Real-Time Sensing of Traffic Information in Twitter Messages
Sara Filipa Lemos de Carvalho · 2010
Traffic issues affect the mobility of many people and the dynamics of the big urban centers. The study of the traffic urban networks is of great importance for the improvement routes and traffic flow. This document intends to introduce a new source of information that can be helpful in the traffic scene analysis: microblogging messages. This new source surpasses some of the disadvantages of traditional traffic sensors, like area coverage and the costs of installation and maintenance. The problem in focus in this study is to investigate whether information can be found, in microblogging messages, that is relevant to the traffic study. The microblogging platform used to collect the messages was Twitter and the objective was to retrieve messages shared by individual users, in opposite to official sources like news agencies. The approach followed to solve the problem was to address it as a text classification problem using SVMs. The solution was divided into two main iterations of built and improvement of two classification models (bootstrapping strategy) to capture traffic messages. In each phase, for each model, the results achieved were registered and the evaluation measures were calculated and compared. Also, the improvements made from one phase to the other were registered. In the end, the two classification models were able to capture a generic traffic message with a precision of more than 80% and a traffic message shared by an individual user with a precision of approximately 50%. In conclusion, the objectives were achieved and the results were considered satisfactory in capturing messages from individual users, although in the capture of general traffic messages the results were a success.