The Promise of Machine Learning for Patent Landscaping

Patent landscaping involves the identification of patents in a specific technology area to understand the business, economic, and policy implications of technological change. Traditionally, patent landscapes were constructed using keyword and classification queries, a labor-intensive process that produced results limited to the scope of the query. In this paper, we discuss the advantages and disadvantages of using machine learning to produce patent landscapes. Machine learning leverages traditional queries to construct the data necessary to train the machine learning models, and the models allow the resultant landscapes to extend more broadly into areas of technology not expected a priori. The models, however, are “black boxes” that limit transparency into their underlying reasoning. To illustrate these points, we summarize two landscapes we recently conducted, one in mineral mining and another in artificial intelligence