Big Data Analytics for Program Popularity Prediction in Broadcast TV Industries

The precise and timely prediction of program popularity is of great value for content providers, advertisers, and broadcast TV operators. This information can be beneficial for operators in TV program purchasing decisions and can help advertisers formulate reasonable advertisement investment plans. Moreover, in terms of technical matters, a precise program popularity prediction method can optimize the whole broadcasting system, such as the content delivery network strategy and cache strategy. Several prediction models have been proposed based on video-on-demand (VOD) data from YouKu, YouTube, and Twitter. However, existing prediction methods usually require a large quantity of samples and long training time, and the prediction accuracy is poor for programs that experience a high peak or sharp decrease in popularity. This paper presents our improved prediction approach based on trend detection. First, a dynamic time warping-distance-based K-medoids algorithm is applied to group programs' popularity evolution into four trends. Then, four trend-specific prediction models are built separately using random forests regression. According to the features extracted from an electronic program guide and early viewing records, newly published programs are classified into the four trends by a gradient boosting decision tree. Finally, by combining forecasting values from the trend-specific models and the classification probability, our proposed approach achieves better prediction results. The experimental results on a massive set of real VOD data from the Jiangsu Broadcasting Corporation show that, compared with the existing prediction models, the prediction accuracy is increased by more than 20%, and the forecasting period is effectively shortened.