An Incremental Model For Multi-Label Classification Of Tweets Based On Label Co-Occurrence

The tremendous growth in the field of ICT encourages use of social media like Twitter as an effective platform for communication between local government and resident of the city. Citizens share numerous urban issues which need to be categorized properly as it decides helps to decide priority of the issue. Multiclass classification which is the traditional algorithm to classify the complaints into multiple classes loses some important information as people tend to write their inconveniences spreaded across multiple domains in a same complaint. Hence, this research work has developed a framework to identify multiple issues discussed in a tweet by performing multi-label classification. One of the most used multi-label algorithms is ‘Classifier chain’ method. However, the performance of the classifier chain algorithm varies with the label order used during classification. So our approach used the concept of label correlation to derive the optimal label sequence and implement the classifier chain algorithm for multi-label classification. In this study, 14 types of most important issues faced by the citizens are considered like potholes, traffic, illegal parking etc. Our experiments clearly indicate that the adoption of label correlation concept has a positive influence on the results.