面向网络舆情的文本知识发现算法对比研究

Comparative research on text knowledge discovery for network public opinion

针对网络舆情分析领域,研究了系统聚类、String Kernels、K最近邻算法(K-nearest neighbor,KNN)、SVM(support vector machine)算法以及主题模型5种聚类算法。以网络舆情数据为对象集,以R语言环境为实验工具,比较了这5种算法的优势与劣势,同时进行了仿真实验。实验结果表明,主题模型相对于其他算法在文本聚类方面具有更好的适用性,其中,主题模型中的CTM(correlated topic model)方法更适合于类别关系的探索与发现,而Gibbs抽样方法则在文本聚类上的表现优于CTM方法。

According to the field of network public opinion analysis,five clustering algorithms：system clustering, string kernels,K nearest neighbor algorithm,support vector machine algorithm and topic models were studied.A com-prehensive comparative research of these five algorithms was conducted by using network public opinion data as data set and R language environment as experimental tool.At the same time,simulation experiments were carried out to com-pare these five algorithms’strengths and weaknesses.Experimental results show that＂topic model＂has better applica-bility than other algorithms in terms of text clustering.After further experiments we also found in topic models,CTM （Correlated Topic Model）method is more suitable for the exploration and discovery of class relations while Gibbs sam-pling method on the performance of text clustering method is better than the CTM method.