ColNet: Embedding the Semantics of Web Tables for Column Type Prediction

Automatically annotating column types with knowledge base(KB) concepts is a critical task to gain a basic understandingof web tables. Current methods rely on either table metadatalike column name or entity correspondences of cells in theKB, and may fail to deal with growing web tables with in-complete meta information. In this paper we propose a neu-ral network based column type annotation framework namedColNetwhich is able to integrate KB reasoning and lookupwith machine learning and can automatically train Convolu-tional Neural Networks for prediction. The prediction modelnot only considers the contextual semantics within a cell us-ing word representation, but also embeds the semantics of acolumn by learning locality features from multiple cells. Themethod is evaluated with DBPedia and two different web ta-ble datasets, T2Dv2 from the general Web and Limaye fromWikipedia pages, and achieves higher performance than thestate-of-the-art approaches