Identifying Actionable Information from Social Media for Better Government-Public Relationship

摘要

Recent years have witnessed a sudden growth of online social media (OSM). Among the various OSM sites, Twitter, an online micro-blogging website has attracted interest of millions of users who use it to access, publish and spread (share) news at a quick pace much faster than conventional news mediums. Latching on this trend, governments across the world have started using Twitter to communicate, engage with their citizens and increase their visibility and popularity. Social media makes it simpler and convenient for citizens to voice their opinions and reach out to a large section of people, and also for the government to listen to their grievances and concerns. While the data on Twitter is quite informative, it presents a challenge for analysis because it is disorganized and highly voluminous with high velocity. Every day, government accounts receive thousands of messages from the general public in the form of opinions, concerns, and grievances. It becomes an extremely troublesome task for the concerned ministries in the government, to manually filter out the irrelevant messages and respond to the genuine ones. In this paper, we attempt to address this problem by using learning algorithms to automatically classify the user generated messages as either actionable (those which can be acted upon by the government ministries) or non-actionable. We have considered eight supervised learning algorithms to classify the messages into actionable or non-actionable and compared their outcomes to determine the most effective learning algorithm. Our results indicate that random forest classifier gave the best results with an aggregate accuracy of more than 0.94 and an F-score of 0.94.