Revisiting Web Data Extraction Using In-Browser Structural Analysis and Visual Cues in Modern Web Designs

摘要

Recent trends in website design have an impact on methods used for web data extraction. Many existing methods rely on structural analysis of web pages and, with the introduction of CSS, table-based layouts are no longer used, while responsive design means that layout and presentation are dependent on browsing context which also makes the use of visual clues more complex. We present DeepDesign, a system that semi-automatically extracts data records from web pages based on a combination of structural and visual features. It runs in a general-purpose browser, taking advantage of direct access to the complete CSS3 spectrum and the capability to trigger and execute JavaScript in the page. The user sees record matching in real-time and dynamically adapts the process if required. We present the details of the matching algorithms and provide an evaluation of them based on the top ten Alexa websites.