本文主旨在討論臺灣的商標侵權實務判決中，法官常見考量之審酌因素對於法官最終判決賠償金額的影響。在法院判決中，法官首先要先認定侵權人具有侵害商標權之事實，須負賠償商標權人之義務，才進行侵害商標權賠償金額之審酌，我們把這稱作樣本自我選擇問題 (Sample selection bias)。國內不乏探討計算侵害商標權賠償金額之文獻，其中大多數以個案進行商標法第71條第1項第3款之合理性探討，以及整理歸納法官在判決實務上應審酌賠償金額過高或過低之因素。然而，過往研究並未以計量方式分析常見法官審酌因素對於判決結果之影響，無法確定歸納出之審酌因素對於判決賠償金額在實證上確實具有影響力。
本文整理出常見之法官審酌因素，利用 Python 程式語言搜索政府公開資料庫之判決書，以規律性文字從判決書中提取法官審酌內容並加以量化後，使用赫克曼兩階段估計法 (Heckman two-stage model)，以常見法官審酌因素對法定賠償金額進行估計。赫克曼兩階段估計法可以良好的處理樣本自我選擇問題，至今已被眾多國內外學者拿來進行實證上之應用，包括且不限於法學實證。此外，本文亦對判決書內容進行詞性斷詞，利用計算辭彙出現於案件數之頻率，找出非一般學界認定之法官審酌因素，並探討新的審酌因素對判決結果之影響。以及，針對屬於智慧財產法院之案件進行迴歸模型估計，以評估專業法院之法官審酌條件對於判決結果的影響。
實證結果發現，估計模型確實有樣本自我選擇問題，利用「善意使用」作為合適之排除變數 (Exclusive variable)，僅對法官判定被告具有侵權事實具有顯著影響力，對於判賠金額並無影響。而在計算斷詞結果後，可以將出現頻率高之辭彙歸類成不同之商品類型，本研究也對判決書提到此新的審酌因素對於判決結果之影響進行討論。並且，針對屬於智財院的案件進行估計後，發現專業法院在審酌條件上的評估更具有一致性，相較於整體法院更為專業化。

The main purpose of this paper is to explore the influence of the judges' common considerations on the assessment of damages in the infringement cases of the Trademark Act of Taiwan. In the litigation procedures, it is necessary to first determine that the infringer has the infringement of the trademark right, and must bears the obligation of the trademark owner, then, the judge will consider the amount of compensation for damages of trademark infringement, which we call it “sample selection bias”. There is no shortage of literature on the assessment of the amount to be compensated for trademark infringement in Taiwan, most of which only discuss on the rationality of Article 71, Paragraph 1, subparagraph 3 of the Trademark Law by Case, and sort out the discretionary factors that judges should consider whether the damages for trademark infringement cases is over identify. However, the previous studies have not measured the impact of the judges' common considerations to the judgment with Metrological analysis, we could not be sure that the discretionary factors in the judgment have an empirical influence on the amount of the statutory compensation.
This paper sorts out the common discretionary factors of trademark infringement cases, and collects the related court verdicts from the platform of Government Open Data with Python. After extracting and quantifying the ratio decidendi without obiter dicta from the court verdicts with regularity text, we apply the Heckman two-stage model to estimate the influence of the common discretionary factors on the amount of the compensation for damages of trademark infringement. The model solves sample selection bias well and has been used by many domestic and foreign scholars for the empirical applications with literatures that including and not limited to empirical legal studies. In addition, this paper also uses word segmentation and part of speech analysis on the court verdicts to calculates the frequency of occurrences of the vocabulary, which is called word type frequency, for identifying the discretionary factors of trademark infringement which were not recognized by general and exploring the influence of the new discretionary factors on the judgment. Furthermore, we distinguish the court to which the case belongs, and estimate the influence of the common discretionary factors on the compensation amount by cases belongs to Intellectual Property Court.
The empirical results show that the sample selection bias exists. In Heckman two-stage model, we use "using another person’s registered trademark with bona fide" to be the exclusive variable, which only has significant influence on the judge determination whether the defendant had the infringement fact, but has no effect on the amount of compensation for damages of trademark infringement. In addition, after calculating the word type frequency, the vocabulary with high frequency of occurrence can be classified into different commodities types, which will be the new discretionary factors of trademark infringement and the new explanatory variable in the model. This paper also discusses the influence of the new discretionary factors on the judgment. Moreover, after estimating the cases that belong to the Intellectual Property Institute, we find that the evaluation of the discretionary factors of the professional court is more consistent and more professional than the overall court.

中文部分
王怡蘋 (2014)，「商標侵害之損害賠償計算-以民國 100 年修法為核心」，《輔仁法學》，48，1-56。
王昱庭 (2011)，「侵害商標權損害賠償計算標準之研究─ 以我國實務判決為中心」，碩士論文，私立東海大學法律學系研究所。
王璽欽 (2007)，「應用資料探勘技術於專利分析之研究」，碩士論文，私立銘傳大學資訊管理學系研究所。
何君豪 (2007)，「階層式分群法在民事裁判要旨分群上之應用」，碩士論文，國立政治大學資訊科學系研究所。
李駿翔 (2003)，「應用資料探勘分類技術於專利分析之研究」，碩士論文，私立中原大學資訊管理研究所。
林琬真、郭宗廷、張桐嘉、顏厥安、陳昭如和林守德 (2012)，「利用機器學習於中文法律文件之標記、案件分類及量刑預測」，《Computational Linguistics and Chinese Language Processing》，17(4)，49-68。
林筱瓴 (2013)，「文字探勘在判決書上之應用-以著作權法民事賠償為中心」，碩士論文，國立臺灣大學科際整合法律學研究所。
邵子案 (2012)，「基於文字探勘的專利分析與預測」，碩士論文，國立清華大學通訊工程研究所。
袁曉東、蔡學輝和許豔霞 (2016)，「我國專利侵權賠償制度實施效果及法定賠償數額影響因素研究」，《情報雜誌》，36(5)，192-198。
高愷均 (2015)，「臺灣智慧財產法院損害賠償判決實證研究」，碩士論文，臺灣大學科際整合法律學系研究所。
張晉芬和杜素豪 (1985)，「薪資性別差異的趨勢」，《台灣的社會變遷1985~2005：社會階層與勞動市場》，217-250。
陳姵君和劉湘宜 (2013)，「新法有關侵害商標權損害賠償計算標準之初探──以「法定賠償數額」 計算標準之審酌因素為中心」，《智慧財產權月刊》，170，23-43。
陳政瑜 (2018)，「基於文字探勘技術探討司法裁判書之撰寫一致性：以刑事訴訟停止羈押聲請裁定書為例」，碩士論文，國立臺灣大學資訊管理學研究所。
陳貞諭 (2010)，「專利關鍵字之分析與探勘」，碩士論文，私立逢甲大學工業工程與系統管理學研究所。
陳恭平、林常青和張永健 (2016)，「法實證研究方法進階導論：設限資料」，《台灣法學雜誌》，303，81-92。
陳麗珣 (2016)，「商標侵權之金錢損害賠償實證研究─以金額酌定問題為中心」，元照出版有限公司。
曾元顯 (2004)，「專利文字之知識探勘：技術與挑戰」，《現代資訊組織與檢索研討會》，111-123。
黃國昌、林常青和陳恭平 (2010)，「勞資爭議協調程序之實證研究--以「政府協調」與「民間協調」之比較為中心」，《中研院法學期刊》，7，209-267。
黃詩淳、邵軒磊 (2018)，「酌定子女親權之重要因素：以決策樹方法分析相關裁判 」，《國立臺灣大學法學論叢》，47(1)，299-344。
經濟部智慧財產局 (2012)，「侵害商標權損害賠償計算標準之研究」，《經濟部智慧財產局成果報告書》。
鍾佳樺 (2012)，「專利權人類型與專利移轉之實證分析：以 Ocean Tomo 為例」，碩士論文，國立中央大學產業經濟研究所。
羅郁仁 (2011)，「中文專利指標及文字探勘之研究」，碩士論文，國立臺北科技大學工業工程與管理研究所。
 
英文部分
D. Black and P. Ciccolo (2004), "Machine Learning for Patent Classification," [Online], Available:http://cs229.stanford.edu/proj2005/BlackCiccoloMachineLearningForPatentClassification.pdf. [Accessed Feb. 05, 2019].
Buchanan, B. G., and Headrick, T. G. (1970), "Some Speculation about Artificial Intelligence and Legal Reasoning," Stanford Law Review, 42, 40-62.
Liu, C. L., Chang, T. C., and Ho, J. H. (2004), "Case Instance Generation and Refinement for Case-Based Criminal Summary Judgments in Chinese, Journal of Information Science and Engineering," Journal of Information Science and Engineering, 29(4), 783-800.
Wooldridge, J. M. (2001), Econometric Analysis of Cross Section and Panel Data, Boston, MA: Massachusetts Institute of Technology.

資料來源
政府資料開放平臺：
https://data.gov.tw/
智慧財產法院：
http://ipc.judicial.gov.tw/ipr_internet/index.php