UPDATE: Video of this presentation is now available online.
Next week we have Alex Schulz from the Technical University of Darmstadt who will giving a talk about his work on using social media data along with machine learning, and semantic dictionaries (i.e., WordNet), to automatically detect small scale incidents, such as car crashes, shootings, and fires.
I saw Alex present a paper co-authored with Petar Ristoski at ICWSM during a really interesting workshop titled When the City Meets the Citizen. In that paper they analyzed Twitter data from Seattle and Memphis. One of their findings was that average citizens (labeled I and blue in the figure below) were often the first to report shootings (53% of the time), much earlier than other people that one would expect such as Emergency Management Organzations (EMO), journalists/bloggers devoted to emergencies (EMJ), general journalists/bloggers, or other types of governmental and non-governmental organizations (ORG).
Here is what Alex has to say about his presentation:
Microblogs are increasingly gaining attention as an important information source in emergency management as a lot of valuable situational information is shared, both by citizens and official sources. However, current analyses focus on information shared during large scale incidents, with high amount of publicly available information. In contrast, in this talk we present the results of several studies on every day small scale incidents. The comparably low amount of information shared per event makes this significantly harder.
First, we show the results of a machine learning experiment for automatically detecting relevant information related to small scale incidents. With a precision of 82.2%, we are able to detect three different types of small scale incidents in microblogs. Second, we highlight the value of information present in microblogs for increasing situational awareness. For that, we demonstrate that incidents detected based on microblogs are correlated with real-world incidents. We also show the results of our analysis on user behavior during this type of incidents.
Axel Schulz is a PhD student and works as a Research Associate at the
Telecooperation Lab, Technische Universität Darmstadt and SAP Research, Darmstadt. He received his Diploma in Business and Computer Science from the same department in 2011. His PhD thesis is focused on making user-generated content usable for decision making in urban management. In this case, he focuses on the steps needed to fuse different types of user-generated content: geolocalization,
classification, and filtering of microblogs and sensor data based on
semantic web technologies and machine learning.
Come learn more about this at Alex’s talk on Tuesday August 27, 1:30pm, at Microsoft Research’s Building 99 (room 1915A). For non-Microsofties, please contact me at email@example.com.