深度學習模型於客服對話文本多標籤分類之研究;Research on Deep Learning Model for Multi-Label Classification of Customer Service Dialogue

NCUIR > School of Management at National Central University > Executive Master of Information Management > Electronic Thesis & Dissertation > Item 987654321/95474

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95474

Title:	深度學習模型於客服對話文本多標籤分類之研究;Research on Deep Learning Model for Multi-Label Classification of Customer Service Dialogue
Authors:	林勇志;Lin, Yung-Chih
Contributors:	資訊管理學系在職專班
Keywords:	多標籤分類;深度學習;獎懲矩陣;客服對話文本;multi-label classification;deep learning;reward-penalty matrix;customer service dialogue
Date:	2024-06-26
Issue Date:	2024-10-09 16:53:29 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著數位科技的快速發展，傳統的客服模式已經發生了巨大的改變。企業需要透過數據分析，從多種管道蒐集和洞察顧客資訊，以提供個人化的服務。然而疫情期間大量專業人力流失，企業面臨著業務推展的困境，亟需在短期內尋求有效的解決方案。但龐大的數據，包括客服語音資料在內，需要經過複雜的處理和轉化才能加以應用，傳統的人工標註和冗長的分類作業已無法滿足現實需求。為了因應這些挑戰，本研究利用深度學習技術，提出了一種基於BERT預訓練模型的創新深度學習模型RPC-BERT，通過融合自適應權重衰減、自適應學習率和自定義機率獎懲係數，有效地降低了多標籤分類中的類別不平衡問題。所提出的RPC優化矩陣能有效地應用於客服對話內容的多標籤分類，相較其他深度學習模型，在準確率、精確率、召回率及F1各項評估指標皆有更佳的表現。此外透過實際案例進行模型的可用性驗證，以RPC優化矩陣機制來進行客服對話文本多標籤的分類處理，其結果除了符合研究之結論外，確認可運用於企業日常實務作業中，並滿足降低人力成本，提升作業效率的需求。;In the wake of rapid digital technological advancements, traditional customer service paradigms have undergone substantial metamorphosis. Enterprises are now compelled to leverage data analytics to collate and derive insights from multifarious channels of customer information, with the aim of delivering personalized services. However, the pandemic period has precipitated a significant exodus of specialized human capital, presenting corporations with formidable challenges in business expansion. This exigency necessitates the expeditious identification of efficacious solutions. The voluminous data, inclusive of customer service voice data, demands intricate processing and transformation prior to application. Conventional manual annotation and protracted classification procedures have become inadequate in meeting contemporary demands. To address these challenges, this research harnesses deep learning technologies to propose an innovative deep learning model, RPC-BERT, predicated on the BERT pre-training model. Through the integration of adaptive weight decay, adaptive learning rate, and customized probability reward-penalty coefficients, the model effectively mitigates class imbalance issues in multi-label classification tasks. The proposed RPC optimization matrix demonstrates efficacious application in multi-label classification of customer service dialogue content. Compared to other deep learning models, it exhibits superior performance across various evaluation metrics, including accuracy, precision, recall, and F1 score. Furthermore, the model′s applicability is validated through practical case studies. The implementation of the RPC optimization matrix mechanism for multi-label classification of customer service dialogue texts not only corroborates the research conclusions but also confirms its viability for integration into daily enterprise operations. This approach satisfies the dual objectives of reducing human resource costs and enhancing operational efficiency.
Appears in Collections:	[Executive Master of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	43	View/Open

社群 sharing

Loading...