With the proliferation of online social platforms, users today find themselves engaging each other on multiple social platforms. For example, users may like posts of their friends on Facebook, retweet users whom they have followed on Twitter, or share photos on Instagram. Besides engaging each other in social activities, users also leverage on online social platforms for collaborative works. For example, software engineers have used social collaborative platforms such as GitHub and Stack Overflow for software development. The study on user behaviors in multiple social platforms have many real-world applications, such as it allows us profile users more effectively and it enables us to build better recommender systems. Thus, this proposal seeks to model users across multiple social platforms. In particularly, we focus on modeling two aspects of cross platform user data, namely: (a) user relationships and (b) user topical interests.
On analyzing of user relationship in multiple social platforms, we conduct empirical study on relationships of users who have accounts in multiple social platforms. Specifically, we propose measures to analyze how user maintain and manage their friendships across multiple social platforms. Extending from the findings in our empirical study, we also conduct link prediction in the context of multiple social platforms; i.e., to infer missing links between users which are likely to exist in two social platforms.
On modeling of user interests across multiple social platforms, we conduct two studies. In the first study, we derived measures to analyze the similarity in user’s topical interests across multiple social collaborative platforms in the context of software engineering. In the second study, we propose a generative model that can learn the latent topics and the users’ topic-specific platform preferences from content generated by the users in multiple social platforms. We evaluate our proposed model’s ability to learn users’ topical interests using likelihood and perplexity, and compare the results with state-of-the-art baseline models. For evaluation of our generative model’s ability to learn the topic-specific preferences, we conduct predictive experiments to predict which platform a user is likely to publish the post after we have learnt its topics.
