Social-based traffic information extraction and classification

Social networks such as Twitter and Facebook are popular, personal, and real-time in nature. We found that there exists a significant number of traffic information such as traffic congestion, incidents, and weather in Twitter. However, an algorithm is needed to extract and classify the traffic information before publishing (re-tweeting) and becoming useful for others. Traffic information was extracted from Twitter using syntactic analysis and then further classified into two categories: point and link. This method can classify 2,942 traffic tweets into the point category with 76.85% accuracy and classify 331 traffic tweets into the link category with 93.23% accuracy. Our system can report traffic information real-time.

I. Introduction

Twitter has become a very popular micro-blog social network. It has been used as real-time text information dissemination. Tweetple or Twitter users are tweeting (to send text to display on profile page) in average of 55 million tweets (text-based posts composed of up to 140 characters) a day and 37 percent of Twitter's active users use their phones to tweet. In Thailand, there were 1,191,760 tweets from 79,705 Twitter users in a single day (April 19,2011), and 36.22% of them tweet from their phones (count form clearly Twitter mobile application only). These statistics show us that more than one-third of Twitter users are sending real-time messages of interesting events, such as weather reports, accidents, and traffic conditions. As for traffic-related messages alone, we found during a period of April 19, 2011-April 30, 2011 that there are about 2000 daily tweets that relay traffic information (e.g. traffic condition, accident) in Thailand.