以主成份分析及類神經網路為基之專利重要性評估方法論

Method of Estimate for the Significance of Patent Base on Principal Components Analysis and Neural Network

林啟瑋(Chi-Wei Lin)

指導教授 ： 張瑞芬

國立臺北科技大學/管理學院/工業工程與管理研究所/碩士(2010年)

https://doi.org/10.6841/NTUT.2010.00081

摘要

企業購買專利已經成為發展技術或是保護自己的手段，但由於全球專利數量越來越龐大，如何在龐大的專利群中，分析出高品質與價值的專利，目前已有頗多專利研究文獻，其大多針對專利技術文件進行分析及議價探討，但評估出專利的價值，需要考量的因素相當複雜，資料的蒐集及所依據的指標必須相當精確，故常在議價的過程中花費大量的時間。本研究目的在縮短專利價值判別的科學性及所需時間，故研究流程為先行蒐集彙整可行專利價值判斷指標因素，將之整理為十二項影響專利品質之指標。透過主成份分析找出關鍵之影響指標，並將這些專利指標，選擇訓練專利樣本投入倒傳遞類神經網路的訓練。訓練前使用者必須先針對各篇專利品質的定義作確認，本研究利用專利是否有買賣之事實作為判斷其品質的依據，將訓練樣本投入訓練後，可分析出指標的訓練權重，即可將未知之專利，透過已知的權重計算出各篇專利的品質分類結果，依此方法將複雜的專利文件作初步品質分類，降低使用者人為判斷的時間及成本，並增加分類法的一致性。

關鍵字

專利品質 ； 專利指標 ； 主成份分析 ； 倒傳遞類神經網路

並列摘要

Technical patent have the way of protection for many business. However the number of patents increases rapidly which may hinder companies global competitiveness. It is important to identify the key patents and basic patents. Now, there are many research reports focusing on patent investigation. In order to estimate the value of a patent, there are many complex factors to be considered. Then, time and effort are often spent process of patent evaluation. The purpose of this research is to enable the shortening of time in patent investigation. The first step of this research is to collect the technology-specific patents which will be the training samples. The second step is to survey literature obtained the relevant patent indicator and integrated the relevant indicators into twelve main factors for patent quality analysis. The third step is to discover the key impact factors using the principal component analysis approach. Afterword the key impact factors are used input parameters for the back-propagation neural (BPN) network model training. In this study, we use the patent transaction (i.e., the IP trading status) judge the patent quality. The patents which are traded with IP usage rights, are considered higher quality patent. Otherwise they are lower quality. After the BPN model is trained by the set training samples, the model, item, can be applied to predict the unknown patents’ quality and forecast their potential for IP trades. The patent significantly high accuracy in patent quality evaluation based on the proposed methodology.

並列關鍵字

Patent Quality ； Patent Indicator ； Principal Component Analysis ； Back-propagation Neural Network

參考文獻

[21] 黃翊軒，本體論為基之智慧型專利文件分類方法論研究，碩士論文，清華大學，工業工程與工程管理學系，新竹，2006。

[4] Cybenko, G., "Approximation by superpositions of a sigmoidal function," Mathematical Control, Signals, and Systems, Vol. 5, No. 4, 1989, pp. 303-314.

[5] Hair, J., Anderson, R., Tathan, R. and Black, W., Multivariate Data Analysis, Macmillan, NY, 1998 , pp.79-82.

[6] Hornik, K., Stinchcombe, M., and White, H., "Multi-layer feedforward networks are universal approximations," Neural Networks, Vol. 2, No. 5, 1989, pp. 336-359.

[7] Kaiser, H.F., "The application of electronic computers to factor analysis," Educational and Psychological Measurement, Vol. 20, pp. 141-151.