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

A Study of the Consistency of Judicial Decisions Based on Text Mining: Using Corpuses from Rulings on Applications for Suspension of Detention

陳政瑜(Jheng-Yu Chen)

指導教授 ： 曹承礎

國立臺灣大學/管理學院/資訊管理學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.02717

摘要

近年來，隨著我國的言論自由風氣，漸從以往封閉保守，轉趨開放多元，加上網際網路的普及與社群媒體的快速發展，使得輿論對於法官的裁判，往往存有批評。

隨著資訊技術的發展，開始有「人工智慧與法律」領域的出現，亦即透過資訊技術解決法律領域的各類問題，特別是更客觀的司法裁判。

一般而言，要做到上述司法裁判方面的人工智慧，需收集以往法官既已做出的裁判書相關資料，透過文字探勘（Text Mining）過程加以進行整理、分析、訓練等，據以建立決策模型，使得往後若有新的個案出現，即得透過此決策模型得到決策結果。然而，倘若過往裁判書在撰寫上之一致性不夠，將間接影響決策模型的決策品質。從而原本期望透過資訊技術，來協助法官更有效率的進行裁判工作，同時增加裁判公平性的理想目標，仍難有實現之日。

本研究之研究目的，即在於運用文字探勘技術，以我國刑事訴訟上停止羈押聲請之裁定書為研究分析對象，探討影響研究對象文本撰寫一致性之關鍵因素，以及這些關鍵因素與裁判結果的關聯。

研究結果顯示，研究對象文本的撰寫一致性受到8個中性因子影響，且此8個中性因子對於裁判結果亦有影響。從而可認為就研究對象而言，尚需於實務上改善撰寫一致性上的品質，方有可能進一步利用資訊技術增加我國法院裁判的客觀性。

關鍵字

人工智慧與法律 ； 文字探勘 ； 歧視意識資料探勘 ； 停止羈押聲請 ； 司法裁判品質

並列摘要

In recent years, the atmosphere about freedom of speech is much opener in Taiwan. Moreover, the development of Internet and social media pushes many comments about judicial decisions.

With the development of information technology, there is a field called “AI and Law” so that we can solve various problems about law, for example, more objective judicial decisions.

In general, we need to collect the history texts and to arrange, to analyze and to train them by the process of text mining before we apply AI field to judicial decisions. The whole process lets us get a decision model so that we can after put new cases to it, and get some useful results directly. However, if the consistency about these history texts on writing is not enough, the quality of the decision model will be affected indirectly, and the hope helping judges to make judicial decisions more efficiently and fairly will not come true.

In this study, we analyzed corpuses from rulings on applications for suspension of detention by using text mining, and tried to find factors of the consistency on writing, and the association between the factors we found and the decisions made by judges.

In the result, we found 8 neutral factors about the consistency on writing, and found the association between the 8 factors and the decisions made by judges. Therefore, we realized that we need to improve the consistency on writing in practice before we can use information technology to enhance the objectivity of judicial decisions.

並列關鍵字

AI and Law ； Text Mining ； Discrimination-aware Data Mining ； Applications for Suspension of Detention ； Quality of Judicial Judgement

參考文獻

2.林琬真、郭宗廷、張桐嘉、顏厥安、陳昭如、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 Vol. 17, No. 4, December 2012, pp. 49-68。

3.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，國立臺灣大學科際整合法律學研究所碩士論文，2013年。

6.Keh-Jiann Chen & Ming-Hong Bai (1998), Unknown Word Detection for Chinese by a Corpus-based Learning Method, Computational Linguistics and Chinese Language Processing vol. 3, no. 1, February 1998, pp. 27-44.

1.何君豪(2007)，階層式分群法在民事裁判要旨分群上之應用，國立政治大學資訊科學學系碩士論文，2007年。

4.Kevin D. Ashley, Stefanie Bruninghaus (2009), Automatically classifying case texts and predicting outcomes, Artificial Intelligence and Law 2009, Volume 17, Issue 2, pp. 125-165.