Research On Automatic Categorization Technology For Chinese Patent Documentation

Patent Classification can accelerate the speed of retrieval of patent documents and facilitate the management of patent literature, which plays quite an important role. At present, since the quantity of China's patent applications improves fast each year, and the method of traditional manual classification has become increasingly difficult to carry out, the achievement of the Chinese patent automatic classification has great significance.In this paper, the background and research status of patent classification are introduced in detail, and the problem of patent classification is defined. Based on detailed analysis of the classification techniques, this article focuses on three main questions of patent classification, feature selection, feature evaluation algorithm and classification method, and the concrete contents are as follows:First, a feature selection method based on IPC domain knowledge is put forward, which imports IPC domain knowledge into feature selection, and set up concept space for each class, and makes use of relativity between character and concept space to select feature. Second, a feature evaluation algorithm method based on theme is put forward, which confirms evaluation by the relativity between character and theme, and makes the text representation of patent approach that of article's theme.Third, a multiple classifier fusion method based on estimated priori probability is put forward, which ensures probability weights by effect of each single classifier,and a linear fusion according to weight is presented to multiple classifier fusion.In this paper, the above-mentioned experimental methods are compared on standard corpus. The experimental results show that the feature selection method can improve classification accuracy, except for the high complexity, and the feature evaluation algorithm method is superior to traditional method which can effectively improve the classification results, and classifier fusion method can effectively improve the classification accuracy.