下坂 正倫,築地 毅, 坪内 孝太, 西 賢太郎, 前田 啓輔

2015年度人工知能学会全国大会(第29回), 2015/5


Machine Learning Data Science Misc.

"This paper describes a hierarchical Bayesian method to model daily people activity patterns in a city (urban dynamics) with large-scale spatio-temporal data obtained from mobile phones. Thanks to the superiority of hierarchical Bayesian modeling, we can extract appropriate number of activity patterns among cities. In addition to existing models using a Dirichlet process (DP) as a prior distribution, we construct latest models using a hierarchical Dirichlet process (HDP). HDP models can consider the mixture rates of patterns in each area unlike DP models. The results of our experiment with large-scale spatio-temporal data, about 40 million logs per day show that our HDP models have better performance than existing DP models."

携帯電話から得られる大規模な位置履歴情報を用いた都市動態モデリング(External Site Link)