Machine learning for patent intelligence: opportunities and challenges

Abstract

The analysis of large data volumes for decision-making has evolved from a sideline to a key driver of economic success in the business world of today. As being particularly relevant for technology-oriented organization, patent intelligence – the retrieval, pre-processing and analysis of patent information – has become a relevant means for organization-relevant decisions. This dissertation sheds light using techniques from machine learning for patent intelligence tasks. After summarizing the current literature streams of patent intelligence and machine learning, four publications outline opportunities and challenges that may arise from using supervised or unsupervised machine learning techniques for patent intelligence. For example, supervised machine learning may guide decision making by reducing noise in predictions, for example. Unsupervised machine learning may be useful to explore latent associations between patents when analyzing computationally challenging patent datasets. However, both techniques impose challenges regarding the complexity of the configuration space as well as the transparency and explainability of their underlying algorithms. Implications of this dissertation offer two trade-offs, i.e. in-house versus external procurement and high performance vs. low explanability, and relevant gaps need being addressed by this dissertation open up avenues of further research.