The Myth of Tuning

Most enterprise search vendors admit that their algorithms provide generally poor results. To compensate, they ask customers to "tune" the search algorithms to meet their specific needs. Sounds good so far, right? In reality, this customer-implemented "tuning" in the context of enterprise search involves significant manual work to create the relevancy algorithm. Common methods employed are boosting metadata terms during indexing, boosting specific documents via scripts or formulas, and boosting specific term weightings at query time. These and other approaches require a significant amount of manual programming work on behalf of the customer, are often based on trial and error approaches, and are just hard to maintain. After all, do you want to run your business or get your Ph.D. in search?

The biggest problem with this "tuning" approach is the complexity of administering the code and scripts. Search administrators, representing the needs of thousands of users in their organization, must manually assign weights or boosts to specific metadata, documents, or query terms. These assignments are made through complex configuration scripts and algorithms that must be coded into the engine. Anytime a value is modified, the engine must be restarted, and sometimes the content re-indexed. Therefore, tuning search is a bit like the Butterfly Effect. In the classic example, mathematician and meteorologist Edward Lorenz asks: If a butterfly beats its wings in Brazil, will it set off a tornado in Texas?

Well, tuning search in the enterprise can likewise set off a tornado of search results chaos! Using these technologies, the act of "tuning" is actually more comparable to building your own search algorithm. We at Google, on the other hand, recognize that search is a complex science with hundreds of factors that must be taken into account. The Google Search Appliance leverages the work of thousands of engineers to get the correct answer, right out of the box, with no "tuning" required. The system also adapts to the needs of your enterprise, taking into account corpus specific factors and learning how users query, when they misspell words, and how often content is changing.

So we recommend you focus your efforts on more business-style tuning, including creating intuitive user interfaces, uploading your companies unique acronyms and vocabulary as synonyms and suggested queries, and promoting key results using KeyMatch. The Google Search Appliance will take care of the rest, providing fast, secure, and accurate access to information throughout your enterprise.