互联网上的攻击性言论严重扰乱了正常网络秩序,破坏了健康交流的网络环境。现有的检测技术更关注文本中的鲜明特征,难以发现更隐晦的攻击方式。针对上述问题,提出融合反讽机制的攻击性言论检测模型BSWD(Bidirectional Encoder Represent...互联网上的攻击性言论严重扰乱了正常网络秩序,破坏了健康交流的网络环境。现有的检测技术更关注文本中的鲜明特征,难以发现更隐晦的攻击方式。针对上述问题,提出融合反讽机制的攻击性言论检测模型BSWD(Bidirectional Encoder Representation from Transformers-based Sarcasm and Word Detection)。首先,提出基于反讽机制的模型Sarcasm-BERT,以检测言论中的语义冲突;其次,提出细粒度词汇攻击性特征提取模型WordsDetect,检测言论中的攻击性词汇;最后,融合两种模型得到BSWD。实验结果表明,与BERT(Bidirectional Encoder Representation from Transformers)、HateBERT模型相比,所提模型的准确率、精确率、召回率和F1分数指标大部分能提升2%,显著提高了检测性能,更能发现隐含的攻击性言论;同时,与SKS(Sentiment Knowledge Sharing)、BiCHAT(Bidirectional long shortterm memory with deep Convolution neural network and Hierarchical ATtention)模型相比,具有更强的泛化能力和鲁棒性。以上结果验证了BSWD检测隐晦攻击性言论的有效性。展开更多
HTTP-flooding attack disables the victimized web server by sending a large number of HTTP Get requests.Recent research tends to detect HTTP-flooding with the anomaly-based approaches,which detect the HTTP-flooding by ...HTTP-flooding attack disables the victimized web server by sending a large number of HTTP Get requests.Recent research tends to detect HTTP-flooding with the anomaly-based approaches,which detect the HTTP-flooding by modeling the behavior of normal web surfers.However,most of the existing anomaly-based detection approaches usually cannot filter the web-crawling traces from unknown searching bots mixed in normal web browsing logs.These web-crawling traces can bias the base-line profile of anomaly-based schemes in their training phase,and further degrade their detection performance.This paper proposes a novel web-crawling tracestolerated method to build baseline profile,and designs a new anomaly-based HTTP-flooding detection scheme(abbr.HTTP-sCAN).The simulation results show that HTTP-sCAN is immune to the interferences of unknown webcrawling traces,and can detect all HTTPflooding attacks.展开更多
文摘互联网上的攻击性言论严重扰乱了正常网络秩序,破坏了健康交流的网络环境。现有的检测技术更关注文本中的鲜明特征,难以发现更隐晦的攻击方式。针对上述问题,提出融合反讽机制的攻击性言论检测模型BSWD(Bidirectional Encoder Representation from Transformers-based Sarcasm and Word Detection)。首先,提出基于反讽机制的模型Sarcasm-BERT,以检测言论中的语义冲突;其次,提出细粒度词汇攻击性特征提取模型WordsDetect,检测言论中的攻击性词汇;最后,融合两种模型得到BSWD。实验结果表明,与BERT(Bidirectional Encoder Representation from Transformers)、HateBERT模型相比,所提模型的准确率、精确率、召回率和F1分数指标大部分能提升2%,显著提高了检测性能,更能发现隐含的攻击性言论;同时,与SKS(Sentiment Knowledge Sharing)、BiCHAT(Bidirectional long shortterm memory with deep Convolution neural network and Hierarchical ATtention)模型相比,具有更强的泛化能力和鲁棒性。以上结果验证了BSWD检测隐晦攻击性言论的有效性。
基金supported by National Key Basic Research Program of China(973 program)under Grant No.2012CB315905National Natural Science Foundation of China under grants 61172048,61100184,60932005 and 61201128the Fundamental Research Funds for the Central Universities under Grant No ZYGX2011J007
文摘HTTP-flooding attack disables the victimized web server by sending a large number of HTTP Get requests.Recent research tends to detect HTTP-flooding with the anomaly-based approaches,which detect the HTTP-flooding by modeling the behavior of normal web surfers.However,most of the existing anomaly-based detection approaches usually cannot filter the web-crawling traces from unknown searching bots mixed in normal web browsing logs.These web-crawling traces can bias the base-line profile of anomaly-based schemes in their training phase,and further degrade their detection performance.This paper proposes a novel web-crawling tracestolerated method to build baseline profile,and designs a new anomaly-based HTTP-flooding detection scheme(abbr.HTTP-sCAN).The simulation results show that HTTP-sCAN is immune to the interferences of unknown webcrawling traces,and can detect all HTTPflooding attacks.
