泛在网络环境下用户兴趣建模与移动推荐方法研究

Under the ubiquitous network environment, traditional recommendation is difficult to meet the requirement of personal recommendation for users in different situations, different locations, different purposes and different time. In order to enhance the quality and efficiency of personalized recommendation, this study first explores the inherent relationship and quantitative expression method between user interest drift and context scenarios. We construct a multi-dimensional user interest model which merges mobile scenarios, user preferences, behavior trajectory and mobile social network. Next, we reveal the multiple topology relation and temporal hybrid index storage structure of user interest model and recommendation project through the separating of spatial position and topological property set. Then, the mobile context is divided into inner stability preferences and outer opportunity interest. On this background, a mobile context heuristic search algorithm based on context-awareness is proposed, and an intelligent recommendation algorithm based on emotional semantic which integrates item forest classification system is also presented; We adopt the Kalman filtering method to fast filter the result sets of collection of non-stationary random process based on temporal priority. Finally, a recommendation effectiveness evaluation model is designed to analyse and verify the model and algorithm. Our research findings will contribute novel ideas and research reference to the study of user interest model and context-awareness recommendation.

传统推荐难以满足泛在网络环境下不同情境、不同位置、不同目的及不同时期的个性化推荐需求。为提升推荐质量及效率，本研究首先探寻上下文情境与用户兴趣迁移之间的内在关联与定量表达方法，构建融合移动情境、用户偏好、行为轨迹及移动社交网络的多维度用户兴趣模型；然后，通过分离空间位置与拓扑属性集，揭示用户兴趣模型与推荐项目之间的多元拓扑关联与时空混合索引存储结构；随后，将移动上下文分为内层稳定偏好与外层机遇兴趣，设计基于情境感知的移动上下文启发式搜索算法；结合商品森林分类体系设计基于情感语义的智能推荐算法；采用卡尔曼滤波方法对非平稳随机过程所汇集结果集进行时间域优先快速过滤；最后，设计推荐效用评价模型对所提模型与算法的有效性进行分析与验证。本课题的研究将为泛在网络环境下的用户兴趣建模与情境感知推荐研究提供新思路与研究参考。

传统的推荐难以满足泛在网络环境下不同情境、不同位置、不同目的及不同时期的个性化推荐需求。为提升推荐质量及效率，本研究首先探寻上下文情境与用户兴趣迁移之间的内在关联与定量表达方法，构建融合移动情境、用户偏好、行为轨迹及移动社交网络的多维度用户兴趣模型；然后，通过分离空间位置与拓扑属性集，揭示用户兴趣模型与推荐项目之间的多元拓扑关联与时空混合索引存储结构；随后，将移动上下文分为内层稳定偏好与外层机遇兴趣，设计基于情境感知的移动上下文启发式搜索算法；结合商品森林分类体系设计基于情感语义的智能推荐算法；我们采用卡尔曼滤波方法对非平稳随机过程所汇集结果集进行时间域优先快速过滤；最后，我们设计推荐效用评价模型对所提模型与算法的有效性进行分析与验证。本课题的研究将为泛在网络环境下的用户兴趣建模与情境感知推荐研究提供了新思路与研究参考。