[3E3-GS-2] Machine learning: market analysis

[3E3-GS-2-04] On the number of samples needed for estimating opinions in social networks

Masato Shinoda¹, 〇Yuko Sakurai², Satoshi Oyama³ (1. Nara Women's University, 2. AIST, Kyushu University, 3. Hokkaido University)

Keywords:Opinion Estimation, Social Networks, PAC Learning

We use the PAC learning framework to evaluate the number of samples needed to estimate the overall proportion of opinions propagated in a social network. While existing studies have only considered binary opinions, this study uses the graph dimension and Natarajan dimension, which are generalizations of the VC dimension, to give upper and lower bounds on the number of samples when multiple values are considered.