| dc.contributor | 金融系 | |
| dc.creator (作者) | 江彌修 | |
| dc.date (日期) | 2021-10 | |
| dc.date.accessioned | 7-Apr-2026 13:17:26 (UTC+8) | - |
| dc.date.available | 7-Apr-2026 13:17:26 (UTC+8) | - |
| dc.date.issued (上傳時間) | 7-Apr-2026 13:17:26 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=181929 | - |
| dc.description.abstract (摘要) | 在此兩年期的研究計畫提案中，我們希望辨認出足以強化企業違約預測模型準確度的具體關鍵：我們指出，基於潛在語意主題分析，有效的企業違約預測必須源於其特徵參數的具體辨認及最適配置。針對此議題，本計劃預計首先透過LDA模型（及其延伸）挖掘公司財務報表中文字的潛在主題，將這些主題變數、傳統財務變數與總體經濟變數相結合作為模型之特徵變數集合。基於此LDA模型的分析基礎，我們預計利用不同的機器學習方法（支持向量機、決策樹以及人工神經網路）進行企業的違約機率預測與破產與否的分類預測上。更重要的，我們將透過建立完成的主題模型衡量其參數擾動之於模型預測能力之影響：就違約機率的預測上，我們將以複雜值以檢驗參數的配置之於各模型的預測表現，在分類預測上，我們將以真實分類與預測分類之混淆矩陣來建立預測模型的最適參數配置。藉由本研究計畫的分析結果，我們將闡述文本訊息所揭露的潛在語意特徵之於公司違約預測所扮演的重要角色。 | |
| dc.description.abstract (摘要) | Over a two-year period, this research proposal intends to quantify the vital issues that attribute to possible enforcements of the projected accuracy in corporate default predictions. We aim to explore, based on latent semantic topic learning, how an effective default prediction can be achieved with an optimal allocation of the latent topic-characteristics. To proceed, we shall first rely on the Latent Dirichlet Allocation (LDA) model (and its variants) to extract the latent topics (relating to corporate defaults) that embedded within corporate financial reports. These latent topics, once identified, are then combined with the traditional macro- and financial- default related risk factors to form a set of model attributes. Using these model attributes, machine learning techniques (Support Vector Machines, Decision Trees, and Neural Networks) are then introduced to proceed with the predications of corporate default probabilities and the binary predication of bankruptcy events. Most importantly, we shall investigate how the default prediction outcomes are subject to attribute perturbation. For the prediction of corporate default probabilities, we shal employ the measure of Perplexity value to examine the impacts of different model-attribute configurations on the prediction outcomes. For the binary predication of bankruptcy events, we shall postulate about an optimal configuration for model attributes by studying the information revealed by the (classifier’s’) confusion matrix. Over all, this research project intends to establish the important role of latent semantic learning over corporate default prediction. | |
| dc.format.extent | 116 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | 科技部, MOST108-2410-H004-080-MY2, 108.08-110.07 | |
| dc.subject (關鍵詞) | 文字探勘; 主題模型; 企業財務預警; 機器學習 | |
| dc.subject (關鍵詞) | text mining; topic model; corporate default prediction; machine learning | |
| dc.title (題名) | 基於潛在語意主題學習的企業違約預警 | |
| dc.title (題名) | Corporate Default Prediction Using Latent Semantic Topic Modeling | |
| dc.type (資料類型) | report | |
