Title: Scalable Detection of Emerging Topics and Geo-spatial Events in Large Textual Streams

Abstract: Social media are a popular source for live textual data. This data poses several challenges due to its size, velocity, and heterogenity. Existing methods for emerging topic detection often are only able to detect events of a global magnitude such as natural disasters, or they can only monitor user-selected keywords or a curated set of hashtags. Interesting emerging topics may, however, be of much smaller magnitude and may involve the combination of two or more words that are not yet known in beforehand.
The presentation will discuss:
(i) A significance measure that can detect emerging topics early, long before they evolve into "hot tags".
(ii) An efficient "online" heavy-hitters type algorithm able to track these statistics for all words and word-pairs with only a fixed amount of memory, and without predefined keywords.
(iii) How to incorporate location information into this process to both allow reporting the locality of events as well as detecting local-only geo-textual patterns.
(iv) Demonstrate the usefulness and scalability on a data set of a billion tweets.