基于大数据和多维张量分解的在线旅游个性化推荐模型研究

互联网背景下，在线旅游平台累积了海量旅游信息数据，为提升信息价值，使游客快速从“过载”信息中获取所需，个性化旅游推荐研究显得十分迫切。旅游消费受多种因素影响，而传统的旅游推荐模型局限于分析用户与产品间的二维关系，处理多维信息时会使数据的结构信息丢失。张量具有处理多维数据的天然优势，基于张量分解的推荐模型既可增加信息表达的维度、又可填补未知元素，从而帮助解决传统推荐模型面临的数据稀疏和冷启动的问题。基于此，本项目拟运用数据挖掘、机器学习等方法对在线旅游消费的时间情景、地域特征和群组偏好进行深度挖掘，通过张量分解模型和迭代算法，构建基于“用户-旅游产品-时间情景-地域特征”时空多维张量分解的旅游推荐模型，并结合个体与群组的消费偏好进行个性化推荐。最后，进行数据实验及仿真分析，验证推荐模型的实用性和有效性。项目研究成果将丰富和完善旅游推荐理论体系，为在线旅游企业个性化推荐提供理论指导与方法支撑。

With the rapid development of online tourism, in order to enhance the value of information, the research on personalized recommendation model of online tourism agency is very urgent. However, traditional recommendation models only consider two-dimensional relationship between tourist and tourism product, which has a serious problem of cold start. And traditional recommendation models often ignore many factors affecting tourism preference, such as travel date, tourism product category, tourist living region and tourist cluster. To overcome these problems, tensor decomposition method will be presented, which can fill in the unknown elements and formulate the behavioripreference of the recommended objects in a multi-dimensional and multi-level way. So, based on data mining and machine learning, this project will first analyze the relationship between tourist and the tourism product, travel date context, tourist living region and tourist cluster. Secondly, a four-dimensional tensor will be presented, composed by tourist, production category, travel date context and living region. And another four-dimensional tensor including tourist cluster and tourism production cluster will also be constructed. Then, two tensor decomposition recommendation models are proposed based on tensor decomposition models and iteration heurist algorithms. These personalized tourism product recommendation models can help improving the recommendation accuracy of traditional models in the case of sparse user behavior data. Finally, the numerical experiments and comparative simulation will be conducted based on the practical case data to verify the practicability and effectiveness of the model and algorithm. To sum up, the results of this project will enrich and improve the existing theoretical system of tourist behavior analysis and recommendation model, and provide theoretical guidance and method support for online tourism enterprises to recommend personalized tourism products.