Modeling Patent Legal Value by Back Propagation Neural Network

以倒傳遞神經網路建構專利法律價值的評價模型

本研究基於專利的法律本質，提出專利法律價值的評價模型。專利侵權訴訟的「判罰金額」被定義為專利可被量化的法律價值。經由調查美國德拉瓦州、加州與德州等三州聯邦地方法院的資料，得到4,289個專利相關訴訟案例，再藉由縝密的篩選得到65個具有判罰金額的有效樣本。本研究採用了17個量化的專利指標，經由因素分析將其縮減為9個萃取因子。由多元迴歸分析篩選出對判罰金額影響較為顯著的6個萃取因子。再倒傳遞神經網路建構專利評價模型，其中6個萃取因子作為輸入變數，判罰金額則為輸出變數。藉由誤差分析與驗證，此專利評價模型具有預測能力，可應用於評估專利訴訟所可能產生的判罰金額，並協助原告與被告兩造雙方的談判。

This study aimed at the basis of patent law and proposed a valuation model for the monetary legal value of patents. The damage award of a patent infringement lawsuit was deemed to be the legal value of a patent. 65 Effective samples of infringement lawsuits were extracted from 4,289 patent related lawsuits in U.S. district courts of Delaware, California and Texas. 17 quantitative patent indicators were summarized to describe dimensions of a patent. 9 extracted factors were generated from the 17 patent indicators by Factor analysis. The multi-regression was applied to find the nonlinearity between the damage award and 9 extracted factors. 6 extracted factors which having higher influence to the damage award were thereby suggested via the multi-regression. The Back Propagation Neural Network was applied to construct the valuation model, wherein the 6 extracted factors were the inputs and the damage award was the output. The proposed valuation model was validated to have the predictive power by error analysis. It accommodated to valuate the possible damage award or to negotiate the settlement fee in disputing patent infringement lawsuits. It also contributed to patent transaction, patent licensing and hypothecation of intangible assets, etc.