A New Architecture of an Intelligent Agent-Based Crawler for Domain-Specific Deep Web Databases

A key problem of retrieving, integrating and mining rich and high quality information from massive Deep Web Databases (WDBs) online is how to automatically and effectively discover and recognize domain-specific WDBs' entry points, i.e., searchable forms, in the Web. It has been a challenging task because domain-specific WDBs' forms with dynamic and heterogeneous properties are very sparsely distributed over several trillion Web pages. Although significant efforts have been made to address the problem and its special cases, more intelligent and effective solutions remain to be further explored. In this paper, a new architecture of an intelligent agent-based crawler (iCrawler) for domain-specific Deep Web databases has been proposed to address the limitations of the existing methods. The iCrawler, based on intelligent learning agents and domain ontology, and a series of novel and effective strategies, including a two-step page classifier, a link scoring strategy, etc, can improve the performance of the existing methods. Experiments of the iCrawler over a number of real Web pages in a set of representative domains have been conducted and the results show that the iCrawler outperforms the existing domain-specific Deep Web Form-Focused Crawlers (FFCs) in terms of the harvest rate, coverage rate and time performance.