基于朴素贝叶斯算法的中文垃圾短信过滤模型研究

Research on Chinese Spam SMS Filtering Model Based on Naive Bayes Algorithm

摘要: 针对传统关键词过滤方法在中文垃圾短信识别中自适应差、误报率高的问题，本文实现并评估了一套基于朴素贝叶斯算法的高效、轻量级过滤工具。首先，构建中文短信语料库，并进行分词、去停用词等预处理；其次，采用TF-IDF结合N-gram进行特征提取，并通过消融实验量化各处理步骤的贡献；最后，基于Scikit-learn实现多项式朴素贝叶斯分类器，通过网格搜索进行参数调优。实验结果表明，该工具在测试集上准确率达88.2%，F1值为0.897，误报率仅6.6%，在训练效率和资源占用上相较对比算法具有一定优势。该工作为垃圾短信治理提供了一种工程上可行、易于部署的轻量级技术参考。

Abstract: To address the issues of poor adaptability and high false-positive rates in traditional keyword-based filtering methods for Chinese spam SMS identification, this paper implements and evaluates an efficient and lightweight filtering tool based on the Naive Bayes algorithm. First, a Chinese SMS corpus is constructed and preprocessed through tokenization and stop-word removal. Second, TF-IDF combined with N-gram is employed for feature extraction, with the contribution of each preprocessing step quantified through ablation studies. Finally, a Multinomial Naive Bayes classifier is implemented using Scikit-learn and optimized via grid search for parameter tuning. Experimental results show that the tool achieves an accuracy of 88.2% and an F1-score of 0.897 on the test set, with a false-positive rate of 6.6%, demonstrating certain advantages over comparison algorithms in training efficiency and resource consumption. This work provides a lightweight, practically feasible, and easily deployable technical solution for spam SMS governance.