MINING OPINIONS ABOUT TRAFFIC STATUS USING TWITTER MESSAGES

In this paper, we have described system for mining opinions from traffic status tweets. The opinions are categorized as positive (p) and negative (n). We collected a corpus of around 5000 traffic related tweets using twitter API. These tweets were then cleansed from unwanted contents and processed. The processed tweets were converted into a bag of words and manually labeled as “p” or “n” appropriately. This labeled dataset was then split into training and test set in the ratio of 80:20. Various classifier algorithms were trained on this set and validated. Based on the performance of the algorithms the top 7 among them were chosen to form an ensemble model. This ensemble model was then used to classify the test set. We obtained an F-measure of 87.15 which indicates that our system is quite competitive in mining opinions about traffic status from tweets.