生鲜产品的多变量SVR需求预测——基于在线评论的顾客感知因素提取

为探索在线评论数据中蕴含的顾客感知产品信息对生鲜产品需求量预测准确度的影响，针对生鲜产品电商平台中大量评论数据，利用网络爬虫技术和Word2vec模型建立产品特征词库，提取主要需求预测影响因素，并基于产品特征词库对评论文本分类将影响因素量化，构建多变量SVR需求预测模型，同时运用粒子群算法对SVR模型中的主要参数进行优化，在此基础上进行实证分析。结果表明:1)Word2vec模型能挖掘在线评论数据中顾客关注的产品特征，有效提取顾客感知的需求预测影响因素；2)与单变量SVR模型相比，加入评论中顾客感知因素的多变量SVR在预测产品需求量时误差更小。利用在线评论中顾客感知因素建立多变量SVR需求预测模型能有效提高生鲜产品需求量预测准确度。

Multi-variable SVR demand forecast model for fresh products: Factor extraction of customer perception based on online reviews

In order to explore the influence of customer perceived product information contained in online review data on the accuracy of fresh product demand prediction, based on the large amount of review data of a fresh product e-commerce platform, web crawler technology and Word2vec model are used to build a product feature database, extract the main factors influencing demand forecast, and quantify the influencing factors based on the product feature database to classify the review text. A multi-variable SVR demand forecast mode is then constructed and the particle swarm algorithm is adopted to optimize the main parameters of the SVR model. On this basis, an empirical analysis and verification is carried out. The results show that: 1)The Word2vec model can mine the product features that consumers pay attention to in online review data, and the influencing factors on consumer perception for demand forecasting can be extracted effectively. 2)Compared with the univariate SVR model, the multivariate SVR with customer perception factors added has smaller error in predicting product demand. In conclusion, using customer perception factors in online reviews to establish the multivariate SVR demand forecasting model can improve the accuracy of fresh product demand forecasting effectively.