3.2 Implementation
We have implemented a prototype system exploit- ing the method as illustradted in Figure 2.
This system is composed by four main modules. The first module is the module for the extraction of target documents which consist of “problem to be solved” section or “solutions” section in patent sum- mary from the patent document collection. The sec- ond one is the module to generate the concept base and document vectors. The third module is the clus-tering module which classifies the target document’s vectors into several groups. In the forth module, the similarity calculation module, the similarity degree of word vectors and each vector of the center of gravity for the cluster are computed to generate the candidate of the cluster’s label.