Reimbursement Account Information Sheet.

Accurate classification of budget items is a critical component of financial reimbursement, as it determines the legitimacy and regulatory compliance of financial expenditures. Currently, manual classification of reimbursement budget items faces to two challenges of inefficiency and inaccuracy. This is primarily due to the labor-intensive nature of the task, which increases the likelihood of selecting incorrect categories. To address these challenges, this study proposed a WeNet-Random Forest (WeNet-RF) model, which leverages speech recognition technology (WeNet) and Random Forest (RF) to improve efficiency and classification accuracy. WeNet-RF includes four steps: speech identification, features extraction, items classification, and evaluated model.This study compared WeNet-RF with Convolutional Neural Networks (CNN), Logistic Regression (LR) and K-Nearest Neighbors (KNN). WeNet-RF was verified by 50 real financial reimbursement records, and the results show that accuracy rate, precision rate, recall rate, and F1 score of WeNet-RF all are 90.77%. The findings provide a robust solution for improving financial management processes, and a reference model to financial management system.

预算科目精准分类是财务报销的核心环节，其直接决定了财务支出的合法性与监管合规性。当前，人工对报销预算科目进行分类面临效率低下与准确率不足两大挑战，这主要源于该任务需耗费大量人力，进而增加了类别选择错误的概率。为解决上述问题，本研究提出了WeNet-随机森林（WeNet-Random Forest, WeNet-RF）模型，该模型融合语音识别技术（WeNet）与随机森林（Random Forest, RF）算法，以提升分类效率与准确率。WeNet-RF模型包含四个步骤：语音识别、特征提取、科目分类与模型评估。本研究将WeNet-RF与卷积神经网络（Convolutional Neural Networks, CNN）、逻辑回归（Logistic Regression, LR）以及K近邻（K-Nearest Neighbors, KNN）进行了对比实验，并通过50条真实财务报销记录对模型进行验证。实验结果表明，WeNet-RF的准确率、精准率、召回率与F1值均达到90.77%。该研究成果为优化财务管理流程提供了可靠解决方案，同时也为财务管理系统提供了可借鉴的参考模型。