基于多主题和网络模型的社交媒体电子医疗用户推荐研究

In the scenarios of big data in social media, it has new characteristics that are multiple topics and complicate network structure in the field of e-healthcare. How effectively combining the social tag and social network to give users precise recommendations is a key issue. It decides if users can grab correct medical information and make reasonable treatment decision. This project combines the topic model of social tag and social network model to study e-healthcare recommendation in social media. Firstly, this project chooses Weibo.com as research bed. We collect personal information to build text set of social tag in the topics of diabetes and its related diseases. Secondly, we use latent dirichlet allocation (LDA) model to extract topics, and improve the parameters of that model to the relationship of user-tag-topic. We use the clustering algorithm to gain the multiple topic-based clustering communities. Thirdly, we build and optimize the exponential random graph model (ERGM) according to the follow-relationship social network of e-healthcare users. Based on this model, we design a link prediction algorithm to compute the node link probability of recommendation list. Finally, this project combines the multiple topics analysis of users’ tags and link probability prediction of network to deeply filter network recommendation list. It will give precise recommendation results in line with user psychological preferences. The recommendation method proposed in this project is evaluated by recommender systems metrics. The innovative research results of this project can provide theoretical and methodological support of e-healthcare network recommendation for chronic disease community. It also provides a scientific evidence for e-healthcare patients to make optimal medical decisions and to serve information of preventing complicated diseases.

社交媒体平台的大数据环境下，在电子医疗领域具有多主题和复杂网络结构新特征。如何有效结合社会标签和社会网络这两种因素给出用户精准推荐，是关系患者能否获取正确医疗信息和做出合理医疗决策的关键问题。本项目将社会标签的主题模型与社会网络模型结合，研究社交媒体上的电子医疗推荐。首先，本项目以新浪微博为研究平台，收集糖尿病以及相关疾病主题的用户个人信息构建社会标签文档集合。其次，利用LDA模型方法提取主题，并改进模型参数形成用户-标签-主题关联关系，利用聚类算法得到基于多主题的用户聚类社区。最后，根据电子医疗用户的关注关系形成的社会网络，构建和优化指数随机图模型。在此模型基础上设计连接预测算法，得到网络节点连接概率的推荐列表。本项目的创新研究成果可以为慢性病群体的电子医疗网络推荐提供理论和方法支持，并为电子医疗患者做出最优医疗决策及预防相关性疾病等信息服务提供科学依据。

如何有效结合社会标签和社会网络这两种因素给出用户精准推荐，是关系患者能否获取正确医疗信息和做出合理医疗决策的关键问题。本项目将社会标签的主题模型与社会网络模型结合，研究社交媒体上的电子医疗推荐。首先，本项目以新浪微博为研究平台，收集糖尿病以及相关疾病主题的用户个人信息构建社会标签文档集合。其次，利用LDA模型方法提取主题，并改进模型参数形成用户-标签-主题关联关系，利用聚类算法得到基于多主题的用户聚类社区。最后，根据电子医疗用户的关注关系形成的社会网络，构建和优化指数随机图模型。在此模型基础上设计连接预测算法，得到网络节点连接概率的推荐列表。本项目的创新研究成果可以为慢性病群体的电子医疗网络推荐提供理论和方法支持，并为电子医疗患者做出最优医疗决策及预防相关性疾病等信息服务提供科学依据。