基于时间序列分析的微博突发话题检测方法

针对微博信息噪音大、新颖度难以判断的问题，在动量模型的基础上进行优化，提出了基于时序分析的微博突发话题检测方法。通过动量模型提取候选突发特征后，对特征的动量时间序列分别借鉴信号频域分析理论和股票趋势分析理论进行建模，分析特征的频域特性来识别频繁伪突发特征，分析特征的新颖程度来识别间歇性伪突发特征，合并过滤后的有效突发特征形成突发话题。微博数据实验表明，该方法有效提高了突发话题检测的准确率和F值。

Bursty topic detection method for microblog based on time series analysis

Detecting bursty topics from microblogs was an important task to understand the current events attracting a large number of internet users.However,the existing hods suitable for news articles cannot be adopted directly for microblogs.Because microblogs have unique characteristics compared wi formal texts,including diversity,dynamic and noise.A detection method for microblog bursty topic was proposed based on time series analysis,which was an op-timization method of momentum model.The candidate bursty features were extracted by momentum model.The time se-ries of feature's momentum were modled by frequency domain analysis theory and stock trend analysis theory.The fre-quently pseudo-bursty features were filtered according to analysis results of frequency-domain characteristics.The inter-mittently pseudo-bursty features were filtered according to the novelty analysis result through stock trend theory.The bursty topics were finally emerged with combination of effective bursty features.The experiments are conducted on a real Sina microblog data set.It show that the proposed method improves the precis and F-measure remarkably compared with the momentum modle.