基于LERT-CRNN-KAN的110警情分类方法研究

doi:10.11871/jfdc.issn.2096-742X.2025.02.010

数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (2): 96-108.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.02.010

doi: 10.11871/jfdc.issn.2096-742X.2025.02.010

基于LERT-CRNN-KAN的110警情分类方法研究

刘卓娴¹(),石拓^2,^*(),胡啸峰¹

1.中国人民公安大学，信息网络安全学院，北京 100038
2.北京警察学院，公安管理系，北京 102202

收稿日期:2024-11-05 出版日期:2025-04-20 发布日期:2025-04-23
通讯作者: 石拓
作者简介:刘卓娴，现为中国人民公安大学在读研究生，CCF学生会员，主要研究方向为自然语言处理、警情挖掘等。
本文主要承担工作为方法提出，实验执行，论文撰写。
LIU Zhuoxian, is currently a graduate student at the People’s Public Security University of China and a student member of CCF. Her main research areas include Natural Language Processing (NLP), police situation mining, etc.
In this paper, she is mainly responsible for proposing methods, conducting experiments, and writing the manuscript.
E-mail:1240875185@qq.com|石拓，博士，现为北京警察学院教授，主要研究方向为人工智能、公安大数据等。
本文主要承担工作为问题提出，数据支撑，论文修改。
SHI Tuo, Ph.D., is a professor at the Beijing Police College. Her main research areas include artificial intelligence and public security big data.
In this paper, she is mainly responsible for proposing the research problem, providing data support, and revising the manuscript.
E-mail: stshi8808@sina.com
基金资助:
北京市自然科学基金“基于多源数据的北京市电信网络诈骗被害分析及精准预防策略研究”(9244025)

Research on 110 Emergency Call Incident Classification Method Based on LERT-CRNN-KAN

LIU Zhuoxian¹(),SHI Tuo^2,^*(),HU Xiaofeng¹

1. Department of Information Network Security, PPSUC, Beijing 100038, China
2. Department of Public Security Management, Beijing Police College, Beijing 102202, China

Received:2024-11-05 Online:2025-04-20 Published:2025-04-23
Contact: SHI Tuo

摘要/Abstract

摘要：

【目的】为了有效解决基层公安机关在处理110警情特别是电信网络诈骗警情时的人工分类效率低和自动化分类效果差的问题，进一步提升警力资源的利用效率和实战效能。【方法】构建一种融合KAN算法、语言学信息增强文本预处理方法（LERT）、卷积神经网络（CNN）和双向长短期记忆网络（BiLSTM）的多通道神经网络警情分类模型。【结果】实验使用中国北方B市的真实警情数据进行测试，结果表明，在警情三分类任务中达到了91.9%的分类准确率，对比消融实验证实该模型优于基线模型。【结论】模型有效解决了110警情数据的分类问题，为基层公安机关提供了一种高效的智能化分类工具，满足了实战要求。其他应用场景有待进一步探索。

关键词: KAN, 警情分类, 多通道神经网络

Abstract:

[Purpose] This paper aims to address the issues of low manual classification efficiency and poor automated classification faced by grassroots public security organs in handling 110 emergency call incidents, especially those related to telecommunication network fraud, and to further enhance the utilization efficiency of police resources and operational effectiveness. [Method] A multi-channel neural network police call classification model integrating KAN algorithm, LERT (Linguistically-motivated bidirectional Encoder Representation from Transformer) for linguistic information-enhanced text preprocessing, CNN (Convolutional Neural Network), and BiLSTM (Bidirectional Long Short-Term Memory) is constructed. [Result] Experiments conducted using real 110 emergency call incident data from a city in northern China demonstrate that the model achieves a classification accuracy of 91.9%, and ablation experiments confirm its superiority to baseline models. [Conclusion] The model effectively addresses the classification problem of 110 emergency call incident data, providing an efficient and intelligent classification tool for grassroots public security organs and meeting operational requirements. Other application scenarios await further exploration.

Key words: KAN, 110 emergency call incident classification, multi-channel neural network

刘卓娴,石拓,胡啸峰. 基于LERT-CRNN-KAN的110警情分类方法研究[J]. 数据与计算发展前沿, 2025, 7(2): 96-108.

LIU Zhuoxian,SHI Tuo,HU Xiaofeng. Research on 110 Emergency Call Incident Classification Method Based on LERT-CRNN-KAN[J]. Frontiers of Data and Computing, 2025, 7(2): 96-108, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2025.02.010.

图/表 14

图1

图2

图3

表1

表2

表3

表4

表5

表6

表7

图4

图5

图6

表8

参考文献 31

[1]	谢玉鹏. 江西省公安机关警情数据采集与综合应用研究[D]. 南昌: 江西财经大学, 2017.
[2]	汤小军, 宋华. 提升基层派出所处置复杂警情能力研究[J]. 河北公安警察职业学院学报, 2021, 21(1): 5-8.
[3]	朱国富, 孙丽丽. 警情分类处置提效能创满意[N]. 人民公安报, 2022-09-20(006).
[4]	公安部: 2023年共破获电信网络诈骗案件43. 7万起[J]. 中国防伪报道, 2024(3): 7-8.
[5]	黄文婷. 公共秩序的界定: 基于香港判例的分析[J]. 港澳研究, 2022(1): 35-50.
[6]	刘鹏. 多发性侵财犯罪的打击与预防对策[J]. 四川警察学院学报, 2015, 27(5): 26-32.
[7]	张在翔. 数据挖掘技术在警情分析及预测上的应用[D]. 上海: 复旦大学, 2008.
[8]	殷小科, 王威, 王婕, 等. 分层文本分类在警情数据中的应用[J]. 现代计算机, 2021, 27(23): 86-90.
[9]	章磊, 王攀, 何芬. 自然语言处理在警情智能分析中的应用[J]. 警察技术, 2021(5): 39-43.
[10]	程春惠, 何钦铭. 面向不均衡类别朴素贝叶斯犯罪案件文本分类[J]. 计算机工程与应用, 2009, 45(35): 126-128.
[11]	张静, 高子信, 丁伟杰. 基于BERT-DPCNN的警情文本分类研究[J/OL]. 数据分析与知识发现, 1-15[2025-01-02]. http://kns.cnki.net/kcms/detail/10.1478.G2.20240313.1318.008.html.
[12]	王孟轩, 张胜, 王月, 等. 改进的CRNN模型在警情文本分类中的研究与应用[J]. 应用科学学报, 2020, 38(3): 388-400.
[13]	DEVLIN J, CHANG M W, LEE K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[J]. arxiv preprint arxiv:1810.04805.
[14]	CUI Y, CHE W, LIU T, et al. Revisiting pre-trained models for Chinese natural language processing[J]. arxiv preprint arxiv:2004.13922.
[15]	SUN C, QIU X, XU Y, et al. How to fine-tune bert for text classification?[C]// Chinese computational linguistics:18th China national conference, CCL 2019, Springer International Publishing, 2019: 194-206.
[16]	JEREMY H, SEBASTIAN R. Universal Language Model Fine-tuning for Text Classification[J]. arxiv preprint arxiv:1801.06146.
[17]	KIM Y. Convolutional Neural Networks for Sentence Classification[C]// Alessandro Moschitti, Bo Pang, Walter Daelemans. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP). Doha, Qatar: Association for Computational Linguistics, 2014: 1746-1751.
[18]	HARTMANN J, HUPPERTZ J, SCHAMP C, et al. Comparing automated text classification methods[J]. International Journal of Research in Marketing, 2019, 36(1): 20-38.
[19]	ZHAO W X, ZHOU K, LI J, et al. A Survey of Large Language Models[J]. arXiv preprint arXiv:2303.18223.
[20]	THOPPILAN R, DE FREITAS D, HALL J, et al. LAMDA: Language Models for Dialog Applications[J]. arXiv preprint arXiv:2201.08239.
[21]	ZHANG C, YANG Z, HE X, et al. Multimodal intelligence: Representation learning, information fusion, and applications[J]. IEEE Journal of Selected Topics in Signal Processing, 2020, 14(3): 478-493.
[22]	GAO J, LI P, CHEN Z, et al. A Survey on Deep Learning for Multimodal Data Fusion[J]. Neural Computation, 2020, 32: 829-864.
[23]	LIU Z, WANG Y, VAIDYA S, et al. Kan: Kolmogorov-Arnold Networks[J]. arxiv preprint arxiv:2404.19756.
[24]	CUI Y, CHE W, WANG S, et al. Lert: A linguistically-motivated pre-trained language model[J]. arxiv preprint arxiv:2211.05344.
[25]	LAI S, XU L, LIU K, et al. Recurrent convolutional neural networks for text classification[C]// Proceedings of the AAAI conference on artificial intelligence. 2015, 29(1).
[26]	YAMASHITA R, NISHIO M, DO R K, et al. Convolutional neural networks: an overview and application in radiology[J]. Insights into imaging, 2018, 9: 611-629.
[27]	MUSTAPHA R, ZGALLAI W A, OZSAHIN D U, et al. Chapter 2-Convolution neural network and deep learning[J]. Artificial Intelligence and Image Processing in Medical Imaging, 2024: 21-50.
[28]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[29]	STAUDEMEYER R C, MORRIS E R. Understanding LSTM-A Tutorial into Long Short-Term Memory Recurrent Neural Networks[J]. arxiv preprint arxiv:1909.09586.
[30]	MAHTO D, YADAV S C. Hierarchical Bi-LSTM based emotion analysis of textual data[J]. Bulletin of the Polish Academy of Sciences Technical Sciences, 2022: e141001.
[31]	TA H T. BSRBF-KAN: A Combination of B-splines and Radial Basic Functions in Kolmogorov-Arnold Networks[J]. arxiv preprint arxiv:2406.11173.

名称	版本
numpy	1.23.5
openpyxl	3.0.10
pandas	2.1.4
python	3.11.7
scikit-learn	1.3.0
tokenizers	0.15.0
transformers	4.36.2
operating system	Win11
openpyxl	NVIDIA GeForce RTX 4060
batch_size	8
learning rate	1e-6
epoch	3
optimizer	Adam

名称	值
convolutional dimension	1 d
embedding layer dimension	768
number of convolutional kernels	128
kernel_size	1
stride	1

模型	Acc	P	R	F1
LERT-CNN-BILSTM	0.9167	0.9225	0.9167	0.9168
MACBERT-CNN-BILSTM	0.9056	0.9066	0.9056	0.9056
ROBERTA-CNN-BILSTM	0.5467	0.3986	0.5467	0.4470
BERT-CNN-BILSTM	0.9144	0.9196	0.9144	0.9135

模型	Acc	P	R	F1
LERT-CNN-BILSTM	0.9167	0.9225	0.9167	0.9168
LERT-CNN	0.6589	0.4978	0.6589	0.5502
LERT-BILSTM	0.9100	0.9142	0.9100	0.9100

模型	Acc	P	R	F1
LERT-CNN-BILSTM (+KAN)	0.9189	0.9255	0.9189	0.9189
LERT-CNN-BILSTM	0.9167	0.9225	0.9167	0.9168

基于LERT-CRNN-KAN的110警情分类方法研究

Research on 110 Emergency Call Incident Classification Method Based on LERT-CRNN-KAN

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 31

相关文章 0

编辑推荐

Metrics

本文评价

模型	Acc	P	R	F1	训练速率(it/s)	测试速率(it/s)
LERT-CNN-BILSTM (+KAN)	0.9189	0.9255	0.9189	0.9189	7.50	47.62
LERT-CNN-BILSTM(+MLP)	0.9078	0.9082	0.9078	0.9079	7.35	47.62

模型	Acc	P	R	F1
LERT-CNN-BILSTM (+KAN)	0.9189	0.9255	0.9189	0.9189
CRNN	0.8667	0.8669	0.8667	0.8666
BERT-DPCNN	0.8556	0.8567	0.8556	0.8559