The step similarity comparisons on method patents

The objective of the study is to establish the technologies of automatic similarity analysis and comparison of two method patents in order to reduce the cost and human efforts. We propose a method to compare the similarity between the components and prepositions. The proposed methods automatically extract components, relations between components, steps, and relations between steps, and establish a machine readable tree structures for similarity comparison. Our experiments use the USPTO database [1] as data resource. The study conducts two experiments to evaluate the performance: 1) applying genetic algorithms to learn the weightings in calculation of step similarity and 2) comparing the performance with a statistical similarity calculation method, BLEU. The experimental results show that the system performs better in terms of recall and precision than BLEU methods that are usually adopted in machine translation.