Abstract: With the rapid prevalence and development of online TV series (or teleplays), there is a novel research problem, predicting the popularity of online teleplays. The continuity of teleplay plots makes the popularity of adjacent episodes have a strong correlation. This paper extends the classical autoregressive moving average (ARMA) model. Specifically, this paper considers two modeling strategies, namely multiple episodes and single day, and multiple episodes and multiple days. Both of them use the sharing parameter method to estimate the model parameters. This paper applies the root mean squared error (RMSE) as the evaluation measure, many experiments on a real-world dataset show that the above two strategies can reduce RMSE by 22.0% and 32.3% respectively.

Key words: autoregressive moving average model, popularity prediction, online teleplays, time series, sharing parameters

摘要： 在线电视剧的迅速普及和发展，引发了一个全新的研究问题，即在线电视剧流行度预测。电视剧情节演化的连续性，使相邻剧集的流行度序列具有很强的线性相关性。扩展了自回归滑动平均（autoregressive moving average，ARMA）模型。具体地，采用多集单天和多集多天两种不同的建模策略，使用电视剧之间共享参数方法进行模型参数估计。利用均方根误差（root mean squared error，RMSE）评价预测方法的准确性，在大量的真实数据集上的实验表明，上述两种策略相比于对比方法，可以使RMSE平均分别降低22.0%和32.3%。

关键词: 自回归滑动平均模型, 流行度预测, 在线电视剧, 时间序列, 共享参数