2020/05/14 by Colin Lockard, Prashant Shiralkar, Lockard, Colin +5 · 3 citations
Computer Science · Decision Sciences · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Information Retrieval (cs.IR) #Scientific Computing and Data Management #Topic Modeling #Web Data Mining and Analysis
paper · pdf · doi:10.48550/arxiv.2005.07105
openalex publication_date 2020/05/14 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
In many documents, such as semi-structured webpages, textual semantics are\naugmented with additional information conveyed using visual elements including\nlayout, font size, and color. Prior work on information extraction from\nsemi-structured websites has required learning an extraction model specific to\na given template via either manually labeled or distantly supervised data from\nthat template. In this work, we propose a solution for "zero-shot" open-domain\nrelation extraction from webpages with a previously unseen template, including\nfrom websites with little overlap with existing sources of knowledge for\ndistant supervision and websites in entirely new subject verticals. Our model\nuses a graph neural network-based approach to build a rich representation of\ntext fields on a webpage and the relationships between them, enabling\ngeneralization to new templates. Experiments show this approach provides a 31%\nF1 gain over a baseline for zero-shot extraction in a new subject vertical.\n