社交网络虚拟用户属性推测关键技术研究与实现

发布时间：2018-04-28 21:34

本文选题：社交网络 + 虚拟用户　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：社交网络平台作为一种用户交流和分享信息的虚拟平台,蕴含了大量的用户信息,其中用户属性信息在个性化推荐、精准营销以及舆情引导等方面发挥着重要作用。社交网络平台无法准确获取用户属性,就不能为用户提供优质服务。因此,研究用户属性对社交网络平台具有重要作用。该领域不同研究者对虚拟用户属性定义不一致,导致在研究属性推测问题时,存在结构混乱(父子级关系不明确)、推测方法适用性受限、研究成果可拓展性差等问题。针对这些问题,本文对虚拟用户属性定义以及推测方法进行了深入研究。首先,针对虚拟用户属性定义不一致的问题,研究了虚拟用户属性表述模型。通过分析社交网络用户数据,刻画了描述用户的四个属性维度。同时,针对属性缺失及不准确的问题,提出了一种基于模型分类以及邻居-社团联合更新的虚拟用户属性计算模型。该模型能针对不同基本属性特点,控制更新操作的有无,在节约时间成本的同时,提高了属性推测的准确率。其次,将虚拟用户属性计算模型应用于具体基本属性推测任务。针对属性聚集性强弱的差异,进行了性别和职业两种属性的研究。在性别推测中,针对性分析了用户使用习惯特征,提出了特征选择及基于字典的特征权重计算方法,并引入了一种基于朴素贝叶斯融合的分类算法,弥补了单纯特征叠加效果不佳的问题,提升了分类效果;在职业推测中,改进了现有的主题式特征选择方法,设计了基于更新机制的分类算法,并测试分析了算法中的影响因子。实验证明,该算法在应用于职业推测时,能有效提升分类效果。最后,基于上述研究设计了虚拟用户属性推测原型系统。该系统通过分析待测用户发布内容、链接关系数据,获取自身以及邻居-社团传播的信号,最终推测用户的属性类别,从而用于充实用户属性表述模型,达到提升个性化推荐服务质量的目的。
[Abstract]:As a virtual platform for users to exchange and share information, social network platform contains a lot of user information, among which user attribute information plays an important role in personalized recommendation, accurate marketing and public opinion guidance. Social network platform can not accurately obtain user attributes, can not provide quality services for users. Therefore, the study of user attributes plays an important role in the social network platform. Different researchers in this field have different definitions of virtual user attributes, which lead to the confusion of structure (the relationship between father and son is unclear, the applicability of speculation method is limited, and the scalability of research results is poor). In order to solve these problems, the definition of virtual user attributes and the methods of conjecture are studied in this paper. Firstly, aiming at the problem of inconsistent definition of virtual user attributes, a virtual user attribute representation model is studied. By analyzing the data of social network users, the four attribute dimensions of describing users are described. At the same time, a virtual user attribute computing model based on model classification and joint update of neighborhood and community is proposed to solve the problem of missing and inaccurate attributes. According to the characteristics of different basic attributes, the model can control the availability of update operation and improve the accuracy of attribute speculation while saving time and cost. Secondly, the virtual user attribute calculation model is applied to the specific basic attribute estimation task. Aiming at the difference of attribute aggregation, gender and occupation are studied. In gender speculation, the characteristics of users' usage habits are analyzed, and a method of feature selection and feature weight calculation based on dictionary is proposed, and a classification algorithm based on naive Bayes fusion is introduced. It makes up for the problem of poor superposition effect of pure features and improves the classification effect. In career speculation, it improves the existing thematic feature selection methods, and designs a classification algorithm based on updating mechanism. The influence factors of the algorithm are tested and analyzed. Experiments show that the algorithm can effectively improve the classification effect when it is applied to occupational speculation. Finally, a virtual user attribute inference prototype system is designed based on the above research. By analyzing the content of the users to be tested, linking the relational data, acquiring the signals propagated by themselves and the neighbors and communities, the system finally inferred the user's attribute category, which was used to enrich the user attribute representation model. To improve the quality of personalized recommendation service.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09

【参考文献】