If you are curious about how probabilistic models can be used to analyze tweets for local event detection, this paper is a good start.

In order to overcome the limited abilities of unsupervised machine learning techniques such as Tweet-SCAN to detect and associate local events with tweets, the authors present an interesting approach to this problem. The main idea is to “integrate event detection and topic modeling.” This is enabled by “assigning topic proportions to events instead of assigning them to individual tweets.” It is all implemented and published as WARBLE, which is meant to be a new probabilistic model and learning scheme. This model and learning scheme incorporates two key factors: “a background model that captures spatiotemporal fluctuations of non-event tweets,” and the observation that “the shortness of tweets hampers the application of traditional topic models.”

Event detection in Twitter is notoriously difficult due to noise; this fact affects work with probabilities. Having said that, prerequisites for understanding the paper include a good knowledge of probabilistic theory and the related mathematical jargon. Hence, it is most suitable for researchers in the area rather than the general public. Nonetheless, it is a well-written paper with a robust experimental basis for its claims, that is, with respect to the rules of experimental science.