| 引用本文： |
-
李志鹏,杨阳朝,廖勇,俞能海,吴哲,谢海永,石珺,曾曦.数据驱动的事件预测技术最新研究进展[J].信息安全学报,2022,7(1):40-55 [点击复制]
- LI Zhipeng,YANG Yangzhao,LIAO Yong,YU Nenghai,WU Zhe,XIE Haiyong,SHI Jun,ZENG Xi.A Survey of Recent Advances in Data-Driven Event Prediction Research[J].Journal of Cyber Security,2022,7(1):40-55 [点击复制]
|
|
| |
|
|
| 本文已被：浏览 9229次 下载 7501次 |
码上扫一扫！ |
| 数据驱动的事件预测技术最新研究进展 |
|
李志鹏1,2, 杨阳朝2, 廖勇1,2, 俞能海1, 吴哲2, 谢海永1, 石珺2, 曾曦2
|
|
|
| (1.中国科学技术大学网络空间安全学院 合肥 中国 230027;2.深圳市网联安瑞网络科技有限公司 深圳 中国 518038) |
|
| 摘要: |
| “事件”（event）是指在特定时空发生的对人类社会和自然界产生较为明显影响的事情。社会动乱、暴恐事件、传染病大流行等例子是给国家和社会安全带来严重威胁的“事件”。如果能够提前对这些事件的发生进行有效预测，将有助于做好应对准备，大大减少不必要的损失，因此事件预测技术在实际中具有重大社会应用价值，能够在社会安全、风险感知、传染病防控等方面发挥重要作用。对事件进行科学准确的预测曾经是一个十分具有挑战性的问题，近期大数据和数据挖掘的发展为事件预测技术带来了新的机遇。本文就以数据驱动的事件预测技术最新研究进展做一系统化的综述，介绍了事件预测的形式化建模与性能度量指标，对事件预测技术领域的最新研究成果进行了科学归类与总结，分为频繁模式挖掘、传统分类模型、时间序列预测、时序点过程、地理空间位置预测、事件图谱、无监督方法、多技术融合预测八大类方法，将每类方法做了系统地阐述，接着探讨了事件预测技术的主要应用领域，最后展望了这一技术未来面临的挑战和潜在的研究方向，以期进一步推动事件预测技术的发展与应用。 |
| 关键词: 计算社会学 机器学习 数据挖掘 事件图谱 事件预测 |
| DOI：10.19363/J.cnki.cn10-1380/tn.2022.01.03 |
| 投稿时间：2020-12-19修订日期：2021-04-27 |
| 基金项目:本课题得到国家自然科学基金联合基金项目（No.U19B2044）资助。 |
|
| A Survey of Recent Advances in Data-Driven Event Prediction Research |
|
LI Zhipeng1,2, YANG Yangzhao2, LIAO Yong1,2, YU Nenghai1, WU Zhe2, XIE Haiyong1, SHI Jun2, ZENG Xi2
|
| (1.School of Cyberspace Science, University of Science and Technology of China, Hefei 230027, China;2.Shenzhen CyberAray Co., Ltd., Shenzhen 518038, China) |
| Abstract: |
| An event is usually defined as an incident that takes place in a specific location and time, impacting either the society or the nature in a nontrivial way. Civil unrest, terrorist attacks and pandemics are commonly known events that can pose serious threats to both national security and public safety. Effectively predicting upcoming events is highly desirable in reality, as successful predictions can be of great use in designing countermeasures to prevent or reduce potential losses. Predicting events in advances plays an important role in public security, risk perception and infectious disease prevention and control, etc. Although event prediction used to be a technically challenging task, recent advances in the fields of big data analytics and machine learning have brought promising opportunities in applying these techniques to solve real world prediction problems. This paper presents a systematic survey on data-driven event prediction research studies. We first introduce the formal definition of event prediction problem and the evaluation metrics of various prediction techniques. Then, the state-of-the-art algorithms and schemes proposed in the field of event prediction are summarized and classified. All existing event prediction methods can be classified into 8 categories: frequent pattern mining, traditional classification model, time series model, temporal point processes, geospatial predictive modeling, event knowledge graph, unsupervised machine learning, multi-model fusion method. We present a systematic and comprehensive summary for each category of these methods. After that, we proceed by discussing the main real-world applications of event prediction techniques, including public security, disease prevention, smart city and natural disaster prediction. We conclude this survey by summarizing multiple open research problems and several possible directions on event prediction research. To the best of our knowledge, this study provides a comprehensive summarization of recent research and advances on data-driven event prediction. We hope this paper can be a useful booster in the research of event prediction and applying this technique in solving real world problems. |
| Key words: computational social science machine learning data mining event graph event prediction |
|
|
|
|
|
