智能回复系统研究综述  

Research on Intelligent Reply System: A Survey

作　　者：岳世峰 林政[1] 王伟平[1] 孟丹[1] YUE Shifeng;LIN Zheng;WANG Weiping;MENG Dan(National Engineering Laboratory for Information Security Technologies,Institute of Information Engineering,Chinese Academy of Sciences,Beijing 100093,China;School of Cyber Security,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院信息工程研究所信息内容安全技术国家工程实验室,中国北京100093 [2]中国科学院大学网络空间安全学院,中国北京100049

出　　处：《信息安全学报》2020年第1期20-34,共15页Journal of Cyber Security

基　　金：国家自然科学基金(No.61502478,No.61602467);国家高技术研究发展计划(No.“863”计划)基金资助项目(No.2013AA013204)

摘　　要：网络舆情形成快、影响大,如何对其进行智能导控一直是网络安全中的难点和重点。本文提出使用智能回复系统对网络舆情进行自动引导的观点,然后对智能回复系统研究进行了综述。本文首先介绍了当前智能回复系统的主流研究方向,如视觉问答、基于知识图谱的问答和推理问答等不同类型的智能回复系统;接着根据应用场景的不同分别介绍了垂直领域和开放领域的智能回复系统,然后从技术手段上对实现智能回复系统的各种主流方法进行了详细的介绍和探讨。最后本文总结归纳了当前智能回复系统的自动评价方法以及当前智能回复系统存在的主要问题及未来可能的研究方向。Online public opinion develops rapidly and has a great impact on society. It has always been difficult and important to lead it in network security. In this paper, we propose the idea of using intelligent reply system to lead the online public opinion, and then we provide a general overview of the current intelligent reply system. First, we introduce the current mainstream research methods for the intelligent reply system, including the visual question answering, knowledge-based question answering and inference question answering;second, we demonstrate the task-oriented and untask-oriented intelligent reply syst em in terms of different application scenarios. After that we discussed the mainstream methods of building reply system. Finally we summarize the automatic evaluation methods of the current reply system and the main problems as well as future research directions in this field.

关 键 词：网络舆情 深度学习 问答系统 生成模型 检索模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]