面向微博的用户动态兴趣建模和智能推荐方法研究

In recent years, with the rapid development of mobile Internet and intelligent terminal equipment, social media applications like microblogging platform have produced a large amount of user information, which results in serious information overload. How to achieve better performance of information recommendation has become a research hotspot. In this research, based on the multi-level learning of social media perception data on the microblogging platform, we will study users' dynamic interest modeling and their social network relationships to perform the intelligent recommendation for microblog short texts. Firstly, based on the topic recognition, sentiment classification and the dependence of the microblog texts over time, the topic-emotion-time joint model is constructed to infer users' dynamic interests; Secondly, the social network analysis technology is used to carry out multi-level mining of the user's social network relationships, and the expression and measurement of the trust relationships between the users are constructed; On this basis, the user interests and social network relationships are integrated to achieve the intelligent recommendation of microblog texts; Finally, the evaluation metrics and evaluation system for recommendation results are designed, and the effectiveness of the proposed models and algorithms are examined and verified by comparing with the real microblog data. We further study the optimization of the application of multi-source data fusion in microblog recommendation. This research will provide a new way of thinking and technical reference for predicting users' potential interests in different contexts and preforming intelligent recommendation.

近年来，随着移动互联网和智能终端设备的迅速发展，以微博为代表的社交媒体应用产生了海量的用户信息，并导致了严重的信息过载。如何实现高质量的信息推荐已成为研究的热点。本课题将针对微博平台上社会化媒体感知数据的多层次学习，研究用户动态兴趣建模和社交网络关系以实现微博文本的智能推荐方法。首先，基于微博文本的主题识别、情感分类和时间依赖，构建主题-情感-时间联合模型，以实现用户多维动态兴趣模型；其次，采用社会网络分析技术对用户社交网络关系进行多层次挖掘，构建融合关注网络和交互网络的用户间信任关系的表达和度量；在此基础上，融合用户兴趣和社交网络关系来实现微博文本的智能推荐；最后，设计推荐结果的评价指标和评价体系，结合实际微博数据，对所提出的模型与算法的有效性进行验证，并进一步研究多源数据融合在微博文本推荐中的优化应用。本课题的研究将为预测不同情境下的用户潜在兴趣和实现智能推荐提供新思路与技术参考。

本项目研究面向微博评论文本的用户动态兴趣建模和智能推荐方法，致力于建立有效的微博用户兴趣模型、社交网络关系模型，并融合用户兴趣和社交网络关系来构建有效的微博信息推荐方法，从而实现社会媒体平台上信息的检索与推荐。首先研究微博用户兴趣模型的构建，主要基于微博的用户注册、社交关系和行为数据构建多层分类器以实现用户画像预测和动态兴趣建模。其次，研究微博用户社交网络关系模型的构建，采用社会网络分析技术对用户社交网络关系进行多层次挖掘，构建融合关注网络和交互网络的用户间信任关系的表达和度量；再次，融合用户兴趣和社交网络关系来构建有效的微博信息推荐方法，并进行了实验验证。最后，基于社会媒体平台的用户评论文本研究了融合注意力机制和深度学习网络的智能推荐模型及新兴场景下的推荐模型，并对推荐结果的可解释性进行了初步探讨。研究成果包括发表项目相关论文21篇，其中期刊：CCF A类2篇，CCF B类/SCI二区论文8篇；会议：CCF A类2篇，CCF B类5篇，CCF C类3篇，国内会议论文1篇。此外，撰写学术专著中的一章，申请专利已进入实审4项。