基于小波多尺度的网络论坛话题热度趋势预测.pdfVIP

19 4 V l. 19 N . 4 2009 4 COM PUT ER TECHNOLOGY AND DEVELOPMENT Apr. 2009 1 2 1 张 虹 , 赵 兵, 钟 华 ( 1. 北京城市学院人工智能研究所, 北京 100083; 2. 中国电力科学研究院, 北京 100085) : ( ) , ; ARIMA ; ; , : , : ; ; ARIMA ; : T P3 11 : A : 1673- 629X( 2009) 04- 0076- 04 Hot Trend Prediction of Network Forum Topic Based on Wavelet Multi- Resolution Analysis 1 2 1 ZHANG H ng , ZHAO Bing , ZHONG Hua ( 1. Artificial Intelligence Institute, Beijing City University, Beijing 100083, China; 2. China Electric P wer Research Institute, Beijing 100085, China) Abstract In this paper, a meth d w as pr p sed, w hich c mbined w avelet multi- res luti n analysis t f recast netw rk f rum t pic h t trend . That w as first, using wavelet multi- res luti n t dec mp se and rec nstruct the riginal time series f rmed fr m click numbers r reply numbers f f rum p st, t create ne l w frequency signal an d several high frequency signals. Th en, the appr ximate l w frequency signal is predictedusing ARIMA m del and the high frequency signals are f recasted respectively using neural netw rks that have different parameters. After ne- step- ahead predicti n, the predicted results f these signals are c mbined int the final predicted result f net- w rk f rum t pic h t trend. This pr p sed meth d w as tested t make better predicti n accuracy f r netw rk f rum t pic h t trend. Key words time series; w avelet dec mp siti n and rec nstructi n; ARIMA m del; neural netw rk 0 , Internet ,