近日，我协会成员撰写的论文《Implicit Offensive Speech Detection Based on Multi-feature Fusion》在CCF-C级会议International Conference on Knowledge Science, Engineering and Management (KSEM)上成功发表。第一作者为郭腾达（2021级网安专业本科生），合作作者为林连鑫（2021级网安专业本科生）、郑承平（2021级网安专业本科生）、刘航（2021级网安专业本科生）、涂致坚（2021级网安专业本科生），指导老师为王海舟副教授。
本论文针对网络上攻击言论越发隐晦和猖獗的现状，构建了第一个中文隐性攻击言论数据集，并借助预训练模型和多任务学习实现了高效率的隐性攻击言论检测模型。作者首先对中国最大的社交媒体平台之一的新浪微博展开调研，对平台海量言论特征进行了分析挖掘，并总结出了隐性攻击言论的突出特征：语义、情感、隐喻和谬误特征。在此基础上，该项目依据以上特征和可信分类信源，爬取大量发言数据并进行了筛选和标注，构建了WIOS（Weibo Implicit Offensive Speech）数据集。此数据集为目前最有时效性和可信度的中文攻击言论数据集，包含和谐言论、显性攻击言论和隐性攻击言论三种标签。之后，作者将各特征SOTA模型改进为特异性适配该项目的特征提取器，借助BERT模型提取有效embedding，并设计了更符合隐性攻击言论检测任务的多输入Loss函数，构建了BMA（BERT-Mate-Ambiguity）检测模型。通过实验验证，该模型拥有更优秀的分类能力和更高的鲁棒性，作者还通过表示学习将模型训练结果更加直观地进行了展示，表现出BMA模型的优异性能。
BMA隐性攻击言论检测模型架构图
会议简介：
KSEM全名The International Conference on Knowledge Science, Engineering and Management，是CCF-C级国际会议；主要研究的主题涉及知识科学、知识工程、知识管理、知识图谱等方面的内容。
官方介绍：The importance of studying knowledge science, knowledge engineering and knowledge management has been recognized widely. We could cite Feigenbaum’s words in his book The fifth generation, Artificial Intelligence and Japan’s Computer challenge to the World, for demonstrating the point:
- “Knowledge is not the same as information. Knowledge is Information that has been pared, shaped, interpreted, selected, and transformed”
- “Knowledge is power, and the computer is an amplifier of that power. We are now at the dawn of a new computer revolution…the transition from information processing to knowledge processing”
- “…establish a ‘knowledge industry’ in which knowledge itself will be a salable commodity like food and oil. Knowledge itself is to become the wealth of nations”.
Our Aim
- Making available information about KSEM conferences among people who are interested in it.
- Serving as a representative forum for experts within the KSEM related field.
- Distributing KSEM related information and knowledge, e.g papers, projects, news, discussions and even blogs
论文链接：https://link.springer.com/chapter/10.1007/978-3-031-40286-9_3