A new method for rare feature extraction in patent documents

Patent documents, as large corpora, potentially contains great knowledge. In order to gain more valid information from the massive unstructured data efficiently, a new research area, called patent mining, emerges in recent years, aiming to analyze the patent documents. The popular methods were generally based on the term frequency, which results that a proximate predefined threshold acts an important role in mining process to obtain the valid information. However, such a mining process leaves out rare features lower than threshold. In this paper, we proposed a new method (weighted max confidence) to detect rare features in massive patent documents. The experimental results show that the method can optimize the rare features ming results with a higher precision.