Friday, October 17, 2003

The Myths of Myths of Federated Searching

The Truth About Federated Searching "Not all federated search engines can search all databases, although most can search Z39.50 and free databases. But many vendors that claim to offer federated search engines cannot currently search all licensed databases for both walk-up and remote users. Why? Authentication."

TKS (as do other systems) allows authentication to be presented as needed so that searching can be performed on both secure and free data.

"It's impossible to perform a relevancy ranking that's totally relevant. A relevancy ranking basically counts the occurrence of words being searched in a citation. Based on this frequency of occurrence, items will be moved closer to the top or farther down the results list. Here's the problem: When attempting to relevancy-rank citations, the only words you have to work with are those that appear in the citation. Often, the search word doesn't even appear. "

Relevance calculations, using a histogram, can provide you with better relevancy rankings based on shared nodes. Of course, if the relationship are not there then it can't work. Although, to get around not having shared keywords you can use shared concepts extracted from a text mining tool.

"You can't get better results with a federated search engine than you can with the native database search. The same content is being searched, and a federated engine does not enhance the native database's search interface."

This is totally wrong. The more context, the more metadata that's extracted and related across different data sources, the better (more relevant) the search results.

No comments: