SMX East 2011: The Current State Of Personalized Search
This session, part of the Big Picture track, will look at personalization in its incarnations at Google and Bing, and suggests what marketers should do in a personalized search world.
Moderator: Danny Sullivan, Editor-in-Chief, Search Engine Land (@dannysullivan)
And for some more big picture stuff on the topic of personalized search, Eli Pariser’s keynote this morning was thought provoking and forward looking. Danny says that this follows up on Eli’s presentation and we’re going to check in on the current state of personalized search.
Jack Menzel, a Google product manager, points out that we all know what SMX is. Except if you’re in the UK, where it’s a motocross show, or a Samung camera, or a snow bike, or sports glasses… This is too good an example of an ambiguous query.
So what’s personalization? According to Eli it will end the world and trap us all in a bubble. But there’s a lot included in “personalized” search that’s not personal, but contextual. Language is imprecise. Personalization can help fill in the gaps:
Here are examples:
- Country – Taxes: the IRS vs. her Majesty’s revenues
- Recent searches – Amazon
- Interests – Dominion
- Preferred sites – scrabble dictionary
- Social endorsements – +1 shown in results
Filter Bubble: Personalization is working!
Perfect personalization + Totally passive users + Myopic focus on clicks = VERY BAD DISTOPIA
If we had perfect personalization, if we were to abandon free will, and if we assume that relevance is only based on clicks… this perfect storm would be bad.
- Personalization is far from perfect.
- Users have active info needs.
- Quality relevance is more than just clicks.
The moral of the story:
- Users maintain control over their data.
- Be transparent about ho the data is used.
- There is no canonical result set for a query.
Stefan Weitz takes the podium and says Jack’s lucky Eli’s not here. Eli’s bigger than him.
Deep Thoughts …
- Personalization is not just about re-ranking of results.
- They look at providing the highest value at the lowest cost. What’s the minimum processing power, or info required from user and session, for the greatest result?
- The 5 Ws apply: who where what, when, why
- Often really a proxy for intent.
- Selective personalization. Different models for different classes of queries, and within those classes, which queries react better to personalization. Overpersonalization goes the other way – there can be too much.
Personalization is the new reality.
When: What time is it?
Where: Carmen Sandiego
Why: So Meta
Why do we think these queries with these common characteristic end this way? Again, they look at them as a proxy for intent.
Personalized search has become just plain search. It includes:
- Presentation: how can they look different
- Context: previous behavior
- Dynamics: whats been hanging across the Web and what you’ve seen before
He does a query for solar panels and Rick Perry (separately) and can see his friends’ interactions with sites, directing his attention to them.
There are also differences in rendering. Results liked by friends on Facebook are moved up to page 1 from page 2 or 3.
Entity collapse is also happening. If friends are liking a brand on Facebook, a website on that topic might show up prominently in SERPs as they map that the topics are the same.
CIS is a common medical term, but there are many other definitions, and the phrase is quite ambiguous. If they see you going back to a certain page for an ambiguous query over and over, the result will go up in results. However, their application of this is done carefully. It’s restricted to queries that appear to be navigational in nature.
INTRODUCING: ADAPTIVE SEARCH
Rolling out right now!
It learns from your behaviors.
Search for [jsm] and Jimmy Swaggart Ministries comes up top. Do a search for [bootstrap method] first, and then [jsm] and reranking toward science/math/computers. It’s not just based on the last query – “it’s better”.
Presentation layers are top secret. How’s that for a teaser?
In response to Eli, what about serendipity? Bing has researched the topic: “From X-Rays to Silly Putty via Uranus: Serendipity and its Role in Web Search”. For every result they asked people to rank if the result was relevant and interesting. The point was trying to define what is serendipity anyway?
Look at everyone who did the searches, all their queries, all their ratings, their search history, etc. Can they find correlation between their ratings and their “filter bubble”. Things that were interesting but not relevant hd a higher score than things that were relevant but not interesting. People would rather get results that are interesting.
Danny asks if personalization means SEO is dead? Jack says it actually makes you job easier. You don’t have to be the “best” bus schedule in the world. What does that even mean? You just have to be most relevant for who’s looking for you. It is hard that it’s harder to measure. Danny points out, though, that’s only the case if you’re only measuring ranking.