Patent Claim Structure Recognition

In patents, the claims section is the most relevant part. It is written in a legal jargon, containing independent and dependent claims, forming a hierarchy. We present our work aimed at automatically identifying that hierarchy within complete patent claim texts. Beginning with a short introduction into patent claims and typical use cases for searching in claims, we proceed to show results from a preliminary context analysis with English claims from the European Patents Fulltext (EPFULL) database. We point out some possibilities with which claim dependency is indicated in the text and show a way of identifying them. Additionally, we describe several of the problems encountered, in particular problems resulting from noisy data. Finally, we show results from our internal evaluations, in which accuracies greater than 93% were measured. We also indicate areas of further research.