A1398
Title: Analyzing patent citations using a stochastic generative model
Authors: Yuichiro Yasui - Nikkei Inc. and The Institute of Statistical Mathematics (Japan) [presenting]
Junji Nakano - Chuo University (Japan)
Abstract: Patent citation networks are investigated and a stochastic generative model for their formation process is developed. Existing models based on preferential attachment and triad formation (TF) usually treat the TF selection probability as a constant, but analysis shows that this assumption is inadequate for patent citations, especially in reproducing triangle participation. A model is therefore proposed where the TF selection probability $\beta$ takes one of three values: $0$, $p\; (0<p<1)$, or $1$. For a citing node with out-degree $k$, $\beta$ is set to $1$ when $k \geq k_U$; otherwise, $\beta$ is set to $0$ with probability $\pi_0$ and to $p$ with probability $1-\pi_0$. This formulation considers the observation that TF-generated edges for nodes with out-degree $k$ and selection probability $p$ can be approximated by a binomial distribution with $k-1$ trials and success probability $p$, while empirical data also show inflation at zero and one. Simulation experiments show that the proposed model improves the fit to patent citation networks, remains applicable to academic citation networks, and provides parameters that help distinguish their structural characteristics.
