Automatic Patent Clustering using SOM and Bibliographic Coupling

Abstract

Patents are usually organized in classes generated by the offices responsible for patents protection, to create a useful format to the information retrieval process. The complexity of patent taxonomies is a challenge for the automation of patent classification. Beside this, the high numbers of subgroups makes the classification in deeper levels more difficult. This work proposes a method to cluster patents using Self Organizing Maps (SOM) networks and bibliographic coupling. To validate the proposed method, an empirical experiment used a patent database from a specific classification system. The obtained results show that patents clusters were successfully identified by SOM through their cited references, and that SOM results were similar to k-Means algorithm results to perform this task. This study can contribute to the development of the knowledge organization systems by evaluating the use of citation analysis in the automatic clustering of patents in a constrained knowledge domain, at the subgroup level of current patent classification systems.