This version of the page http://www.google.com.ua/enterprise/relevancy.html (0.0.0.0) stored by archive.org.ua. It represents a snapshot of the page as of 2008-01-12. The original page over time could change.
Google Enterprise

Relevancy

Many users of traditional universal search technology have seen inconsistent search results - some of their queries deliver accurate, well-targeted results while others return irrelevant information. Also, many administrators of search systems have spent time trying to “tune” their system to improve its accuracy only to find that they have made things slightly better for some small set of queries but much worse for the vast majority of queries and users.

Google set out to solve the universal search problem. Companies, government agencies, and educational institutions worldwide have enjoyed Google quality search results for some time. In a survey Google Search Appliance customers overwhelmingly endorse Google as a simple, easy, and effective solution to their universal search challenges:

  • About 75% of customers responding switched to Google from another search provider's product
  • Over 50% of customers who switch to Google, switched due to the poor relevancy their prior search provider delivered
  • Over 90% would recommend the Google Search Appliance to another company like theirs

Google Enterprise has focused on the search needs of companies and organizations for over 5 years. A dedicated team of universal search quality engineers continually refines the sophisticated set of algorithms to make the Google Search Appliance provide fast, relevant results to users’ queries. Taking the work from thousands of Google engineers, the Google Enterprise team then customizes the search algorithms and experience to meet the needs of enterprise customers.


Relevancy – Technology Tailored to Enterprise Needs

back to top


Google has made a name for serving the most relevant results. We feel relevancy is one of the most important factors that influence our users’ experience. Serving relevant results is no easy task. In determining what results to serve and how to rank them, a number of factors are used. For example: number of times a term appears on a given document; whether the query appears on the title; proximity of the query words, and many more. Hundreds of signals and factors are used to evaluate each and every single document, much like a human would do.

The Google Search Appliance leverages the technology tested daily by millions of users at google.com. We complement this technology with proprietary algorithms and models that we develop on an ongoing basis exclusively for our Enterprise customers. There are many variables that exist only in the Enterprise world that greatly affect the relevancy of results served to a user.

Some customers ask to what extent Google universal search technology relies on PageRank. The answer is: very little. PageRank is only one among more than a hundred factors that determine the relevancy of universal search. Even for consumer search – Google.com – PageRank is not the only factor.


The Myth of “Tuning”

back to top


Old-school universal search generally provides poor results “out of the box”. And so the practice of “tuning” search algorithms was born. Some tuning methods involve tweaking metadata terms during indexing, using scripts or formulas to “boost” specific documents or assigning different weight to specific terms during query time.

This approach carries several limitations. First, it’s a huge amount of manual work! It might work for the first few hundred, maybe the first thousand. But beyond that, it doesn’t scale. Complexity of administering scripts and having to re-start the engine or re-index documents anytime a change is made are certainly other examples of “tuning” limitations.

Most importantly, by “tuning” a certain set of results, a search administrator might be unknowingly demoting or penalizing some other terms of queries. This leads to fine tuning for a small subset of users or documents, at the expense of the enterprise as a whole.

What “tuning” is certain to do for you is increase your Total Cost of Ownership (TCO) – what you pay for upfront is only a small fraction of what you’ll have to pay over time. And even worse, there’s the opportunity cost of not finding quickly the right information your business needs.

At Google Enterprise we do all the heavy work for you. The Google search appliances leverage 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.


Measuring Search Quality

back to top

Search systems are comprised of many moving parts: the content that is being searched, the users and their queries and the search engine that brings it all together. Each one of these three components is facing constant change and evolution. Therefore, it is important to measure the quality of your search system on an ongoing basis, to determine how these changes are impacting your users’ experience.

Search quality evaluation starts by collecting queries that were issued to the search system and the results that were returned to the user. Each result is normally inspected by human assessors and judged for its relevance to the query. A result will typically be considered relevant if it is a good answer to the query and satisfies the information needs of the user who issued the query.

Many customers share their evaluations with Google and the search community at large. For example, some Raytheon search experts presented their findings on an universal search evaluation at the Semantic Technology Conference. They ultimately chose Google to provide integrated universal search across their desktop and enterprise content repositories. Most impressive was the response from their user base who participated in the survey: 84% found the "right" answer (the one they were looking for) in the top 3 results. This is just one example of the Google Search Appliance providing outstanding relevance, right “out of the box”, with no tuning required.


Summary

back to top

At Google, we're proud to see that a large majority of enterprise users are easily finding what they're looking for. And we’re happy that our no-tuning-required model is proved time and again. But we're not satisfied – we want to provide the best search experience for everyone. We're working closely with customers and partners to assess where and when our algorithms work best, and how we can further improve them.

Find more about how Google Enterprise consistently delivers the most relevant results at one of our Google Search Appliance or Google Mini seminars.