Ontology With Hybrid Clustering Approach for Improving the Retrieval Relevancy in Social Event Detection

Abstract

Progression in digital technology and the fame of social media sites such as Facebook, YouTube, Flickr etc., necessitate sharing memories. This results in a colossal amount of multimedia content such as text, audio, photographs and video on the web. Retrieving photographs exclusively from web in the large collection is a challenging task. One way to retrieve photographs is by identifying them as events. The automatic organization of a multimedia collection into groups of items, where each group corresponds to a distinct event is described as Social Event Detection (SED). Contextual information, present for each photograph in social media adds semantics to the photographs. For semantic based retrieval, ontology based approaches yield good retrieval results, by reducing the number of false positives. So, the proposed approach moves with domain ontology construction followed by a hybrid clustering approach. Compared to the existing single-pass incremental clustering algorithm, the proposed approach ensures a good f-measure of 0.8608.