常见中文社交平台中网络欺凌语言的检测分析

Detection and Analysis of Cybernetics Bullying Language on Common Chinese Social Network Platforms

当今中文社交平台中网络欺凌语言十分盛行,而传统的平台管理员人工审核的方式已无法有效地对其进行检测与分析.为解决这一难题,首先,我们提取了十几个典型的中文社交平台中的部分样本进行人工标注,构建了一个训练数据集.然后,我们分别使用朴素贝叶斯、支持向量机、长短期记忆神经网络构建分类模型,对未标注的数据进行分类识别处理.实验表明:选取的分类模型均能有效地识别出网络欺凌语言,其准确率分别是0.87,0.79,0.88.其中长短期记忆神经网络综合效果最佳.由此得出的结论为:借助大数据手段建立的分类模型,能快速地检测出社交平台上的原始数据中网络欺凌语言的存在.最后,我们对含有网络欺凌语言的评论与用户等级、发表时间等属性上的相关性做了分析,并拟合出高斯分布模型.

In order to effectively detect the cyberbullying language on Chinese social platforms,a dozen typical Chinese social platforms are selected,and some samples are extracted from them for manual annotation to construct a training data set.On the basis of the training set,three types of classifiers,i.e.Naive Bayes,support vector machine and long-short-term memory neural network,are used to construct a classification model to classify and recognize unlabeled data.Experiments show that the above selected classifiers can effectively identify cyberbullying language with an accuracy rate of 0.87,0.79 and 0.88,respectively.Of the three classifiers,the long-short-term memory neural network has the best effect.It is concluded that the classification model established with the help of big data can quickly detect the original data on social platforms and detect the existence of cyberbullying language.Finally,this paper also analyzes the correlation between offensive comment language and user rank,publication time and other attributes,and fits a Gaussian distribution model.