distinct empirical data set, and compare with doubly recurrent and homogeneous Poisson process model to demonstrate the superiority of our prediction models.

The real-world large scale networks motivate the need for parallel and distributed evaluation of network analysis and computational tasks for computational efficiency and application effectiveness. One of the essential tasks for parallel and distributed evaluation, is to have partitions over the underlying network graph. Over these partitions the computational or network analysis tasks are in turn processed in a distributed or parallel manner. It is interesting to use intrinsic communities of social networks as partitions, to be used as basic components in parallel and distributed computation. We propose two novel graph compaction algorithms that generate the desired compact graph of communities as a preprocessing stage to the parallel and distributed evaluation of computational tasks. To comply with heterogeneity in community structure and size, we use a flexible limit on them. We evaluate the structure and quality of our algorithms and hence its resulting communities over two distinct application networks. We show that the generated community structure, reasonably complies with the modular structure of the network. We evaluate the quality of the partitions, relative to the partitions generated using existing state-of-the-art approach, and compare the approaches to show better quality of our partitions in terms of number of graph cuts.

One of the most intriguing aspects of network analysis is how links or interactions occur over time between a pair of nodes and whether we can have a model to accurately predict the occurrence of links ahead of time, and with what accuracy. In contrast to the existing approaches, this paper proposes a novel Markov prediction model over the time-varying graph of an underlying social network. The model considers the effect of multiple time scales in leveraging temporal analysis for link prediction. The analysis considers fine-grained and coarse-grained time scales, along with associated local (links) and semi-global (clusters) structural evolution, respectively. The model takes into account correlated evolution and rate of evolution in selecting start and end nodes, and the corresponding interaction probability. Finally, we use temporal data of two heavily dynamic real world social networks (e.g., Twitter and Facebook), and a relatively lesser dynamic network (e.g., DBLP) to demonstrate the prediction accuracy that our Markov model outperforms two recent dynamic approaches in the range of 7.5% to 19.81%.

Real world large scale networks exhibit intrinsic community structure, with dense intra-community connectivity and sparse inter-community connectivity. Leveraging their community structure for parallelization of computational tasks and applications, is a significant step towards computational efficiency and application effectiveness. We propose a weighted depth-first-search graph partitioning algorithm for community formation that preserves the needed community dependency without any cycles. To comply with heterogeneity in community structure and size of the real world networks, we use a flexible limiting value for them. Further, our algorithm is a diversion from the existing modularity based algorithms. We evaluate our algorithm as the quality of the generated partitions, measured in terms of number of graph cuts.

The challenge in predicting future links over large scale networks (social networks) is not only maintaining accuracy, but also coping with the time-varying network graph. In contrast to the existing approaches, in this work we propose building a Markov prediction model. It not only incorporates temporal snapshots reflecting the dynamic network graph, but also considers effect of multiple timescales, along with corresponding local and global structural evolution (links and clusters respectively), correlated evolution and rate of evolution. The resulting edge selection in our approach exhibits the power law degree distribution, as exhibited in real world networks. Finally, we use two heavily dynamic real world network temporal data set (e.g. Twitter and Enron) and one relatively less dynamic network data set (e.g. DBLP), and existing state-of-the-art static and recent dynamic measures, to evaluate the prediction accuracy of our proposed Markov model and show that it out performs existing approaches.

Evaluating influential nodes is one of the fundamental problems in large scale networks having wide range of applications. The centrality metric, in particular betweenness centrality plays a significant role in ranking influential nodes. Existing exact algorithms for evaluating betweenness centrality metric consider the entire network and hence incur high computational cost. In this paper, we reduce computational cost by leveraging network structural properties. We propose a community detection algorithm that uses right-skewed nature of degree distribution with incremental accumulation and semi-local optimal node selection giving computational cost O(|V| 2 - m|V|k 2 ), where k, |V| and m represent average degree, number of vertices and modularity respectively. Additionally, we use predefined upper bound (O(√|V|)) equation on the number and size of communities to propose an algorithm for evaluating exact betweenness centrality indices that exploit the dense intra-modular and sparse intermodular connections of large scale networks, leading to the computational cost of O(|V| 2 + 1/2|V| 3/2 log|V|). We validate our algorithms using real world networks. The computational cost incurred due to community detection and betweenness centrality evaluation holds irrespective of graph density and out performs existing exact algorithms. To the best of our knowledge this is the first work to leverage structural properties in community detection and exact betweenness centrality evaluation over large scale networks.

Estimating influential nodes in large scale networks including but not limited to social networks, biological networks, communication networks, emerging smart grids etc. is a topic of fundamental interest. To understand influences of nodes in a network, a classical metric is centrality within which there are multiple specific instances including degree centrality, closeness centrality, betweenness centrality and more. As of today, existing algorithms to identify nodes with high centrality measures operate upon the entire (or rather global) network, resulting in high computational complexity. In this paper, we design efficient algorithms for determining the betweenness centrality in large scale networks by taking advantage of the modular topology exhibited by most of these large scale networks. Very briefly, modular topologies are those wherein the entire network appears partitioned into distinct modules (or clusters or communities), wherein nodes within the module (that likely share highly similar profiles) have dense connections between them, while connections across modules are relatively sparse. Using a novel adaptation of Dijkstra's shortest path algorithm, and executing it over local modules and over sparse edges between modules, we design algorithms that can correctly compute the betweenness centrality much faster than existing algorithms. To the best of our knowledge, ours is the first work that leverage modular topologies of large scale networks to address the centrality problem, though here we mostly limit our discussions to social networks. We also provide more insights on centrality in general, and also how our algorithms can be used to determine other centrality measures.