基于时间序列聚类和ARMA模型的检索量预测

Prediction of Search Data Volume Based on Time-Series Clustering and ARMA Models

为了通过预测分析检索量数据来指导商家调整产品开发及经营策略,将检索量数据组织为时间序列,对其用自回归滑动平均(ARMA)模型进行建模预测.先将时间序列进行聚类,仅对聚类中心序列进行ARMA模型识别,同类序列用该模型进行近似建模预测;经过数据预处理、相似性分析、基于相似度的聚类、时间序列预测等过程,得到检索量数据的预测值,并将其与检索量的实际值做比较.结果表明,用同一个ARMA模型拟合相似时间序列的方法具有可行性,且有较高的预测准确率.从聚类结果还可看出,同品牌产品的检索量数据趋于聚成一类,这为检索词关系的挖掘提供了参考.

In order to guide the adjustment of product development and business strategy by predicting and analyzing the search data volume,the data of search volume are organized into time series that is modeled and predicted using the autoregressive moving average（ARMA） models.Then,the set of time series is modeled by clustering;the cluster centers are modeled using ARMA models;and the same-class series is fitted with the models approximately to obtain the predicted values.Moreover,after such operations as data preprocessing,similarity analysis,similarity-based clustering and time-series prediction,the search data volume is predicted and is compared with the actual one.Experimental results show that it is feasible and accurate to model similar time series with the same ARMA model.In addition,clustering results indicate that the search data volume of the products with the same brand tends to be clustered together,which provides a reference for the relationship mining of search terms.