Berlin 2015 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 18: Social Networks
SOE 18.2: Talk
Thursday, March 19, 2015, 11:30–11:45, MA 001
Proposal of segmentation method of time series in the blogosphere — •Kazuya Satou1, Hideki Takayasu2,3, and Misako Takayasu1 — 1Tokyo Institute of Technology 4259,Nagatsuta-cho, Midori-ku, Yokohama-shi, Kanagawa, 226-8502 Japan — 2Sony Computer Science Laboratories, 3-14-13, Higashigotanda, Shinagawa-ku, Tokyo, 141-0022 Japan — 3Meiji University, 4-21-1, Nakano, Nakano-ku, Tokyo, 164-8525, Japan
Time series of the words written in the blogosphere have various types of changes, and it is difficult to extract trends by conventional parametric methods.
In this presentation, we introduce a sequential segmentation method of time series based on Fisher's exact test. In this method, various types of trends are automatically detected by comparing the p-value with that of randomized time series. Since Fisher's exact test is nonparametric, and can be tested with small number of samples, the proposed method can be potentially applied for the general time series.
In addition, we combine the method with the random diffusion model which detects large discontinuous changes in blog time series.
We discuss the results of detection accuracy of the proposed method by using artificial time series and compare with that of conventional methods. Finally, we show the segmentation results applied for real blog time series.