Event Related Data Collection from Microblog Streams

Many studies have established that microblog streams, e.g., Twitter and Weibo, are leading indicators of emerging events. However, to statistically analyze and discover the emerging trends around these events in microblog message streams, e.g., popularity, sentiments, or aspects, one must identify messages related to an event with high precision and recall. In this paper, we propose a novel problem of automatically discovering meaningful keyword rules, which help identify the most relevant messages in the context of a given event from fast moving and high-volume social media streams. For the specified event, such as {#trump} or {#coronavirus}, our technique automatically extracts the most relevant keyword rules to collect related messages with high precision and recall. The rule set is dynamic, and we continuously identify new rules that capture the event evolution. Experiments with millions of tweets show that the proposed rule extraction method is highly effective for event-related data collection and has precision up to 99% and up to 4.5X recall over the baseline system.