Wednesday, December 13, 2006

AlwaysOn panel: What is the data telling us?

There are some interesting tidbits on personalization in this July 2006 AlwaysOn panel, "What Is the Data Telling Us?", with Peter Norvig (Google), Jim Lanzone (Ask), Usama Fayyad (Yahoo), and Michael Yavonditte (Quigo).

The panel moderator, Bambi Francisco, focused on privacy issues at the beginning, and the panelists appeared a little reluctant to talk. Usama Fayyad started off early by saying:
Knowing what people do collectively or in segments of special interests gives you a lot of very interesting information and a lot of leverage in terms of product and making things more relevant, including making advertising more relevant, and makes a better service.
A bit generic, but it is a good framing of the problem. We are trying to use aggregate data to make search and advertising more relevant and useful.

Bambi continued poking at the privacy issue, sparking Peter Norvig to say that Google really does not need or want to know everything about you. As Peter explained, building up some uber profile of everything you have ever done is less important than focusing on your recent history:
What's important is not you as an individual, but it's the role you are playing at the moment. When you are looking for one particular piece of information, I don't want to know about you so much as I want to know about all the other people in the same situation and what they did then.

And I'd rather know about what is your history for the last five minutes as you try to solve this problem than know about your history for the last five years.
Exactly right. What matters is your current mission, what you are trying to do right now. We can help by paying attention to what you are doing right now and helping you get it done.

Jim Lanzone chimed in around here, both talking about how users will not do a lot of up-front work in search and expanding on Peter's point about helping people with the problem they are currently trying to solve:
Most users are actually very lazy. While some high end users might use products that require tagging, the vast majority of people won't.

The behavior they will use is to iterate on a search engine. That one white box is just so easy for them to put in whatever is in the top of their head ... then the average searcher will review a result page in 5 seconds or less ... they get clues and then they will iterate their search.

That's why the average search session will have 3 or 4 searches ... That is part of the game for them, is finding a clue, iterating their search, getting more specific, and then finding what they need.

It's not worth their time to sit there and toggle a bunch of things in advance of their query, to then hopefully get a better result. It just saves them time to start going.
At this point, Bambi seemed to shift focus a bit and ask a bunch of questions about personalization and recommendations. Again, Bambi was not getting a lot of answers, but most of the answers she did get were fairly negative toward the idea of personalized search.

For example, Usama said, "You really can't read the searcher's mind," a statement that reminded me of a quote from former A9 CEO Udi Manber: "People will learn to use search better but have to invest the thinking -- we are not in the mind reading business." I was surprised to see Peter echo this point, saying something to the effect that Google would have to be clairvoyant to guess user intent given a search of a couple keywords.

I think both of these statements miss the point of personalized search. The idea is not do to something with nothing. That would be magic, mind reading. No, the idea behind search personalization is to add data about what a searcher has done -- especially what a searcher just did -- to refine the current search.

If the couple keywords in a search are too vague, looking back at a searcher's history may help disambiguate it. If a searcher is iterating and not finding what they want, paying attention to what they just did and did not find can help us narrow down on what they might need.

The entire talk is good fun, worth watching. Usama is focused on Yahoo Answers and social search. Jim talks mostly about search experience and making search easy. Peter adds clarity on a few points and has a few amusing anecdotes. Do not miss Peter's joke around 53:23 in the video about a haiku he found of some searches in the logs, "a story of ... frustration and release", very funny.


Anonymous said...

To me, this is the difference between "contextual" and "behavioral". Context is now. Behavior is what you've shown over time. Just thinking in terms of advertising, we've found that context is much more valued by the user (higher click-thru rates, eCPM, etc.) and the advertiser.

Obviously the field of search has to prioritize context first, and behavior second. I suspect at Findory you're more focused on behavior-driven personalization.

I agree with your opinion about "behavior" being able to supplement "context".

Peter's comment "I don't want to know about you so much as I want to know about all the other people in the same situation and what they did then" particularly resonated with me. At Others Online, we're trying to *show* you those people in real time and let you connect with them. It goes beyond search though.

jeff.dalton said...

There is a lot of research and working going into making search more context aware. For example:

* Where is the user located? (geographic location or even physical location: home, work, car, etc...)

* When is the query being executed? (what time of day, year, month, etc...). For example, are they more likely to be water vs. snow skiing?, etc...

* What is the user doing when the query is executed?

* Can you bias the query towards the documents the user has in their collection? (bookmarks, images, e-mail, word documents, video games, etc...)

Anonymous said...

Interesting summary. I am amused by Peter's comment that Google cares only about the current search terms of a user not necessarily their history. I think one reason for that is that companies like Yahoo and Google cant afford to store the data which at Yahoo gets to be over 10 Terabytes a day or something similar.

If storing that data was possible, they would definitely do it. I dont buy the "only current" argument. On the other hand, startups in the personalization search space can afford to do this cos they dont have the scale that Google or Yahoo has.

Anonymous said...

At this point, Bambi seemed to shift focus a bit and ask a bunch of questions about personalization and recommendations. Again, Bambi was not getting a lot of answers, but most of the answers she did get were fairly negative toward the idea of personalized search.

That seems kinda strange to me, because a few months later (Septembrish? Octobrish? I can't remember offhand) when Peter gave a talk at PARC, he was all over the idea of personalized search. He talked about the amazing things Google was doing toward search personalization. In fact, for years the storyline coming out of the Googleplex was that users not only do not want to do a lot of work, but that users will not do a lot of work. And personalization is the way to help them avoid doing more work (as opposed to the "tools" approach, i.e. relevance feedback and query expansion).

Does this mean Google is actually starting to change its mind, and go in more of the "tools" direction? If so, does that mean they are finding that users really are willing to do a little bit more work, when given the proper tools in the proper contexts? I've always thought that users really are willing to do more work: witness the fact that a single search session involves the user typing in the same query, with slight variations, 3 or 4 times. That in itself is a lot of work right there! So if you can give the user intelligent tools to help guide them, then you might narrow that down to 2 or 3 iterations.

Anonymous said...

Greg - thanks for sharing this information - these types of panels always provide some do insight into where people are thinking on an issue. I thing I would add (knowing you are on a path to personalized search ;-)...

People in general will take the path of least resistance - in all areas of their life - and using the internet is not different. Patterns (and habits) are developed over time and they are almost impossible to break (once entrenched).

Just because some "person" creates a way to do something that they believe is "so simple, everyone should be doing it this way" goes against basic human nature.

Great Post - thanks for sharing.