考虑竞争的电影首映日票房集成预测模型研究

电影首映日票房预测对该日排片、后续放映日票房及总票房有显著影响。在构建考虑竞争的电影首映日票房预测变量集的基础上，建立首映日票房集成预测模型。首先使用多元线性回归(multiple linear regression, MLR)、支持向量回归(support vector regression, SVR)、套索回归（Least absolute shrinkage and selection operator, Lasso）和极端梯度提升（Extreme Gradient Boosting, XGBoost）等算法建立基学习器，随后使用XGBoost算法作为原学习器构建堆栈集成预测模型，最后利用收集到的数据进行对比实验。实验证明，加入竞争变量的电影首映日票房预测变量集适用于首映日票房预测；相比单一模型，提出的集成预测模型的准确性、泛化性能和稳定性均有提升，相比较传统预测方法对首映日票房预测更准确。提出的集成预测模型有助于提升首映日票房排片的有效性。

Movie Release-day Box Office Prediction Model Considering Competition

Forecasting of the release day box office has a significant impact on film schedule of the release day, box office on the following days and total box office. On the basis of building the variable set of box office prediction on the film release, an ensemble prediction model of box office was established to provide support for the forecasting method of the cinema line. Using multiple linear regression(MLR), support vector regression(SVR), Lasso regression(Lasso), and XGBoost(Extreme Gradient Boosting) algorithm to build the base-learner, and then, heterologous ensemble prediction model using XGBoost was built based on the base learner. Experiments reveal that the box office prediction variables set contain competitive variables is more suitable for box office prediction on the release day. Compared with the single model, the proposed ensemble prediction model improves the accuracy, generalization performance and stability. Far more, ensemble model shows more accurate than the traditional method in box office prediction on the release day. The proposed model may be helpful to improve effectiveness film schedule of the release day.