基于主题挖掘的旅游运营商个性化推荐策略及其优化研究

With the rapid development of tourism industry and increasing competition faced by tour operators, personalized attraction recommendation considering both the preference of tourists and the profitability of tour operator is of important theoretical and practical significance. Firstly, we propose an extended latent dirichlet allocation model considering time context and expert knowledge for topic analysis on characteristics of tourist attractions. We also propose an algorithm for topic analysis on preferences of tourist groups taking into account trust relationship in social networks, based on which a personalized attraction recommendation strategy for tourist groups is developed. We then, based on the key cost factors analysis of tour operators' personalized attraction recommendation process and stochastic demand of tourist groups, propose a two-stage stochastic programming model that involves joint chance-constraint and a mixed integer programming model based on Graph Expander theory, respectively, aiming at different cases of single tourist origin and multiple tourist origins, and recommendation strategy optimization is conducted accordingly. Furthermore, the solution algorithms for the above two models are developed based on the sample average approximation method and the cutting-plane method, respectively, and the algorithmic performance is tested through extensive numerical experiments. Finally, comprehensive analysis on the profitability of tour operator, service level, and algorithm robustness is conducted to verify the effectiveness of the proposed personalized attraction recommendation strategy and the two optimization models. Based on it, we summarize the managerial insights for tour operators.

随着旅游业的迅猛发展和旅游运营商日益激烈的竞争，综合考虑游客偏好和旅游运营商收益水平进行个性化景点推荐具有重要的理论和现实意义。项目首先提出基于时间情境和专家意见的扩展潜在狄利克雷分布主题模型进行景点主题特征分析，同时研究考虑信任关系的游客群主题偏好，在此基础上提出针对游客群的基于主题偏好的个性化景点推荐策略。随后，针对单游客来源地和多游客来源地两种情况，分析旅游运营商开展个性化景点推荐的关键成本要素以及游客需求的随机性特征，分别提出带有机会约束的随机优化模型以及基于图扩展理论的混合整数规划模型，进行推荐策略优化研究。进一步，分别基于样本均值近似方法和切平面法提出上述两个优化模型的求解算法，并通过数值计算实验分析算法的综合性能。最后，从旅游运营商的利润水平、游客需求的满足水平以及数值计算结果的鲁棒性等方面，分析所提出的个性化景点推荐策略和优化模型的有效性以及对旅游运营商的管理启示。

随着旅游业的迅猛发展和竞争日趋激烈，综合考虑游客偏好和收益水平对游客进行个性化旅游景点推荐，对于提升旅游运营商的竞争力显然具有重要的理论和现实意义。本项目的研究主要包括三方面内容。首先，根据时间情境和专家意见提出基于潜在狄利克雷分布的扩展LDA模型，利用季节性语境信息提取景点特征；同时构建了基于相似性信任和熟悉性信任的双重信任关系模型，提出具有双重信任增强的季节感知旅游景点推荐方法。在综合考虑季节主题偏好和信任关系的基础上，提出基于游客群的推荐策略，提高了个性化推荐的水平。其次，针对单游客来源地的情况，从旅游经营者的角度，对随机游客需求下基于成本的景点推荐提出两阶段优化模型。在得到最符合游客主题偏好的基础上，考虑成本因素，建立具有联合机会约束的随机优化模型，对初步推荐方案进行优化。该优化模型能显著降低旅游运营商的推荐成本，同时保持较高的服务水平和游客满意度。最后，针对多游客来源地的情况，基于图扩展理论提出了考虑不确定游客需求下top-N推荐与运营成本之间的权衡的在线旅游推荐模型。基于图论中用于刻画欧式图连通性的扩展图理论，对游客随机需求进行转化处理，将在线旅游推荐问题构建为基于扩展器图的优化模型，并提出一个求解这一大规模非线性离散优化问题的切割平面程序，该模型实现了较高的需求满足率。本项目的研究，一方面是对推荐系统理论和方法的丰富与完善，项目虽然针对旅游景点个性化推荐进行研究，所提出的模型及方法对于其他产品或服务的个性化推荐也具有重要的借鉴意义；另一方面，本项目的研究成果对于旅游运营商提升个性化推荐水平和效益也具有重要的现实意义。