Abstract
Popularity of publications, such as CDs, books, and movies, is critical to circulations and incomes. However, an erroneous prediction of popularity of publications causes unnecessary costs, or lost due to underproduction. Hence, the analysis of popularity of products has become an important issue. Our purpose in this research was to detect the trend before a publication becomes popular. Generally, the time series of popularity of a product can be divided into three phases – the slow-start phase, the fast-growing phase, and the slow-end phase. We proposed a two-stages detecting algorithm, which monitored the popularity with a CUSUM mechanism, verified the monitoring by comparing the distributions of past and future data to find the outbreak point, predicted the future trend of popularity, and then detected the time that the growth slows down. Thus, the data series was divided into the three mentioned phases. Through some simulation results with real data, the rate of accuracy on detecting outbreak points was over 90%, and over 80% on detecting cool-down points. This exhibits that the proposed algorithm improves efficiency and accuracy while predicting popularity of publications.