Research On Application Of Data Mining In Analyzing And Forecasting The Patent Information

With the development of computer technique, especially the rapid development and widespread use of the database technique, the amount of data in every walk of life has been growing. Because it is difficult to take full advantage of the useful knowledge stored in these data by the traditional approaches, Data Mining has come into being. Data Mining is defined as a process that gets information and knowledge which are connotative, unknown and useful from practical data which is substantial, incomplete, noise, ambiguous and stochastic.Patent information set is the world's largest set of technical information which practically includes technological achievements about all application areas. With the growing commercial competition, enterprises urgently want to know the patent information of their business competitors and use these information maximum widely. Through using the method of Data Mining to deal with patent information, the life cycle and the stage of technical development have been found out, the geographical distribution and the competitor distribution of patent data have been gotten too, so it can help people with reducing duplicate research and invalid work and making investment more active and rational.This thesis focuses on the feasibility of describing the patent information by the method of Data Mining. Firstly, the basic techniques for analyzing and forecasting the patent information have been introduced in this thesis. Secondly, taking chip packaging patent for example, the basic patent data has been analyzed in SQL 2000 database and auto regression model, generalized regression neural network model and grey model have been taken to deal with the data of chip packaging patent in Matlab in order to get the period of the data and forecast the number of the data next time. Using text categorization technique, the frequency of the keyword item has been computed and general rules of the development of technique have been found out. Using k-nearest neighbors model, patent documents have been classified and the similar patent documents have been extracted for manual interpretation. Finally, the program of patent search and analysis system for custom request has been designed, and the techniques which achieve it have been given out.