Generic metadata representation framework for social-based event detection, description, and linkage

Abstract

Various methods have been put forward to perform automatic social-based event detection and description. Yet, most of them do not capture the semantic meaning embedded in online social media data, which are usually highly heterogeneous and unstructured, and do not identify event relationships (e.g., car accident temporally occurs after storm, and geographically occurs near soccer match). To address this problem, we introduce a generic Social-based Event Detection, Description, and Linkage framework titled SEDDaL, taking as input: a collection of social media objects from heterogeneous sources (e.g., Flickr, YouTube, and Twitter), and producing as output a collection of semantically meaningful events interconnected with spatial, temporal, and semantic relationships. The latter are required as the building blocks for event-based Collective Knowledge (CK) organization, where CK underlines the combination of all known data, information, and metadata concerning a given concept or event. SEDDaL consists of four main modules for: i) describing social media objects in a generic Metadata Representation Space Model (MRSM) consisting of three composite dimensions: temporal, spatial, and semantic, ii) evaluating the similarity between social media objects’ descriptions following MRSM, iii) detecting events from similar social media objects using an adapted unsupervised learning algorithm, where events are represented as clusters of objects in MRSM, and iv) identifying directional, metric, and topological relationships between events following MRSM’s dimensions. We believe this is the first study to provide a generic model for describing semantic-aware events and their relationships extracted from social metadata on the Web. Experimental results confirm the quality and potential of our approach.