TwitterTrends: a spatio-temporal trend detection and related keywords recommendation scheme

Twitter is a very popular online social networking service that enables its users to post and share text-based messages known as tweets. Even though one tweet may contain at most only 140 characters, the number of tweets generated daily is enormous and hence, collectively, they can give important clues to the resolution of interesting issues such as those associated with public opinion and current trends and the retrieval or recommendation of hot multimedia contents. In this paper, we propose a spatio-temporal trend detection and related keyword recommendation scheme for tweets called TwitterTrends. Our scheme can identify hot keywords and recommend their related keywords at a given location and time by analyzing user tweets and their metadata such as GPS data. The scheme is based on a client–server collaboration model for efficiency. The client on the user device manages user interactions with the Twitter server, such as the writing and uploading of tweets. In addition, it selects candidate trend keywords from tweets by simple filtering, collects user location data from the mobile user device, and sends them to our trend processing (TP) server. Our scheme can show trend keywords and their related keywords intuitively and expand them on the fly by displaying relevant keywords collected from portal sites such as Wikipedia and Google. The TP server collects candidate trend keywords and metadata from all the clients and analyzes them to detect spatio-temporal trend keywords and their related keywords by considering their co-occurrence in tweets. Our scheme is very robust in that it can handle typical input events such as abbreviations and typing errors that occur when writing tweets on mobile devices as well as provide supplementary keywords from portal sites. We implemented a prototype system and performed various experiments to demonstrate that our scheme can achieve satisfactory performance and scalability.