About KSE >






Change Point Detection for Information Diffusion Tree by Prof. Motoda Hiroshi

We would like to invite all of you a seminar by Prof. Motoda Hiroshifrom AFODR/AOARD and Osaka Univ. on December 4. 2015(Fri)


Speaker : Prof. Motoda Hiroshi


Date & Time : 2:30 p.m ~ , Dec. 4. 2015(Fri)


Place :  E2-1, #1222 Seminar Room


Title :  Change Point Detection for Information Diffusion Tree


AbstractWe propose a method of detecting the periods in which bursts of information diffusion took place from an observed diffusion sequence data over a social network, explicitly taking the network structure into account. This is different from most of the change detection approaches in which all the diffusion information is projected on a single time line and the detection is searched in this time axis. In reality information diffusion takes place along a diffusion path. Each such a path has multiple descendants on the way and the path splits into each child node from which another new diffusion path starts. Thus, change in diffusion is both spatial and temporal. We assume a generic information diffusion model in which time delay associated with the diffusion follows the exponential distribution and the bursts are directly reflected to the changes in the time delay parameter of the distribution.  The shape of the parameter's change is approximated by a step function along each path and the problem of detecting the change points and finding the values of the parameter is formulated as an optimization problem of maximizing the likelihood of generating the observed diffusion sequence. Time complexity of the search is almost proportional to the number of observed data points and the method is very efficient.  We first demonstrate that the proposed method can detect the bursts using a real Twitter data, and show that a method that does not consider the network structure gives very different results with many more incorrect change points using Kleinberg's bust detection method as a representative approach of this kind. We then investigate the performance of the proposed method more extensively using synthetic data for which we artificially embed the change points (thus the ground truth is known) and compare the results with Kleinberg's method. The experimental results reconfirm the satisfactory performance of the proposed method.


Short BioHiroshi Motoda is a professor emeritus of Osaka University and currently works, as a scientific advisor, for AFOSR/AOARD (Asian Office of Aerospace Research and Development, Air Force Office of Scientific Research, US Air Force Research Laboratory). As a scientific advisor, he manages a small number of AOARD granted projects in Asian countries.  His original research background is nuclear engineering, but for the last 25 years he has been working in the area of artificial intelligence, scientific knowledge discovery, knowledge acquisition, machine learning, data mining and information diffusion.  He received his Bs, Ms and PhD degrees all in nuclear engineering from the University of Tokyo. He is a member of the the steering committee of PAKDD, PRICAI, DS, WI and ACML. He received the best paper awards from Atomic Energy Society of Japan (1977, 1984) and from Japanese Society of Artificial Intelligence (1989, 1992, 2001), the outstanding achievement awards from JSAI (2000), the distinguished contribution award for PAKDD (2006) and PRICAI (2014), Okawa Publication Prize from Okawa Foundation (2007) and outstanding contribution award from Web Intelligence Consortium (2008). He is a fellow of JSAI.







File Download:
  • 페이스북 공유
  • 트위터 공유
  • Google+ 공유
  • 인쇄하기