Query Hidden Attributes in Social Networks

Microblogs and collaborative content sites such as Twitter and Amazon are popular among millions of users who generate huge numbers of tweets, posts, and reviews every day. Despite their popularity, these sites only provide rudimentary mechanisms to navigate their sites, programmatically or through a browser, like a keyword search interface or a get-neighbors (e.g., friends) interface. Many interesting queries cannot be directly answered by any of these interfaces, e.g., find Twitter users in Los Angeles that have tweeted the word "diabetes" in the last year. Note that the Twitter programming interface does not allow conditions on the user's home location. In this paper, we introduce the novel problem of querying hidden attributes in microblogs and collaborative content sites by leveraging the existing search mechanisms offered by those sites. We model these data sources as heterogeneous graphs and their two key access interfaces, LocalSearch and ContentSearch, which search through keywords and neighbors respectively. We show which of these two approaches is better for which types of hidden attribute searches. We conduct experiments on Twitter, Amazon, and RateMDs to evaluate the performance of the search approaches.