An Event Data Extraction Method Based on HTML Structure Analysis and Machine Learning

This paper proposes an event data extraction method that extracts business event data, such as coupons, tickets, sales campaigns, etc., from a homepage or blog of shops and pushes them to users. Users no longer need to browse their favorite shops´ homepage one by one. The method supports comprehensiveness and effectiveness for event data obtainment. This proposition works into two tasks: web page block segmentation and event data identification. The first task segments the web page into blocks. Each of the blocks includes information, such as title, notification, date, etc. Relating to event information. Many related works suppose web page block segmentation based on specific tags, vision, function, etc. In this research, we propose a web page block segmentation method based on HTML document structure analysis. The second task is used to identity event data from segmented blocks. We propose a method to implement event data identification based on machine learning. We show the results of a verification experiment. Experimental data are from 96 shops located in two underground shopping streets UNIMALL and ESCA, at a train station in the city of Nagoya (Japan). Because the event data identification depends on the Japanese language, this method is available for all the Japanese home page.