| dc.description.abstract | 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. | en_US |