A fine-grained taxonomy of tables on the web

We propose a classification taxonomy over a large crawl of HTML tables on the Web, focusing primarily on those tables that express structured knowledge. The taxonomy separates tables into two top-level classes: a) those used for layout purposes, including navigational and formatting; and b) those containing relational knowledge, including listings, attribute/value, matrix, enumeration, and form. We then propose a classification algorithm for automatically detecting a subset of the classes in our taxonomy, namely layout tables and attribute/value tables. We report on the performance of our system over a large sample of manually annotated HTML tables on the Web.