融合网络环境下情景感知的个性化推荐机制

The project aims at the characteristics of user’s information demands and behaviors under multi-network fusion environments, and focuses on the new needs of the personalized recommendation. We will make a deep research on the mechanism and key technology of the context-aware personalized recommendation for multi-network fusion environments. In view of the multimodality, heterogeneity, high dimension, and dynamics of big user data, based on the tensor model, we propose a multi-view modeling method of user context-interest and context-behavior profile to implement the multimodal user data mining. Based on the correlation and interactivity in social network, we propose a user impact relation model and a quantitative method by using user interaction modeling and sentiment analysis. Based on both the spatio-temporal statistical characteristics of user behaviors and the correlated patterns between the context and behaviors of user, we propose a context-aware prediction method of user behaviors by using semi-supervised learning and uncertainty reasoning. Due to the user behavior features such as the randomness, fragmentation, and context-dependence, on the basis of the potential correlation among user's attributes, behavior intention, social relations, scene, context, and interests, we propose a context-aware recommendation algorithm and a ranking optimization method by integrating multi-factors. Finally, we will use mobile social applications to fulfill the theoretical proving, and then provide reference for context-ware personalized recommendation mechanism for multi-network fusion environments.

本项目针对融合网络环境下用户信息需求和行为特征，围绕个性化推荐新需求，研究融合网络环境下情景感知的个性化推荐机制和关键技术。针对用户大数据的多模异构性和高维动态性，提出了基于张量空间下异构多视角的用户情景兴趣和情景行为画像建模方法，实现对多模态的动态用户数据挖掘。基于社会网络的关联性和交互性，提出了一种基于用户交互建模和情感倾向分析的用户影响关系模型和定量化方法。基于用户行为的时空统计特性以及情景与行为的关联性，提出了一种基于半监督学习情景感知和不确定性推理的用户实时行为意图预测方法。针对融合网络环境下用户行为的随机性、碎片化和情景依赖性等特征，基于用户的属性特征、行为趋向、情景、社会关系与用户兴趣的内在关联机制，提出了多因素融合的情景感知的个性化推荐方法和排序优化方法。最后通过移动社交应用进行理论验证，为融合网络环境下个性化推荐机制研究提供参考。

本项目针对融合网络环境下用户信息需求和行为特征，围绕个性化推荐新需求，研究融合网络环境下情景感知的个性化推荐机制和关键技术。项目组主要围绕融合网络环境下的用户情景兴趣和情景行为画像建模方法，用户之间的影响关系模型和定量化方法，用户实时行为意图预测理论和方法，个性化推荐方法与排序优化理论和方法展开研究与应用验证。我们分别提出了支持实时可靠的流数据处理的边缘智能框架以用于用户实时情景识别，基于隐知识学习的用户活动实时识别和实时预测模型和方法，个人化用户和项目关系网络创建方法以及融合矩阵分解和邻居模型的个性化推荐模型和方法。.我们对提出的方法和模型进行了大量实验验证，取得了较好的实验结果。相关的重要理论成果发表在国内外高质量期刊和会议上（其中两篇发表在CCF A类期刊 IEEE Transactions on Knowledge and Data Engineering）。在科技成果转化方面，我们做了初步尝试，部分成果在2017 年浙江省秋季科技成果竞价（拍卖）会，拍出60万元；开发的原型系统获得杭州市经济开发区人才创业项目立项支持。项目组在人才培养上取得较好成果，负责人和主要成员在职称和学位上获得晋升。