Hello Dr. Hirschberg thank you for your interest. We are currently able to identify transportation events with 94% accuracy. Unfortunately the localization of the tweets is relatively low based on sole use of twitter location data or user input data. We are currently addressing this by cross referencing locations and landmarks mentioned in tweets by using a geo-dictionary, or gazetteer. By using this inferred data we hope to achieve a higher — and more precise — amount of location data.

That is a future consideration for this project once the initial system is fully deployed. There are trust, reputation, and uncertainty components that will need to be addressed at that time. Once all of these facets have been integrated the VTIS will also accomodate input from users as well as feedback on that form of input as well.

In papers such as “Traffic Observatory: a system to detect and locate traffic events and conditions using Twitter” by Ribeiro et al. my previously mentioned method is discussed. By combining prior knowledge of geolocational data we can provide a more accurate placement of an event. For example, if a landmark such as the Empire State Building was mentioned within a tweet with a transportation event there is no explicit location; however, by accessing a geolocational reference such as a gazetteer we could provide a localization to the tweet that could not be extracted simply from the plaintext.

That is a very interesting consideration. We don’t currently model that; however, if the affect of our system grows that may be necessary functionality that would need to be included.

Thank you for your interest. The user would input some amount of tolerance on a scale of 1-10 where 10 is the highest tolerance. The user would specify this when they created the route via a simple drop down box.

Because of user preference we feel this is an important aspect that should be determined on a user by user basis.