| Summary: | 碩士 === 國立臺北科技大學 === 工業工程與管理研究所 === 96 === The father of innovative problem-solving methods (TRIZ), Genrich Altshuller and his team engaged in patent documents and developed innovative TRIZ. The 40 TRIZ innovative principles had been considered as most popular and fastest to implement in TRIZ applications. Some scholars announced that some innovative principles were overlapped and ambiguous. They tried to cluster innovative principles into certain groups for more feasible in automatic patent classification. It has limited effectiveness for TRIZ users in a specific industry.
The aim of this research is to classify 1,000 Chinese patents of ROC LCD-related industry into TRIZ innovative principles by using text mining techniques. The 1,000 patents have been mapping into alternative TRIZ innovative principles grouping approaches (40 innovative principles by G. Altshuller, 22 categories of innovative principles by H. Cong, and 13 categories of innovative principles by K. Rantanen), respectively. First of all, using on-line auto-tag system provided by Academia Sinica to break every sentence in a document into several keywords and label these keywords manually. Calculating text frequency (TF) and inverse document frequency (IDF) in the corresponding documents. Secondly, chi-square statistics and correlation coefficient approaches are used to select and sort word features highly correlated to TRIZ innovative principles. Then, TFIDF and weight-TFIDF values for selected keywords are calculated and further incorporated with classifiers such as Support Vector Machines (SVM) and K-Nearest neighbor classifier (KNN). Finally, SVM and KNN evaluate the performances of 1,000 Chinese R.O.C patents of LCD-related industry with respect to 3 TRIZ innovative principles grouping approaches as mentioned. As far as Chinese patents concerned, experimental results show that 40 TRIZ innovative principles merging into certain groups may not be enhance the efficiency of patent classification. Clustering TRIZ innovative principles into certain groups incur poor result and increase the complexity of classification. Both of SVM and KNN perform well and the accuracy between SVM and KNN was less than 2%. However, the best result of classification only had 65.5%. Even though, the result of Chinese patent classification into TRIZ innovative principles could encourage LCD industry to generate new and feasible ideas.
|
|---|
