Applying WebTables in Practice

We started investigating the collection of HTML tables on the Web and developed the WebTables system a few years ago [4]. Since then, our work has been motivated by applying WebTables in a broad set of applications at Google, resulting in several product launches. In this paper, we describe the challenges faced, lessons learned, and new insights that we gained from our efforts. The main challenges we faced in our efforts were (1) identifying tables that are likely to contain high-quality data (as opposed to tables used for navigation, layout, or formatting), and (2) recovering the semantics of these tables or signals that hint at their semantics. The result is a semantically enriched table corpus that we used to develop several services. First, we created a search engine for structured data whose index includes over a hundred million HTML tables. Second, we enabled users of Google Docs (through its Research Panel) to find relevant data tables and to insert such data into their documents as needed. Most recently, we brought WebTables to a much broader audience by using the table corpus to provide richer tabular snippets for fact-seeking web search queries on Google.com.