以資料探勘技術為基礎之創新發明文件自動分類系統

An Automatic Classify System of TRIZ Document Based on Data-mining Technology

簡丞皓(Cheng-Hao Jian)

指導教授 ： 蔡明標

國立虎尾科技大學/工程學院/自動化工程研究所/碩士(2011年)

摘要

TRIZ創新發明理論與案例式推理系統於產品創新設計理論之應用已有非常重要之成果，而其中影響案例式推理之成功與否的關鍵又是TRIZ-BASED知識庫的完善程度。本文提出專為TRIZ用戶所需之創新發明之文件自動分類系統，以建立完善之TRIZ文件知識庫。此分類方法主要分成三個步驟，首先整理出40個發明原則在專利文件中各種出現的統計形式與次數。第二步驟根據矛盾表的39個改善與惡化之參數，依照參數在文件中出現的頻率高低(可能不只一組參數)判斷文件與哪一組參數最為相關。第三步驟將前兩步驟所分類好之文件庫與矛盾表進行比對驗證，以確認分類的結果是否正確。
在系統實作方面，本文以SQL Server 2008為主要發展工具，收集200餘篇與TRIZ相關之專利文件以及期刊為訓練樣本。本文僅由TRIZ的40個原則中挑出6個原則，依據這6個原則分別使用群集演算法，關聯規則以及決策樹演算訓練模型，分類此200篇文件，第二步則是依據此200篇文件內使用到的矛盾與改善參數再做進一步的細分，然後進行與矛盾表之確認工作，並以80篇創新發明文件作為測試樣本，驗證模型之正確性。最後將分類好的文件庫，提供為輔助產品設計之案例推理系統之知識庫用途。

關鍵字

創新發明 ； 文件分類 ； 專利分類 ； 資料採礦

並列摘要

The Theory of Inventive Problem Solving (TRIZ) and Case-Based Reasoning (CBR) system are often applied by many companies in recent years. And the completeness of TRIZ-based knowledge system is a key focus to influence the result of CBR system which is adopted to aid the product development. In this paper, an automatic classify system of TRIZ-document based on data-mining technology is proposed and built to increase TRIZ-based knowledge data. There are three steps in the development of automatic classify system of TRIZ-document. First, the relative glossary and vocabulary of 40 inventive principles are made arrangements according to the frequency and numbers of their emergence in the document pools. Secondly, the relative vocabulary of 39 improving and worsening parameters are searched and made judgment depending on the frequency and numbers of their emergence in the document pools. Lastly, the prior classified results of step one and step two are verified and made judgment with the contradiction table of TRIZ.
In paper, there are three kinds of algorithms to be adopted to build the model of automatic classify system including decision tree, associated rule and clustering. And Microsoft SQL Sever 2008 is used to build automatic classify system of TRIZ-document. In order to simplify the practical problem, there are 6 inventive principles selected including segmentation, asymmetry, preliminary action, curvature, parameter changes and composite material. The 200 TRIZ-documents about 6 inventive principles are collected for constructing and training the model of automatic classify system and 80 documents of 6 inventive principles for testing the model. Finally, the result of decision tree is the best model for analyzing the correctness with the contradiction matrix.

並列關鍵字

Theory of Inventive Problem Solving (TRIZ) ； Case-Based Reasoning (CBR) ； automatic classify system

參考文獻

[1] Robles, G. C., Innovation management and knowledge management in process and industrial systems engineering: TRIZ and case-based reasoning,Ph.D., INPT-ENSIACET , 2005.

[2] 周聖原，“模組化產品設計-以自行車產品設計為例”，國立虎尾科技大學機械與機電工程研究所，碩士論文，2009。

[7] Theory of inventive problem Solving (TRIZ). Available from: http://www.mazur.net/triz/.

[8] Makarov M. “The seventh edition of the IPC”. World Patent Inform 2000;22:53–8.

[15] Liang C, Naoyuki T, Hisahiro A. “A patent document retrieval system addressing both semantic and syntactic properties”. In:ACL Workshop on Patent Corpus Processing, 2003.