Kouzou Ohara

Learning Continuous-Time Information Diffusion Model for Social Behavioral Data Analysis

We address the problem of estimating the parameters for a continuous time delay independent cascade (CTIC) model, a more realistic model for information diffusion in complex social network, from the observed information diffusion data. For this purpose we formulate the rigorous likelihood to obtain the observed data and propose an iterative method to obtain the parameters (time-delay and diffusion) by maximizing this likelihood. We apply this method first to the problem of ranking influential nodes using the network structure taken from two real world web datasets and show that the proposed method can predict the high ranked influential nodes much more accurately than the well studied conventional four heuristic methods, and second to the problem of evaluating how different topics propagate in different ways using a real world blog data and show that there are indeed differences in the propagation speed among different topics.

Selecting information diffusion models over social networks for behavioral analysis

We investigate how well different information diffusion models can explain observation data by learning their parameters and discuss which model is better suited to which topic. We use two models (AsIC, AsLT), each of which is an extension of the well known Independent Cascade (IC) and Linear Threshold (LT) models and incorporates asynchronous time delay. The model parameters are learned by maximizing the likelihood of observation, and the model selection is performed by choosing the one with better predictive accuracy. We first show by using four real networks that the proposed learning algorithm correctly learns the model parameters both accurately and stably, and the proposed selection method identifies the correct diffusion model from which the data are generated. We next apply these methods to behavioral analysis of topic propagation using the real blog propagation data, and show that although the relative propagation speed of topics that are derived from the learned parameter values is rather insensitive to the model selected, there is a clear indication as to which topic better follows which model. The correspondence between the topic and the model selected is well interpretable.

Learning diffusion probability based on node attributes in social networks

Information diffusion over a social network is analyzed by modeling the successive interactions of neighboring nodes as probabilistic processes of state changes. We address the problem of estimating parameters (diffusion probability and time-delay parameter) of the probabilistic model as a function of the node attributes from the observed diffusion data by formulating it as the maximum likelihood problem. We show that the parameters are obtained by an iterative updating algorithm which is efficient and is guaranteed to converge. We tested the performance of the learning algorithm on three real world networks assuming the attribute dependency, and confirmed that the dependency can be correctly learned. We further show that the influence degree of each node based on the link-dependent diffusion probabilities is substantially different from that obtained assuming a uniform diffusion probability which is approximated by the average of the true link-dependent diffusion probabilities.

Efficient discovery of influential nodes for SIS models in social networks

We address the problem of discovering the influential nodes in a social network under the susceptible/infected/susceptible model that allows multiple activation of the same node, by defining two influence maximization problems: final-time and integral-time. We solve this problem by constructing a layered graph from the original network with each layer added on top as the time proceeds and applying the bond percolation with two effective control strategies: pruning and burnout. We experimentally demonstrate that the proposed method gives much better solutions than the conventional methods that are based solely on the notion of centrality using two real-world networks. The pruning is most effective when searching for a single influential node, but burnout is more powerful in searching for multiple nodes which together are influential. We further show that the computational complexity is much smaller than the naive probabilistic simulation both by theory and experiment. The influential nodes discovered are substantially different from those identified by the centrality measures. We further note that the solutions of the two optimization problems are also substantially different, indicating the importance of distinguishing these two problem characteristics and using the right objective function that best suits the task in hand.

DryadeParent, An Efficient and Robust Closed Attribute Tree Mining Algorithm

In this paper, we present a new tree mining algorithm, DryadeParent, based on the hooking principle first introduced in DRYADE. In the experiments, we demonstrate that the branching factor and depth of the frequent patterns to find are key factors of complexity for tree mining algorithms, even if often overlooked in previous work. We show that DryadeParent outperforms the current fastest algorithm, CMTreeMiner, by orders of magnitude on data sets where the frequent tree patterns have a high branching factor.

Behavioral analyses of information diffusion models by observed data of social network

We investigate how well different information diffusion models explain observation data by learning their parameters and performing behavioral analyses. We use two models (CTIC, CTLT) that incorporate continuous time delay and are extension of well known Independent Cascade (IC) and Linear Threshold (LT) models. We first focus on parameter learning of CTLT model that is not known so far, and apply it to two kinds of tasks: ranking influential nodes and behavioral analysis of topic propagation, and compare the results with CTIC model together with conventional heuristics that do not consider diffusion phenomena. We show that it is important to use models and the ranking accuracy is highly sensitive to the model used but the propagation speed of topics that are derived from the learned parameter values is rather insensitive to the model used.

Cl-GBI: a novel approach for extracting typical patterns from graph-structured data

Graph-Based Induction (GBI) is a machine learning technique developed for the purpose of extracting typical patterns from graph-structured data by stepwise pair expansion (pair-wise chunking). GBI is very efficient because of its greedy search strategy, however, it suffers from the problem of overlapping subgraphs. As a result, some of typical patterns cannot be discovered by GBI though a beam search has been incorporated in an improved version of GBI called Beam-wise GBI (B-GBI). In this paper, improvement is made on the search capability by using a new search strategy, where frequent pairs are never chunked but used as pseudo nodes in the subsequent steps, thus allowing extraction of overlapping subgraphs. This new algorithm, called Cl-GBI (Chunkingless GBI), was tested against two datasets, the promoter dataset from UCI repository and the hepatitis dataset provided by Chiba University, and shown successful in extracting more typical patterns than B-GBI.

Effect of personalization on retrieval and summarization of sports video

Personalization is one of the most important mechanisms to make multimedia systems easy to use. In video applications, its embodiment is to tailor video contents for a particular viewer. For this purpose, we are now developing a system of retrieving and browsing video segments, called video portal with personalization (VIPP). VIPP is characterized by 1) supporting the viewers access to video contents and making a summarized video clip by taking his/her preference into account, and 2) acquiring the viewers profile from his/her operations automatically. In this paper, we discuss the effect of personalization on retrieval and summarization of sports videos on VIPP.

Learning information diffusion model in a social network for predicting influence of nodes

We address the problem of estimating the parameters, from observed data in a complex social network, for an information diffusion model that takes time-delay into account, based on the popular independent cascade IC model. For this purpose we formulate the likelihood to obtain the observed data which is a set of time-sequence data of infected active nodes, and propose an iterative method to search for the parameters time-delay and diffusion that maximize this likelihood. We first show by using a synthetic network that the proposed method outperforms the similar existing method. Next, we apply this method to problems of both 1 predicting the influence of nodes for the considered information diffusion model and 2 ranking the influential nodes. Using three large social networks, we demonstrate the effectiveness of the proposed method.

Efficient mining of high branching factor attribute trees

In this paper, we present a new tree mining algorithm, DryadeParent, based on the hooking principle first introduced in Dryade (Termier et al, 2004). In the experiments, we demonstrate that the branching factor and depth of the frequent patterns to find are key factor of complexity for tree mining algorithms. We show that DryadeParent outperforms the current fastest algorithm, CMTreeMiner, by orders of magnitude on datasets where the frequent patterns have a high branching factor.