ノイズあり属性統計量からの学習 (JSAI2016) Learning from Attribute Statistics with Noise (JSAI2016)

吉川友也. ノイズあり属性統計量からの学習. 第30回人工知能学会全国大会 (JSAI2016). 2016.
Yuya Yoshikawa. Learning from Attribute Statistics with Noise. The 30th Annual Conference of The Japanese Society of Artificial Intelligence (JSAI2016). 2016.

Estimating the attributes of SNS users is an important technique for marketing and advertisement delivery systems.
The attributes are typically estimated in a supervised learning manner.
However, labeling the users with their attributes manually is difficult and expensive.
In this paper, we address the problem of learning a classifier that predicts users’ attributes only from the attribute statistics of websites that the users have visited.
To tackle this problem, we propose a probabilistic generative model for the attribute statistics, in which can capture the users’ attributes as hidden variables.
Moreover, since the attribute statistics of the SNS and the websites may be different, the proposed model is modeled so as to treat the difference as noise variables.
In the experiment, we show that the proposed model outperforms the existing methods on synthetic datasets.