Networks with hierarchical structure: Applications to the patent domain

Abstract

In this paper we introduced a graph-based metric to measure a similarity between weighted sets of classifications codes defined as nodes on hierarchical taxonomy trees. We applied this metric to build relationship networks among companies and to find company peers (communities) in IPR (intellectual-property rights) domain based on patent portfolios. To characterize evolution of patent portfolios for companies we used weighted sets of international patent classification codes (IPC), where each IPC weight corresponds to a number of IPC codes in a company patent portfolio aggregated to a given hierarchy level over a given period of time. We used the suggested graph-based similarity at different hierarchical IPC levels to build corresponding networks and detected communities over different time periods. To track communities evolution in time we developed a cluster-matching algorithm to align community labels over time. Then we study evolution of communities in time to identify changes in a company strategy and its peers at the given time. The suggested methodology may be applied to other domains that include hierarchical classification sets such as trademarks, legal documents, scientific papers, lawsuits etc.