Machine learning for understanding the contextual semantics of tabular web sources

Tables are frequently used in web sources to present relational data in a human friendly manner. Because they are intended for humans, using machines to extract such information is difficult. There are approaches such as wrappers that attempt to solve this problem, but they lack adaptability and require high maintenance. Identifying and extracting information from web tables is not a trivial task, and understanding the semantics of a web table proves to be even harder. In this paper, we introduce a machine learning based approach to understand the semantics in the data residing in tabular web sources. We suggest features that reflect the characteristics of the content in the tables and analyze their impact on the accuracy of the classification process.