| dc.contributor.advisor | 黃詩淳<br>劉昭麟 | zh_TW |
| dc.contributor.advisor | Huang, Sieh-Chuen<br>Liu, Chao-Lin | en_US |
| dc.contributor.author (Authors) | 劉威志 | zh_TW |
| dc.contributor.author (Authors) | Liu, Wei-Zhi | en_US |
| dc.creator (作者) | 劉威志 | zh_TW |
| dc.creator (作者) | Liu, Wei-Zhi | en_US |
| dc.date (日期) | 2023 | en_US |
| dc.date.accessioned | 6-Apr-2023 18:00:45 (UTC+8) | - |
| dc.date.available | 6-Apr-2023 18:00:45 (UTC+8) | - |
| dc.date.issued (上傳時間) | 6-Apr-2023 18:00:45 (UTC+8) | - |
| dc.identifier (Other Identifiers) | G0109753157 | en_US |
| dc.identifier.uri (URI) | http://nccur.lib.nccu.edu.tw/handle/140.119/144044 | - |
| dc.description (描述) | 碩士 | zh_TW |
| dc.description (描述) | 國立政治大學 | zh_TW |
| dc.description (描述) | 資訊科學系 | zh_TW |
| dc.description (描述) | 109753157 | zh_TW |
| dc.description.abstract (摘要) | 有鑒於近年請求扶養費民事訴訟案件有上升之趨勢，考慮伴隨著調解件數增長，未來量能或將難以負荷。而根據法律扶助基金會的年度報告顯示，民事訴訟法律扶助案件中家事案件數量最多的案由為給付扶養費案件，因此本實驗針對其案件進行研究並提出扶養費准駁預測模型與扶養金額預測模型。本實驗以輔助判決為目標，其對象可以是一般民眾、聲請人亦即原告、相對人亦即被告、律師或是法官，對於一般民眾可以透過此模型來了解自己是否能夠獲得其應有的扶養費；對於聲請人與相對人而言則主要是希望兩造都能接受一個共識的扶養金額並多少減輕法官的負擔；對於法官而言則是希望能提供一個客觀數據來參考，並能快速給予公平的裁判。本研究對兩造主張段落進行斷詞、模糊化及向量化，並以機器學習及深度學習為基礎建立二分類模型來進行扶養費的准駁預測，而本實驗會對有無模糊化之兩造主張段落進行向量化與模型預測搭配分別有 TF-IDF 搭配 naïve-Bayes 與 logistic regression 和 SBERT 搭配單純的平均句向量與 BiLSTM 串接並以深度學習方式來訓練與預測，故共計八種方式進行評估 accuracy、precision、recall、F1 score。本實驗也提出對有限且客觀特徵值使用 model tree 來建構迴歸模型進行扶養金額預測，並比較未使用model tree 單純用 linear regression 、使用 model tree 且各分支皆為 linear regression 和使用 model tree 且各分支使用不同的預測模型之三者的 MAE，同時為了節省人工標註特徵值需要大量的人力與時間，讓機器能全自動化進行標註與訓練模型，本實驗有透過 W2NER 來進行特徵值的提取，並進行後續模型的扶養金額預測模型訓練。扶養費准駁預測的實驗中以 logistic regression 的表現最佳其平均的 分數為 0.715，而扶養金額預測的平均 MAE 為 1992.88，然而透過 W2NER 進行自動提取特徵值並搭配後續金額預測模型其平均 MAE 則為 3912.93。透過本實驗模型後望提供未來請求扶養費案件中之兩造乃至法官一客觀參考基準，以期未來能在庭外調解時提供一相對客觀的試算結果供有扶養費爭議之兩造參考並儘早達成共識，亦或給予法官參考數據輔助以期能加速判決之進程，進而減少司法資源的浪費。 | zh_TW |
| dc.description.abstract (摘要) | The needs for mediation are increasing rapidly along with the increasing number of cases of the alimony for the elderly in recent years. According to Legal Aid Foundation’s annual report, cases of the alimony for the elderly has account for the largest number in the foundation’s civil cases. Therefore, this research focus particularly on these cases, offering a prediction mechanism for predicting the outcomes of some prospective lawsuits may alleviate the workload of the mediation courts.This research aims to offer predictions for the judgments and the granted alimony for the plaintiffs of the alimony for the elderly cases in Chinese. For the general public, the predictions can be used to understand whether they can get the granted alimony; for the plaintiff and defendant, it could be a reference for both parties to reach an early consensus; for the judges, it could be an objective data and hopefully speed up the judicial process.To build the current binary classification system for judgments predictions, we segment, blur and vectorize the texts of the judgement documents of the past lawsuits. In the experiment, we vectorize both blur and non-blur documents and train the model using TF-IDF with naïve-Bayes and logistic regression, SBERT with average embedding, SBERT with BiLSTM, total 8 method to train and evaluate the accuracy, precision, recall and F1 score.For the granted alimony predictions, we apply model tree for predicting the judgments and compare with applying only linear regression, model tree with branches using linear regression and model tree with branches using different predicting model for MAE score. Furthermore, we use W2NER to help on feature extraction, saving great amount of manual labeling time.In our experiment, logistic regression has the best performance for judgments predictions and the average score for predicting the judgments is 0.715. For the granted alimony predictions, model tree with branches using different predicting model has the best performance with the average MAE score 1992.88, and the average MAE score via W2NER to perform feature extraction is 3912.93. We hope the results can provide an objective reference for the involved parties to reach an early consensus and provide as supportive data for the courts in order to speed up the process of judgment. | en_US |
| dc.description.tableofcontents | 1 緒論 11.1 研究背景與動機 11.2 研究目的 11.3 主要貢獻 21.4 論文架構 22 文獻回顧 32.1 裁判檢索及推薦系統 32.2 裁判書案件分類及分群 32.3 案件因素分析及結果預測 32.4 小結 43 實驗語料來源與篩選及系統架構 53.1 語料來源 53.1.1 司法院裁判書 53.1.2 扶養費、民事裁判書 63.2 語料篩選 73.3 研究架構 114 扶養費准駁預測模型 134.1 扶養費准駁預測資料前處理 134.1.1 資料擷取與標記 134.1.2 資料模糊化與向量化 164.2 准駁預測模型實驗設計 194.2.1 向量化設定 194.2.2 准駁預測模型與參數設定 194.3准駁預測模型實驗結果比較 275 扶養金額預測模型 295.1 扶養金額預測前處理 295.1.1 篩選案件 295.1.2 客觀特徵擷取 305.2 金額預測模型設計 335.2.1 Model tree 各層設計 345.2.2 Model tree 各分支對應預測方式介紹 355.3 扶養金額預測結果與比較 366 NER 取代人工標註 406.1 W2NER模型 406.1.1 模型介紹 406.1.2 實驗設計與參數設定 426.1.3 訓練結果 436.2 實驗結果 437 結論與未來展望 477.1 結論 477.2 未來展望 48參考文獻 49附錄 51 | zh_TW |
| dc.format.extent | 5109233 bytes | - |
| dc.format.mimetype | application/pdf | - |
| dc.source.uri (資料來源) | http://thesis.lib.nccu.edu.tw/record/#G0109753157 | en_US |
| dc.subject (關鍵詞) | 判決預測 | zh_TW |
| dc.subject (關鍵詞) | 民事案件 | zh_TW |
| dc.subject (關鍵詞) | 給付扶養費 | zh_TW |
| dc.subject (關鍵詞) | legal judgment prediction | en_US |
| dc.subject (關鍵詞) | civil case | en_US |
| dc.subject (關鍵詞) | the issues of alimony | en_US |
| dc.title (題名) | 老年扶養費請求案件之准駁及扶養金額預測 | zh_TW |
| dc.title (題名) | Predicting Judgments and Grants for Civil Cases of Alimony for the Elderly | en_US |
| dc.type (資料類型) | thesis | en_US |
| dc.relation.reference (參考文獻) | 藍家樑，中文訴訟文書檢索系統雛形實作，國立政治大學 資訊科學系 碩士論文，2009。曹錫璋，基於深度學習模型之判決書情境相似檢索技術之研究，國立中興大學 資訊科學與工程學系所 碩士論文, 2021.何君豪，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 資訊科學系 碩士論文, 2007.林琬真，機器學習於中文法律文件之標記與分類，中文計算語言學期刊 第 17 卷 4 期 第49 - 68頁，2012。李右元，透過起訴書輔助法院判決-以竊盜罪為例，國立政治大學 資訊科學系 碩士論文，2021。何君豪，AI 引入民事程序可行性之研究，國立臺灣科技大學 資訊管理系 博士論文，2021。黃詩淳、邵軒磊，以人工智慧讀取親權酌定裁判文本: 自然語言與文字探勘之實踐，臺大法學論叢 NTU Law Journal 第 49 卷 1 期 第195 - 224頁，2020。Zellig S. Harris. Distributional Structure, WORD, Volume: 10, no. 2-3, Pages: 146-162, 1954. Available: https://doi.org/10.1080/00437956.1954.11659520Fabrice Muhlenbach, Long Nguyen Phuoc and Isabelle Sayn. Predicting Court Decisions for Alimony: Avoiding Extra-legal Factors in Decision made by Judges and Not Understandable AI Models. 2020. Available: arXiv:2007.04824黃詩淳，老親扶養費酌定裁判之實證研究，台灣大學數位智能法院、法律科技與接 近正義研討會，2022。Donato Malerba, Floriana Esposito, Michelangelo Ceci and Annalisa Appice. Top-down induction of model trees with regression and splitting nodes. IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume: 26, Issue: 5, Pages: 612-625. 2004.林玠鋒，論家事財產法上法院之裁量調控-以扶養費、家庭生活費用及贍養費之酌付為中心，國立政治大學 法律學系所 博士論文，2014。謝天懷、賴俊穎、黃詩淳，老人扶養費請求事件之實證研究，裁判時報，第 115 期第84 - 95頁，2022。Available: https://doi.org/10.53106/207798362022010115008林岡毅，以資訊技術分析我國離婚贍養費相關裁判，國立臺灣大學 科際整合法律學研究所 碩士論文，2018。陳冠群，中文裁判書之要旨擷取：以最高法院裁判書為例，國立政治大學 碩士論文，2018。P.-H. Li, T.-J. Fu, and W.-Y. Ma, Why Attention? Analyze BiLSTM Deficiency and Its Remedies in the Case of NER, Proceedings of AAAI Conference on Artificial Intelligence, Volume: 34, no. 5. 2019.Gerard Salton and Chris Buckley. Term Weighting Approaches in Automatic Text Retrieval, Information Processing & Management, Volume: 24, Issue: 5, Pages: 513-523. 1988.Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. 2018.Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin. Attention Is All You Need. 2017.N. Reimers and I. Gurevych. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. 2019.Jingye Li, Hao Fei, Jiang Liu, Shengqiong Wu, Meishan Zhang, Chong Teng, Donghong Ji, Fei Li, Unified Named Entity Recognition as Word-Word Relation Classification, Proceedings of the 36th AAAI Conference on Artificial Intelligence, Volume: 36, no. 10., 2022. Available: https://doi.org/10.48550/arXiv.2112.10070 | zh_TW |
