Berlin 2012 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 6: Focus Session: Big Data (Contributed Talks)
SOE 6.5: Talk
Monday, March 26, 2012, 18:15–18:30, HE 101
Analysis and modeling of human behavior observed in cyber space communication data — •kenta yamada1, yukie sano2, hideki takayasu3, and misako takayasu4 — 1Waseda University, Japan — 2Nihon University, Japan — 3Sony CSL, Japan — 4Tokyo Institute of Technology, Japan
Analysis and modeling of human behavior become major targets of twenty-one century science. Especially, huge data of articles in cyber space such as blogs and twitters are attracting attention because the data directly reflect trends and topics in the society.
In this presentation, we report our analysis and modeling of the huge blog database which contains 300,000 bloggers including 70 million articles from 11/01/2006 to 8/31/2011. We observed some characteristic patterns in appearance frequency of some key words per day. We categorize them into three patterns: ordinary words, news words and trendy words. Ordinary words like adverbs are characterized by stationary processes, while the frequency of the news words and trendy words are characterized by non-stationary processes. A news word such as "tsunami" shows a sharp increase by sudden appearance of news and the number decays slowly following a power law. In the case of a trendy word, the number of entries per day increases exponentially.
In order to understand the origin of these characteristic motions, we introduce a state transition type agent-based model similar to the SIR (Susceptible-Infected-Recovered) model which is a basic epidemic model. We show that our simple agent-based model reasonably reproduces these three typical patterns.