Deep Web structure

Abstract

Our current understanding of Web structure is based on large graphs created by centralized crawlers and indexers. They obtain data almost exclusively from the so-called surface Web, which consists, loosely speaking, of interlinked HTML pages. The deep Web, by contrast, is information that is reachable over the Web, but that resides in databases; it is dynamically available in response to queries, not placed on static pages ahead of time. Recent estimates indicate that the deep Web has hundreds of times more data than the surface Web. The deep Web gives us reason to rethink much of the current doctrine of broad-based link analysis. Instead of looking up pages and finding links on them, Web crawlers would have to produce queries to generate relevant pages. Creating appropriate queries ahead of time is nontrivial without understanding the content of the queried sites. The deep Web's scale would also make it much harder to cache results than to merely index static pages. Whereas a static page presents its links for all to see, a deep Web site can decide whose queries to process and how well. It can, for example, authenticate the querying party before giving it any truly valuable information and links. It can build an understanding of the querying party's context in order to give proper responses, and it can engage in dialogues and negotiate for the information it reveals. The Web site can thus prevent its information from being used by unknown parties. What's more, the querying party can ensure that the information is meant for it.