Crane is a tool for searching documents in English based on their similarity to a query. It enables a user to submit a query of any length -- it can even be a large text document. In fact, the more informative the query is the more useful results can be obtained. As an option the tool performs a lexical analysis of the submitted document, e.g. text chunking and steeming for English, and extracts from it a series of small queries that are presented to the user. The user can pick up the most relevant small queries that they want to submit to the search engine. The tool returns a list of Web pages that best correspond to the content of the original query and/or small queries. For this, Crane uses a specialized algorithm for classifying documents by their similarity to the original query, rather than by PageRank-like algorithms that are based on popularity of pages.
We think that this type of tools could become useful in the era of world-wide web computing, when a huge number of services will be available through the Web, with many services offering similar features. Due to security and other non-technical reasons, it may be hard to assume that all these services will be registered in a single UDDI-like directory. Moreover, RESTful services are already represented as Web resources that can be accessed using URI addresses. These resources can be indexed as normal Web pages by popular search engines. In order to lookup for a service, a user (or application) could submit some machine-checkable description of the service as a query to a standard search engine. Our tool would return a list of addresses of services that best match to the description. In this case, the best result is not the most popular service but the one that matches the query.
See a short demo movie in the avi and mov format, showing the main features.
|
|
