Filtering patent maps for visualization of diversification paths of inventors and organizations

In a patent technology network map, almost all pairs of technology classes are connected, whereas most of the connections are extremely weak. This observation suggests the need and also the possibility to filter the network map by removing the negligible and noisy links. But link removal may reduce the power of the network for predicting the cross-field patent portfolio diversification of inventors and inventing organizations. This paper proposes a metric for such predictive power of a patent network, and a method that allows one to objectively choose a best tradeoff between predictive power and the removal of weak links. We show the results that identify filtered networks below the optimal tradeoff, and also remove a degree of arbitrariness compared with other filtering treatments from the literature. On that basis, we further demonstrate the use of filtered technology maps to visualize and analyze the main paths of patent portfolio diversification of a prolific inventor (Leonard Forbes) and a technology company (Google Inc.).