节目收视预测对提高用户体验起到越来越重要的作用，而针对现有收视预测往往仅考虑用户收视行为忽略了用户评论情感因素，以及要求数据量丰富、易受“奇异点”影响、存在过拟合、欠拟合、参数设置困难等问题。本文提出基于粒子群优化算法的混合核最小二乘支持向量机模型，综合考虑了用户收视行为、评论情感两类因素，并结合时间序列及最小二乘支持向量机模型在预测上的优势对节目收视进行预测。本文采用自适应迭代预测方式，以20天为滑动窗口步长，对用户收视序列进行拟合训练，验证了该模型在收视预测上的有效性及适用性。

Program rating prediction plays an increasingly important role in improving the user experience. However, existing rating prediction only considers user rating behavior and ignores the emotional factors from user comments. Moreover, it requires abundant data, is susceptible to the influence of “singularity”, and has problems such as over-fitting, under-fitting and difficult parameter setting. In this paper, a hybrid kernel least squares support vector machine model based on particle swarm optimization algorithm is proposed, which comprehensively considers the two factors of user viewing behavior and comment emotion, and combines the advantages of time series and least-squares support vector machine model in forecasting to predict program viewing. In this paper, the adaptive iterative forecasting method is adopted, and the 20-day sliding window step is used to fit the user viewing sequence, which verifies the effectiveness and applicability of the model in viewing forecast.