基于改进K-近邻算法的电视剧点播量预测方法

摘要

及时、准确地预测电视剧点播量为商业决策提供很大帮助.传统时间序列预测需要大量历史数据,很难满足及时、准确的预测需求.提出一种基于改进K-近邻算法的电视剧点播量预测方法,改进了K-近邻模型,并融入缩放技术和相关系数,结合百度搜索数据和点播量序列的相关性,以前一周每天的点播量为特征,预测电视剧后一天的点播量.在PPTV和优酷数据集上进行实验,比用K-近邻的方法在MAE和MAPE上分别提高了75.5％、95.3％和71.8％、99.3％.%Timely and accurate prediction of TV on demand provides a great help for commercial decision.Traditional time series prediction requires a lot of historical data, and it is difficult to meet the timely and accurate prediction needs.In this paper, an improved KNN algorithm is proposed to improve the prediction of TV on demand.The KNN model is improved, and the scaling technology and correlation index are integrated.Combining the correlation between Baidu search data and the demand quantity sequence, it is characterized by daily demand of the previous week to predict the day after the TV drama demand.Experiments on the PPTV and Youku data sets show an increase of 75.5％, 95.3％,71.8％ and 99.3％ on the MAE and MAPE, respectively, compared with the KNN algorithm.