Research And Application Of Automatic Classification Of Product Patents

Product patent is one of the most important sci-tech information resources, in which the technique information has high utilization value for product improvement and innovation. As the key technology in organizing and processing the huge amount of patent data, text classification can sort them and avoid the disorder of patent information. The traditional approach for management of patents is the International Patent Classification (IPC), which classifies patents according to the application fields of inventions. This classification system is too general and inflexible to find the specific required information quickly and precisely. While artificial classification has the disadvantages in low efficiency, high cost, and poor consistency. Therefore, according to the project sponsored by the National Natural Science Foundation of China (No. 50675197) and the practical requirement of the enterprise, the author makes further research on automatic classification of product patents in this thesis, and develops a product patents classification system, which can do the classification according to any technique requirements. This system is successfully used in automatic classification of patents related to pneumatic tools, to assist the product innovation design supported by patent knowledge.This thesis consists of six chapters. In chapter 1, the author elucidates the background and significance of this subject, and systematically introduces the development process of text classification and the present research status of patent classification and pneumatic tool products, and then confirms the main research contents in this thesis. In chapter 2, the structure of patent document is analyzed, then the definition and evaluation indices of text classification are given, and the general realization method of automatic patent classification is also proposed. In chapter 3, patent representative components selection and text preprocessing are researched, and the corresponding patent representation model is proposed, which provides conditions for the classification. In chapter 4, the author makes comparison research on feature selection, and studies the principles of classification algorithms, including K nearest neighbor, class center vector and naive bayes. The classification performances of these algorithms are compared, and the strategies for improving K nearest neighbor and naive bayes are proposed. Finally the main influential factors of classification performance are researched by orthogonal test to find the optimal factors combination. Chapter 5 introduces the development and realization of the patent classification system in details, which includes the general framework of the system, database design, program structure and key functions design. This system is successfully applied in the classification of patents related to pneumatic tools, and the instance using patent knowledge to innovate some parts of the products is given too. The last chapter is the conclusion, which summarizes all the work in this thesis and gives the outlook for the future research.