Retrieving Attributes Using Web Tables

摘要

In this paper we propose an attribute retrieval approach which extracts and ranks attributes from Web tables. We use simple heuristics to filter out improbable attributes and we rank attributes based on frequencies and a table match score. Hanking is reinforced with external evidence from Web search, DBPedia and Wikipedia. Our approach can be applied to instances (e.g. Canada) to retrieve their attributes (capital, GDP). Our approach has a much higher recall than DBPedia and Wikipedia and that it works better than lexico-syntactic rules for the same purpose.