Probabilistic Graphical Models for Data Mining and Recommendation in Social Media

基于Web2.0的技术基础，社交媒体允许创建和交换用户生成的内容，并支持内容生产者和消费者之间的各种形式的交互。与报纸和电视等传统媒体相比，社交媒体的生产者群体要大得多，也更加多样化，这导致了许多简短而嘈杂的帖子，允许用户对内容发表评论和相互交流，并动态演变。社交媒体有可能向生产者提供有价值的反馈，并允许消费者利用“大众的智慧”来帮助他们做出决定。我们研究计划的长期目标是开发(1)数据挖掘方法来模拟社交媒体中集体行动的复杂动态，使人们能够更好地了解这些媒体并预测未来的事件。(2)推荐方法，针对特定用户推荐值得信赖的相关内容，并使社交媒体网站能够提高用户参与度。为了模拟社交媒体中的复杂效果，我们将探索概率图形模型，这些模型信息量很大，可以自然地整合可用的背景知识。

Social media allow the creation and exchange of user-generated content and support various forms of interactions among content producers and consumers, based on the technological foundations of Web 2.0. Compared to traditional media such as newspapers and TV, social media have a much larger and more diverse group of producers, leading to many short and noisy posts, allow users to comment on content and to communicate with each other, and evolve dynamically. Social media have the potential to provide valuable feedback to producers and to allow consumers to tap into the "wisdom of the crowds" as aid in their decision making. The long-term objectives of our research program are to develop (1) Data mining methods to model the complex dynamics of collective action in social media, enabling a better understanding of these media and the prediction of future events. (2) Recommendation methods that recommend trust-worthy and relevant content, specific to a given user, and enable social media sites to increase the level of user participation. To model the complex effects in social media, we will explore probabilistic graphical models, which are very informative and can naturally integrate available background knowledge.