面向多源数据细粒度情感挖掘的旅游需求预测

Tourism demand forecasting based on multi-source fine-grained sentiment mining

数字经济时代,消费者基于互联网社交媒体平台产生的搜索与评论等内容拓展了旅游需求预测的数据来源.虽然已有研究聚焦消费者对旅游目的地整体情感倾向的挖掘,但是少有研究考虑消费者对餐饮、住宿、交通和服务等细粒度层面的差异化评价对需求预测的影响.本文以九寨沟景区客流量预测为例,采用基于机器学习的细粒度情感分析方法对携程、去哪儿网、大众点评和美团等多源数据进行文本分析,构建涵盖八个维度的细粒度情感指数,对客流量开展一步和多步预测,并与包含搜索引擎指数及整体情感指数的模型进行对比.结果表明:在基于时间序列、机器学习和深度学习三种不同类型的预测模型中,包含细粒度情感指数的模型均能显著提高旅游需求预测的准确性.在样本外预测中,所构建的包含细粒度情感指数的模型较包含搜索引擎指数和整体情感指数的模型,预测精度平均提升17.78%和6.53%.本研究为数字经济时代多维数据驱动的旅游需求预测提供创新研究方法.

In the digital economy era,the content generated by consumers based on internet social media platforms,such as searches and reviews,has expanded the data sources for tourism demand prediction.Existing research focuses on mining and analyzing consumers’overall emotional tendencies towards tourist destinations,but there is little consideration of the impact of their differentiated evaluations on fine-grained levels such as dining,accommodation,transportation,and services on tourism demand prediction.Taking Jiuzhaigou scenic area’s visitor flow prediction as an example,this paper uses machine learning-based fine-grained sentiment analysis methods to analyze multi-source data from Ctrip,Qunar,Dianping,and Meituan,constructing a fine-grained sentiment index covering eight dimensions to predict the scenic area’s visitor flow.This is compared with models containing search engine indices and overall sentiment indices.The results show that,in three different types of prediction models based on time series,machine learning,and deep learning,the models containing the fine-grained sentiment index can significantly improve the accuracy of tourism demand prediction.In the out-of-sample forecasting,the model incorporating a fine-grained sentiment index demonstrates an average prediction accuracy improvement of 17.78%and 6.53%compared to models that include search engine indices and a general sentiment index.This study provides innovative research methods for multi-dimensional data-driven tourism demand prediction in the era of big data.