April 12, 2021
Modular newsgathering makes content aggregation faster, more accurate
The use of small processing modules can significantly reduce overheads on computing systems with limited resources available to them when large amounts of data must nevertheless be processed. Research by a team in Greece described in the International Journal of Web Engineering and Technology shows how that approach can be used for content aggregation, information extraction, sentiment tagging, and visualization tasks.
Iraklis Varlamis and Dimitrios Michail of the Department of Informatics and Telematics at Harokopio University of Athens and Pavlos Polydoras and Panagiotis Tsantilas of Palo Ltd in Kokkoni, Greece, have demonstrated how this modular approach might function well on the social media and news analytics platform, PaloAnalytics. The team shows how their proposed architecture can easily withstand the pressures of increased content load when an issue goes viral on social media, such as when a major event takes place. The micro-modules that replace the monolithic architecture of conventional data-processing systems can quickly release unused resources when the content load reaches its normal flow.
The researchers point out that even from the early days of primitive web crawlers that became the foundation of search engines and other related tools, it was recognized that distributed processing is the only viable way to taming the vast quantities of textual data being generated even way back then. Today, the scale is almost unimaginable with many petabytes of data to be assimilated, aggregated, processed, indexed, and annotated with meaning. The vast realms of the web and social media systems offer us a rich seam to be tapped for information and knowledge if the tools can be built to cope with the bits and bytes.
The team's tests so far were based on analysis of 1500 websites, 10000 blogs, forums, hundreds of thousands of public Facebook pages, Instagram, Twitter, and YouTube updates, across six European nations and in six different languages. Their work shows where improvement might be made to build a powerful analytical tool that would be scalable and allow us to soon mine those enormous knowledge seams efficiently and in an effective way.