Record Linkage for Event Identification in XML Feeds Stream Using ELM

Abstract

Most of the news portals and social media networks are utilizing RSS feeds for information distribution and content sharing. Event identification improves the service quality of feeds providers in the aspect of content distribution and event browsing. However, thriving challenges arise due to representation of structural information and real-time requirement in feeds streams mining. In this paper, we focus on the record linkage problem which classifies stream content into known categories. To realize fast and efficient record linkage over XML feeds stream, we design two classification strategies: a classifier based on ensemble ELMs and an incremental classifier based on OS-ELM. Experimental results show that our solutions provide effective and efficient record linkage for event identification applications.