Download Full Text (2.6 MB)
Dr. Jin Soung Yoo
Department of Computer Science
Indiana University – Purdue University Fort Wayne
IPFW Sigma Xi (Scientific Research Honor Society) Student Research Competition
Best Graduate Student Poster Presentation
IPFW Student Research and Creative Endeavor Symposium Award Winner
Twitter is a free online social networking and micro-blogging service that asks “What are you doing?” In 2014, Twitter’s average monthly active users (MAUs) were 255 million, and mobile MAUs reached 198 million with a year-over-year increase of 25% and 31% respectively.Twitter data is becoming a popular source of social trajectory data (e.g. geo-economic events—how the public’s sentiment affects the stock market; the discovery of unusual social events (i.e. festivals, demonstrations, natural disasters—earthquakes and storms); geographic disease and influenza trends; social questions—are popular events associated with increased public sentiment; and predicting political alignment).Although the defining and detecting of events (i.e. feature types) have long been a research topic, the characteristics of Twitter make it a non-trivial task.For example, Twitter events are usually overwhelmed with “babble” (i.e. noisy, unstructured text); about 40% of Tweets do not include the target event queried by the user; Twitter event detection algorithms need to be scalable given the sheer amount of Tweets; Twitter contents are dynamically changing and increasing in a real-time nature; the majority of Twitter APIs only grant access to a 1% sample of the Twitter data, and concerns about Twitter’s sampling strategy and the quality of the Twitter data have been raised.This work presents three twitter event detection techniques for use in various domains:(1) content analysis and time-series modeling within a geographic area (i.e. frequency bursts); (2) deviation/anomaly detection within a geographic area (i.e. statistical outliers from the norm); and (3) Probabilistic Soft Logic (i.e. probabilistic reasoning in relational domains).Effective Twitter event detection is critical for Twitter data mining and predicting the present (i.e. “contemporaneous forecasting” or “nowcasting”) — a topic of special interest to the central banks and other government agencies.
Computer Sciences | Physical Sciences and Mathematics
Kimmey, David, "Twitter Event Detection" (2015). 2015 IPFW Student Research and Creative Endeavor Symposium. 43.