Personalizing Google Week, we conducted an email interview with Google's Sep Kamvar, who is Lead Software Engineer for Personalization at Google. Sep was the founder of Kaltix, a search engine that was acquired by Google in 2003 and is thought to be the basis of Google's current personalization efforts.As part of our
Here is our interview, with questions derived from a number of the R/WW authors:
R/WW: In a general sense, do you think privacy is less important now than it used to be - due to the popularity of social networks and social software? Because it does seem that to achieve true personalization, some sacrifices need to be made in terms of privacy (note: we're not getting at Google here, but in terms of web technology it is a noticeable trend these days).
Sep: Even despite the current trends, it is incredibly important to design products with the utmost respect for user data and we do this in the following ways:
1. Choice. Our Web History product is an optional product. Those who don't want it, can opt to not use the product.
2. Transparency. For those users who opt in to the Web History product, we show them their previous queries, so that our users can see all the data that is used to personalize their search results.
3. Control. Our Web History users have the ability to pause Web History at any time, or go back and delete individual items.
4. Data Portability. Our Web History users can export their web history data to another service through an RSS feed.
R/WW: In May at the Google Personalization Workshop, you mentioned that Google wants to compute PageRank for every single person. How far away is a 'Personal PageRank' system for mainstream (Mom and Pop) people, who may not have a Google Account yet and are probably a long way off using iGoogle? Is an Account necessary, or can PageRank be personalized in other ways, without Mom and Pop even realizing it perhaps?
Sep: We have various levels of personalization. For those who are signed up for Web History, we have the deepest personalization, but even for those who are not signed up for Web History, we personalize your results based on what country you are searching from. As we move forward, personalization will continue to be a gradient; the more you share with Google, the more tailored your results will be.
To answer your question as to whether Mom and Pop would realize it, it's important for us to be transparent about what we're using to personalize your search results, and we will continue to strive towards that as we increase the levels of personalization in search.
R/WW: Does Google use the information gleaned from GMail to improve the targeting of search ads, or even the branding?
Sep: No, we don't use GMail for personalization of search or search ads.
R/WW: Is personalization really a problem in search? Doesn't the context of the query give the information that the search engine needs?
Sep: Peter Fleischer's article in the Financial Times gives a nice rundown of some examples of when personalization is useful. [Ed: Peter Fleischer is global privacy counsel for Google]
R/WW: A few people in the comments for this week's Read/WriteWeb poll mentioned that Google's personalized search ended up just giving them the same results over and over again, which wasn't helpful for them. In other words, they wanted to search for new things, not things they'd already found previously. This may explain why 9% of poll respondents say personalized search has given them worse results. Do you have any explanation or response to those people?
Sep: That's interesting feedback. Giving the user a diverse set of results for a query is important to us, and for this reason we have tended towards making the effects of personalization subtle. Most of the individual rankings changes that we do at Google (both in personalization and outside of personalization) have an effect that's not highly perceptible to our users as a large change.
In terms of the specific concern as to whether personalization would give them the same results repeatedly, this is a reasonable concern, which is why the majority of our algorithms aren't simply boosting those results that you've seen before, but rather taking into account things like your location and your interest in order to give results that are both relevant to your query and to you.
R/WW: Thank you Sep for the interview.
Note: photo is via Google Blogoscoped