Query Routing: Finding Ways in the Maze of the DeepWeb

This paper presents a source selection system based on attribute co-occurrence framework for ranking and selecting Deep Web sources that provide information relevant to users requirement. Given the huge number of heterogeneous Deep Web data sources, the end users may not know the sources that can satisfy their information needs. Selecting and ranking sources in relevance to the user requirements is challenging. Our system finds appropriate sources for such users by allowing them to input just an imprecise initial query. As a key insight, we observe that the semantics and relationships between deep Web sources are self-revealing through their query interfaces, and in essence, through the co-occurrences between attributes. Based on this insight, we design a co-occurrence based attribute graph for capturing the relevances of attributes, and using them in ranking of sources in the order of relevance to user’s requirement. Further, we present an iterative algorithm that realizes our model. Our preliminary evaluation on real-world sources demonstrates the effectiveness of our approach.