%0 Journal Article %T 融合共现网络特征与知识增强语义梯度提升电子邮件分类
Gradient Boosted Email Classification through Integration of Co-Occurrence Network Features and Knowledge-Enhanced Semantics %A 艾均 %A 邹智洋 %A 苏湛 %A 耿爱国 %A 马菀言 %J Modeling and Simulation %P 222-237 %@ 2324-870X %D 2025 %I Hans Publishing %R 10.12677/mos.2025.143217 %X 本文针对现有电子邮件分类算法缺乏知识网络特征，并且训练复杂度较高的问题，应用复杂网络理论和知识增强语义模型，设计了一种基于电子邮件知识共现网络特征和知识增强语义的梯度提升算法，研究如何利用电子邮件知识网络和增强深度学习模型的知识表征来提升分类算法性能。首先，利用词汇共现度构建基于电子邮件知识的共现网络；其次，采用维瓦尔第算法将共现网络的节点映射到张量空间，生成对应知识节点空间嵌入；然后，计算共现网络模型的中心性特征并与维瓦尔第语义空间嵌入相结合，再融合知识增强语义模型生成的文本语义特征；最后，使用梯度增强算法实现电子邮件分类学习。在实验中，相较于现在的领先模型，本文算法在准确率、精确率和召回率等指标上均有明显提升，验证了其有效性，揭示了电子邮件知识网络特征能够有效增强现有模型的性能，提供了对其表征能力的有效补充。
In this paper, for the problem that existing email classification algorithms lack knowledge network features and have high training complexity, a gradient boosting algorithm based on email knowledge co-occurrence network features and knowledge enhancement semantics is designed by applying the complex network theory and knowledge enhancement semantics model to study how to improve the performance of classification algorithms by using the email knowledge network and knowledge representation of the augmented deep learning model. Firstly, the lexical co-occurrence is used to construct a co-occurrence network based on email knowledge; secondly, the Vivaldi algorithm is used to map the nodes of the co-occurrence network to the tensor space to generate the corresponding knowledge node space embedding; then, the centrality feature of the co-occurrence network model is calculated and combined with the Vivaldi semantic space embedding, and then the text semantic features generated by the knowledge-enhanced semantic model are fused; finally, the gradient boosting algorithm is used to achieve email classification learning. In the experiments, compared with the current leading model, the algorithm in this paper has obvious improvement in the indexes of accuracy, precision and recall, which verifies its effectiveness and reveals that the email knowledge network features can effectively enhance the performance of the existing model and provide an effective complement to its representational capability. %K 电子邮件分类， %K 文本分类， %K ERNIE， %K 复杂网络， %K XGBoost
Email Classification %K Text Classification %K ERNIE %K Complex Network %K XGBoost %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=109610