Patent Thickets Identification

Patent thickets have been identified by various citations-based techniques, such as Graevenitz et al (2011) and Clarkson (2005). An alternative direct measurement is based on expert opinion. We use natural language processing techniques to measure pairwise semantic similarity of patents identified as thicket members by experts to create a semantic network. We compare the semantic similarity scores for patents in different expert-identified thickets: those within the same thicket, those in different thickets, and those not in thickets. We show that patents within the same thicket are significantly more semantically similar than other pairs of patents. We then present a statistical model to assess the probability of a newly added patent belonging to a thicket based on semantic networks as well as other measures from the existing thicket literature (the triples of Graevenitz and Clarkson’s density ratio). We conclude that combining information from semantic distance with other sources can be helpful to isolate the patents that are likely to be members of thickets.