Personalized and Customized Search
Moderator Danny Sullivan says that it was very hard to get speakers for this topic because many people aren’t sure what they’re doing yet as far as personalization and customization are concerned. But Bryan Horling, Software Engineer, Personalized Search, Google, is hoping to give an overview of what’s going on, what changes to expect and the history behind personalized and customize search.
Here’s an example using “dinner” as a query. The top two results are a Wikipedia article and a site with recipes. Basically, the results will help you research what dinner is or find a site that requires additional work. In a personalized version, there’s a result for a manicotti recipe and a result for an area restaurant. This shows the difference between what a user might want and what is being provided, and the two are miles apart. If every person was to write out a list of what they’d expect from the query “dinner”, everyone’s would be different, so it’s not a one-size-fits-all approach. [I've heard that about personalization somewhere before... --Susan]
Why personalize? You get the user to the right information as quickly as possible. Many searches are inherently ambiguous, so they can be a challenge to answer correctly for the individual user.
He won’t be talking about: search preferences, iGoogle, custom search engine, subscribed links, and Google Desktop, although these things can all affect a user’s search experience.
Early personalization history at Google
- Kaltix acquisition (2003): Three guys from Stanford wrote some cool things about personalizing PageRank.
- Personalized search on Google Labs (2004): Explicit, specify your interests.
- Personalized search launched (2005): Implicit, based on your Web history.
Web history records a user’s queries and clicks and gets a good idea of what the user is actually interested in. It’s better than models where users enter their interests because sometimes people don’t answer in the way that best shows their interests.
Principles of Personalization
User privacy is key. Without the trust of users, no one’s going to allow their information to be used. This is done through:
- Transparency: Inform when and what changes are made.
- Security: Sensitive info, including personalization based on that info is only available when signed in.
- Control: Users can edit or delete underlying data or turn the service off.
On the search results page, click on “Web History” for a page that displays Web history. The date and time of all activity is shown, and there’s a calendar where heavy search days occurred as well as a place that categories are sorted, from images to news to blogs and videos. This is the transparency arm.
Web history control is letting the user remove or pause Web history. Individual items can be removed from the history. A user can also clear their entire Web history. This has been around for a few years.
More recently, there are customized results for locations. By clicking on the “more details” link, there is info shown for the location and you can compare results if the Web history was not applied.
- Using the searcher’s geolocation to affect search.
- Different levels of granularity.
- Both explicit and implicit information.
Country level localization will serve results that apply to the country you’re in. He uses the query “got talent” as an example, and in the UK the result is Britain’s Got Talent while in the U.S. the result is America’s Got Talent.
Regional localization is, again, intuitive. When querying “metro” in D.C., you’ll get results for public transit, but if you search in San Francisco, you’ll get results for the publication.
City localization will show local results, especially for queries with strong local intent, like “pizza”. That query will result in local business results for pizza in your town or city.
- Using the searcher’s personal context to rank results
- Recent searches (short term)
- Web history (long term)
Recent searches – Disambiguation
Searching for “jordans” would probably result in the sneaker, but if their recent queries were about furniture, the results will show Jordans the furniture store. Generally, it results in a re-ranking that still includes shoe results.
Web History – Disambiguation
A search for “galaxy” will mostly show pages about space for non-personalized results. But if the searcher has been looking at soccer sites, the LA Galaxy will show up more proficiently.
Web History – Refinding
A result that is visited previously will show that it has been visited and that site’s listing may possibly rise in the position it shows up for.
What does this mean for SEM?
- Collecting metrics
- Seeing how your pages rank
- Easier for people looking for your service to find you.
- Easier to retain customers who prefer your business.
The top position is not winner-take-all. To take advantage of personalized search:
- Create compelling and interesting content.
- Appeal to users, not search engines.
- You can control personalization for your searches. Use search details. Disable it by appending &pws=0 to searches.
- Sign out.
- Firefox extension, greasemonkey script.
- Edit or turn off Web history.
If I want to create personalization on my site itself, how can I do it?
A lot of what we try to address with personalized search is ambiguity, but within a particular site, there’s probably not the same need.
What’s the percentage of people that are actually using this on a regular basis?
All Bryan can say is “a bunch”.
How does this all tie into local business accounts?
He doesn’t know a lot about local business, but the issue is what if you’re looking for a result in an area that you don’t live? We probably aren’t serving those as well as you might in the short term. But in the long term, Web history would help resolve this as it figures out that you like to visit some place. Of course, if you’re looking for something out of town, you’ll probably make it a town specific query. There’s nothing stopping searchers from refining queries.
Could personalization help a site that seems like another site?
People with similar interests will see similar results.
Do you need to have a Google account to be served personalized search results?
Previous queries and geo-based queries don’t require a Google account.