Answering cross-source keyword queries over deep web data sources

A popular trend in data dissemination involves online data sources that are hidden behind query forms, which are part of the deep web. Extracting information across multiple deep web sources in a domain is challenging, but increasingly crucial in many areas. Keyword search, a popular information discovery method, has been studied extensively on the surface web and relational databases. Keyword-based queries can provide a powerful yet intuitive means for accessing data from the deep web as well. However, this involves many challenges. For example, deep web data is hidden behind query interfaces, deep web data sources often contain redundant and/or incomplete data, and there is often inter-dependence among data sources. Thus, it is very hard to automatically execute cross-source queries. This paper focuses on answering cross-source queries over deep web data sources. In our approach, we model a list of deep web data sources using a graph to capture the dependencies among them, and we consider the problem of answering cross-source queries over these deep web data sources as a graph search problem. We have developed a bidirectional query planning algorithm to generate query plans for two types of cross-source queries, which are entity-attributes queries and entity-entity relationship queries.