Understanding Tables on the Web

The Web contains a wealth of information, and a key challenge is to make this information machine processable. In this paper, we study how to “understand” HTML tables on the Web, which is one step further from finding the schemas of tables. From 0.3 billion Web documents, we obtain 1.95 billion tables, and 0.5-1% of these contain information of various entities and their properties. We argue that in order for computers to understand these tables, computers must first have a brain – a general purpose knowledge taxonomy that is comprehensive enough to cover the concepts (of worldly facts) in a human mind. Second, we argue that the process of understanding a table is the process of finding the right position for the table in the knowledge taxonomy. Once a table is associated with a concept in the knowledge taxonomy, it will be automatically linked to all other tables that are associated with the same concept, as well as tables associated with concepts related to this concept. In other words, understanding occurs when computers will understand the semantics of the tables through the interconnections of concepts in the knowledge base. In this paper, we illustrate a two phase process. Our experimental results show that the approach is feasible and it may benefit many useful applications such as web search.