本文旨在建立有效分群，找出死刑與無期徒刑差異。藉由差異建立特徵，並輔助法官判決，進而降低犯人從審判到執行的成本。
目前國在法律文件分類上的研究已經相當成熟，多為用特徵詞與 SVM 的搭配進行案件分類。而中文判決的相關分析卻相當少，原因在於中文的處理上並不像英文一樣好處理，且詞性的標記上有許多困難。由於上述困難，使得法律判決在分析上，需要花費大量人力進行處理，進而使得研究上有難度。
本文利用政府資料開放平台所提供的判決書，找出死刑與無期徒刑判決。並以這些判決作為訓練資料，用以訓練 word2vec 建立語意空間。取得判決向量後，利用 K-medians、SVM 對判決進行分群。然而，在實驗過程中發現，判決中無意義資訊過多，使得準確率無法上升。最終，以「本院經查」中的內容做為代表該判決的主要資訊，並進行否定詞的合併，作為 word2vec 的訓練資料。以倍率詞作為特徵詞，提取出判決中較具代表的特徵，作為該判決的語意，並進行分群。
最終使用倍率詞之後，準確率提升為 94%。而實證結果發現，高倍率詞與低倍率詞中，存在一些差異較大的詞彙，像是「喋血」、「不實」、「泯滅」、「枉顧」、「兒童」、「教化」、「遷善」等詞。將其與死刑及無期徒刑的平均向量計算相關性，發現有的詞明顯屬於死刑，而有的詞介於二者中間。然而，死刑與無期徒刑也如預想的，兩者之間十分相近。但是藉由特徵詞，建立出死刑與無期徒刑的差異，使得兩者可被明確區分。

The purpose of this study was to establish an effective classification to find out the difference between Capital punishment and Life imprisonment. Establishing features by difference and assisting judge’s adjudicate, thereby reducing the cost of trial to execution of criminal.
At present, domestic research on the classification of legal documents was quite complete, it was mostly to classify cases by using the characteristic words and SVM. But, it was seldom used to classify Chinese judgments because the processing method of Chinese was different from English, there were many difficulties in marking its part of speech. Due to the above difficulties, bring about the legal judgment analysis required a lot of manpower to process, which made the research difficult. This study make use of the government data open platform to find out the legal judgement of Capital punishment and Life imprisonment. And use it as training data to train word2vec to establish semantic space. After obtaining the decision vector, utilize K-medians and SVM to classify it. However, during the experiment, it was found that there was too much meaningless information in the judgment, which influence the accuracy rate. Ultimately, the main information on the judgment was based on the contents of the " find and establishment of the Court ", and merged the negative words as training materials for word2vec. Taking rate words as characteristic words, and extract the more representative features of the judgment, and classify it.
Found from empirical results, the death penalty and the life imprisonment, were also as expected, were very similar. However, by using the characteristic words, the differencebetween the death penalty and the life imprisonment was established, so that the two can be clearly distinguished.

中文部份
王兆鵬 (2010)，「台灣死刑實證研究」，《月旦法學雜誌》，No.183，105-130。
陳新民 (2007)，「廢除死刑暨替代方案之研究」，法務部委託研究報告。
邱垂發 (2018)，「重大刑事案件論處死刑之相關法制問題研析」，《國會季刊》，46(3)，110-126。
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」，《中文計算語言學期刊》，17(4)，49-68。

英文部份
Hachey, B., and Grover, C. (2005), “Sequence Modelling for Sentence Classification in a Legal Summarisation System,” ACM Symposium on Applied Computing , 292-296.
Maat de, E., and KRABBEN, K. and WINKELS, R. (2010), “Machine Learning versus Knowledge Based Classification of Legal Texts,” The Twenty-Third Annual Conference on Legal Knowledge and Information Systems.
Quaresma, P., and Gonçalves , T. (2005), “Is linguistic information relevant for the classification of legal texts,” International conference on Artificial intelligence and law, 10,168-176.
Rong, X. (2005), “Word2vec Parameter Learning Explained,” Working paper.
Sulea, O., Zampieri, M., Malmasi, S., Vela,M., , P. Dinu, L., Genabith, J., (2017), “Exploring the Use of Text Classification in the Legal Domain,” Working paper.
政府資料開放平台:
https://data.gov.tw/
裁判書用語辭典資料庫查詢系統－名詞查詢 - 司法院:
http://terms.judicial.gov.tw/