Ontology-driven urban issues identification from social media.

Abstract

The cities worldwide face with many issues directly related to the urban space, especially in the infrastructure aspects. Most of these urban issues generally affect the life of both resident and visitant people. For example, people can report a car parked on a footpath which is forcing pedestrians to walk on the road or a huge pothole that is causing traffic congestion. Besides being related to the urban space, urban issues generally demand actions from city authorities. There are many Location-Based Social Networks (LBSN) in the smart cities domain worldwide where people complain about urban issues in a structured way and local authorities are aware to fix them. With the advent of social networks such as Facebook and Twitter, people tend to complain in an unstructured, sparse and unpredictable way, being difficult to identify urban issues eventually reported. Social media data, especially Twitter messages, photos, and check-ins, have played an important role in the smart cities. A key problem is the challenge in identifying specific and relevant conversations on processing the noisy crowdsourced data. In this context, this research investigates computational methods in order to provide automated identification of urban issues shared in social media streams. Most related work rely on classifiers based on machine learning techniques such as Support Vector Machines (SVM), Naïve Bayes and Decision Trees; and face problems concerning semantic knowledge representation, human readability and inference capability. Aiming at overcoming this semantic gap, this research investigates the ontology-driven Information Extraction (IE) from the perspective of urban issues; as such issues can be semantically linked in LBSN platforms. Therefore, this work proposes an Urban Issues Domain Ontology (UIDO) to enable the identification and classification of urban issues in an automated approach that focuses mainly on the thematic and geographical facets. Experimental evaluation demonstrates the proposed approach performance is competitive with most commonly used machine learning algorithms applied for that particular domain.