With the rapid development of information technology, a large number of sensitive words have appeared, which has brought great harm to network security and social stability. Therefore, how to identify and classify these sensitive information accurately has become an important issue. Combined with the improved artificial network algorithm, an adaptive classification algorithm is proposed in the experiment, which can provide local intelligent classification service according to the classification results. At the same time, the algorithm transforms the clustering model structure of the traditional network algorithm, introduces the classification information, so that it can be applied to the classification problem, and expands the application field of the data field theory. The experimental results show that: (1) From the experimental results of different text quantities, the SOM algorithm assigns the classification task to different levels of nodes, and realizes the modularization of detection. (2) The overall mean results show that the highest recall rate is 87%, which has met the basic grading criteria, and the detection accuracy of sensitive words will also be improved. (3) The experimental results show that the algorithm can accurately classify the sensitive and speed up the parameter optimization, and is superior to the comparison algorithm in many indicators. (4) From the simulation results, compared with the traditional neural network algorithm, the precision and recall of the algorithm are maintained at more than 90% and the loss is less than 0.11.

随着信息技术的快速发展，大量敏感词的出现，给网络安全和社会稳定带来了极大危害。因此，如何对这些敏感信息进行准确识别和分类成为一个重要问题。实验中结合改进的人工网络算法，提出了一种自适应分类算法，可以根据分类结果提供本地智能分类服务。同时，该算法对传统网络算法的聚类模型结构进行改造，引入分类信息，使其能够应用于分类问题，拓展了数据场理论的应用领域。实验结果表明：（1）SOM算法根据不同文本量的实验结果，将分类任务分配给不同级别的节点，实现检测的模块化。（2）总体平均结果显示，召回率最高为87%，已经满足基本的分级标准，敏感词的检测准确率也将得到提高。(3)实验结果表明，该算法能够准确分类敏感对象，加快参数优化速度，在多项指标上优于对比算法。(4)从仿真结果来看，与传统的神经网络算法相比，该算法的查准率和查全率均保持在90%以上，损失小于0.11。SOM算法将分类任务分配给不同级别的节点，实现检测的模块化。（2）总体平均结果显示，召回率最高为87%，已经满足基本的分级标准，敏感词的检测准确率也将得到提高。(3)实验结果表明，该算法能够准确分类敏感对象，加快参数优化速度，在多项指标上优于对比算法。(4)从仿真结果来看，与传统的神经网络算法相比，该算法的查准率和查全率均保持在90%以上，损失小于0.11。SOM算法将分类任务分配给不同级别的节点，实现检测的模块化。（2）总体平均结果显示，召回率最高为87%，已经满足基本的分级标准，敏感词的检测准确率也将得到提高。(3)实验结果表明，该算法能够准确分类敏感对象，加快参数优化速度，在多项指标上优于对比算法。(4)从仿真结果来看，与传统的神经网络算法相比，该算法的查准率和查全率均保持在90%以上，损失小于0.11。已经达到了基本的分级标准，敏感词的检测准确率也会得到提高。(3)实验结果表明，该算法能够准确分类敏感对象，加快参数优化速度，在多项指标上优于对比算法。(4)从仿真结果来看，与传统的神经网络算法相比，该算法的查准率和查全率均保持在90%以上，损失小于0.11。已经达到了基本的分级标准，敏感词的检测准确率也会得到提高。(3)实验结果表明，该算法能够准确分类敏感对象，加快参数优化速度，在多项指标上优于对比算法。(4)从仿真结果来看，与传统的神经网络算法相比，该算法的查准率和查全率均保持在90%以上，损失小于0.11。