基于跨社交媒体网络行为大数据的用户建模关键技术研究

The user modeling of social media networks is an important manifestation of the analysis and application of multi-source big data social media. According to challenges of the high sparsity, massive dynamic, heterogeneous and complex intentions of the user behavior in social media networks, a collaborative analysis is proposed by integrating a large number of common users' multimedia behavior information on different social media networks, which helps us to analyze and understand users' actual information needs, and to solve the problem of user intention gap in knowledge push in the era of big data. On the one hand, we propose a deep learning framework to learn a unified representation for multimedia resources in the social network, so as to solve the problem of expressing the heterogeneous data in a unified knowledge space; on the other hand, we propose to employ the overlapped users cross platforms as a bridge for mining and integrating the static and dynamic association between the multi-source heterogeneous multimedia behavior data and user attributes, based on the cross platform correlation and the context correlation property of user multimedia behavior data in social media network. In this way, we can extract the valuable information from the fragment data scattered in the different network platform to construct the user portrait model, so as to solve the problem of sparsity and single content in the single platform.

社交媒体网络的用户建模是分析与应用多源社交媒体大数据的重要表现。针对社交媒体网络用户行为分析与建模面临的行为数据高稀疏度、海量动态、多元异构和意图复杂等挑战，本项目通过整合大量共同用户在不同社交媒体网络上的多媒体行为信息进行协同分析，深入准确理解用户的信息需求，解决大数据时代知识精准推送中的用户"意图鸿沟"问题，为大规模社交媒体应用提供实际解决方案。一方面，本项目利用深度学习的框架将用户在社交网络上提供的多媒体资源学习出一个统一的表示，从而解决异构数据在同一知识空间的表达问题；另外一方面，利用跨社交媒体用户多媒体行为数据的上下文相关性与跨平台相关性，以用户为桥梁对多源、跨平台用户异构数据与用户属性进行动静态关联挖掘与整合，从这些分散于不同网络平台的碎片化数据中提取有价值的用户画像模型信息，有效解决单一网络的行为稀疏性和内容单一性问题。

本项目研究目标为针对社交媒体网络用户行为分析与建模面临的行为数据高稀疏度、海量动态、多元异构和意图复杂等挑战，利用跨社交媒体用户多媒体行为数据的上下文相关性与跨平台相关性，以用户为桥梁对多源、跨平台用户异构数据进行关联挖掘与整合，从这些分散于不同网络平台的碎片化数据中提取有价值的信息，全方位地分析和理解用户的信息需求，解决大数据时代知识精准推送中的用户“意图鸿沟”问题。经过课题组全体成员四年的努力,达到了课题的预定目标和技术指标。课题组提出了一种基于互编码器的跨社交媒体网络用户多媒体数据一致性表达模型解决多媒体数据的统一表示问题；提出隐性支持向量机模型对隐变量进行推断学习的属性之间以及用户行为与平台个性化人口属性关联发现；提出主题情感模型来挖掘旅游文本信息中的潜在主题和情感以及二者的耦合关系的用户兴趣建模方法。课题执行期间完成任务书中预期成果指标。