Clustering Social Media Users Using Categorical-Valued Functional Data Analysis
提出一种基于分类函数数据的方法，根据用户发帖行为的相似性对Twitter用户进行聚类，帮助公司识别有潜力积极推广品牌的消费者。
Social media provides more insight into consumer behavior than companies have ever had, and firms can interact with consumers on social media to increase their brand loyalty and address concerns they might have. However, it is critical for companies to evaluate whether the consumers they interact with have the potential to positively promote the firm. This can be challenging, especially when limited information is available about social media users. Our work proposes a flexible methodology to cluster many Twitter users based on the similarity in their posting behavior to solve this problem. We provide a framework that views users’ high-frequency postings during a specified timeframe as densely-observed categorical functional data, and propose to cluster them using latent user-specific characteristics. This leads to an interpretable and computationally-efficient algorithm and enables us to gain insights into the posting behavior of social media users. While our methods are inspired by a Twitter application they can be applied to understand posting behavior across various social media platforms. Finite-sample properties of the methods are investigated through simulations. This method is implemented in the function catfdcluster() in the R package catfda.