复杂情境下概率生成模型与深度学习融合的POI智能推荐方法

POI推荐为满足基于位置的社交网络中用户个性化服务需求、减轻信息过载起到关键作用。针对现有POI推荐系统面对数据稀疏、信息多源异构、用户兴趣多样及真实场景等复杂情境下难以高效应用的关键难题，本项目将POI推荐转换为用户移动行为时空序列预测问题，研究复杂情境下概率生成模型与深度学习融合的POI智能推荐方法。主要内容包括：（1）顾及时空特征和活动目的的个性化用户画像构建，精细刻画用户特征；（2）融合语义信息的POI知识图谱表示学习，准确描述POI之间的多维关联关系；（3）基于分层注意力机制和概率生成模型的下一个POI预测，实现多特征融合智能推荐；（4）POI智能推荐原型系统研发与实验分析，验证模型和方法的有效性。本项目旨在提高POI推荐的精准度、用户满意度，为构建POI智能推荐系统提供基础理论与方法，促进“互联网+POI服务”的大众化普及应用。

POI recommendation is of great significance for satisfying the users' personalized service needs and mitigating information overload in location-based social networks. However, under complicated scenarios with data sparsity, multisource heterogeneity, diversity of user interest, the current POI recommendation system is not able to be applied efficiently. This study aims at transforming POI recommendation into spatio-temporal sequence prediction of user mobile behaviors to propose a POI intelligent recommendation method integrating probability generative model and deep learning with complex context information. The study primarily includes: (1) Constructing personalized user profile considering spatio-temporal features and activity purpose of users to precisely portray users characteristics; (2) Analyzing the multi-dimensional relations between POIs with representation learning of POI knowledge graph with semantic information; (3) Predicting next POI based on hierarchical attention model and probability generative model to achieve intelligent recommendation with multi-feature fusion; (4) Developing prototype system for POI intelligent recommendation and making experimental analysis to evaluate the performance of the model and method. This study is crucial to improve the precision and user satisfaction of POI recommendation, provide basic theory and methods for the construction of POI intelligent recommendation system, thus eventually promote popularization and application of Internet Plus POI services.