The Pros & Cons Of Personalized Search
In the past two weeks I’ve had the opportunity to talk to both Marissa Mayer and Matt Cutts about the impact of personalization at Google. While the initial storm following Google’s announcement seems to be dying down somewhat, the ripple effects can still be felt throughout the SEM industry. In future columns I’ll probably be coming back to the personalization angle on a semi-regular basis because I really believe it’s fundamentally important to the next evolution of search. But in the next little while I’ll probably move on to other topics. Before I do move on, however, I did want to take a look at some of the criticism that has come out about personalization and provide my view on why personalization is the logical next step for search.
Crying wolf about personalization
To be honest, I’m surprised at the amount of push back to search engine personalization.
One of the biggest dissenters of personal search is Michael Gray. Both on his own blog and various comments posted on other blogs, he’s been a consistent critic of Google’s move towards personalization. Here’s just a sample of Michael’s arguments against personalization.
2+2 should always be 4 no matter where you are. It shouldn’t be 3.7 because in Brazil they like things slimmer and lighter and it shouldn’t be 4.2 in Italy because they have an aversion for waifishly thin numbers. It also shouldn’t be 3 for me because I’ve demonstrated a dislike for the number 4 for based on prior search history, but really like the number 3…
I think I get Michael’s point here but I disagree with the context. Two plus two equals four is a mathematically defined truth. It’s universal and unquestionable. But search is all about disambiguating intent from a few words formulated in your mind. There’s no universality, because the intent is solely yours. And that’s the problem the search engines are currently facing. For search to move to the next one level, disambiguation of individual intent is critical.
Semantic mapping primer
Anyone who’s heard me speak in the past has probably heard me talk about the concept of semantic mapping. For those of you who haven’t, here’s a 30 second explanation: When we launch a query on a search engine, we take a concept in our mind and try to distill it into a single query. Concepts are usually complex things, made up of a number of aspects. For example, let’s say I’m looking to replace my PDA. There are probably brands of PDA’s I’m familiar with and features that are important to me. I may be looking for independent reviews, and I may be looking for one that’s compatible with one particular data network provider. These and a number of other factors make up my concept. The various aspects of this concept can probably be translated into words, creating a number of labels for my concept. These words create a semantic map, a literal translation of my concept. But the semantic map may contain dozens, or hundreds of words and I’m certainly not going to take the time to type them all into a search engine. Remember, our average interaction with the search engine is some somewhere under eight seconds. So I distill this complex concept down into a single phrase. That’s all I give the search engine to work with.
The search engine’s job is to take that single phrase and try to provide me with the results that most closely match my semantic map. And now you see the problem that presents itself to the search engine. How can it possibly know all the words that are in my semantic map?
Matching the user’s intent
One way is to make the user do the heavy lifting. This could play out in a couple of different ways. On search engines as we currently know them, it probably means an iterative searching process. We try a phrase, see how close the matches are, and if we’re not satisfied that we’re getting the best match to our intent, we can continually tweak our phrases until the results look more relevant. But this is a time-consuming process. It is, however, the process we’ve become used to and we feel like we’re in control of it.
Another way the user could help the search engine disambiguate intent is though a new interface. One interesting approach to this is the one used by Quintura. If you want to see how semantic mapping looks in action take a look at Quintura’s interface. Quintura takes your original query and provides a semantic map based on the most likely matches to the phrase.
Example of Quintura “Cloud” for the query “pda”
You can click on words in the map and the query refines itself as you do so, narrowing the focus of the map so that it hopefully gets closer and closer to the concept in your head. Again this can be a time-consuming process and puts a lot of the onus on the user, not the background technology.
Disambiguation in the background
Ultimately, search has to solve the problem in a more transparent way. While there are some users who feel comfortable with putting the time in to refine their query and help the engine with the task of disambiguating intent, the majority of us don’t want to make that big an investment. The tool has to be much simpler and quicker to use. So that brings us back to the question, how do engines disambiguating intent without us giving it any more information to work with at the point of query? This is where personalization comes in. In the current online reality, there are really only a few places that the search engine can look to help define intent without depending on further information from the user:
- They can look at your past history and learn more about you by what you have already done
- They can look at the context of the task you’re currently engaged in, hoping that it will give some clues to what you’re looking for
- And finally, if they know something about you and your social, geographic and demographic cohort, the engine can hope that there is a similarity of thinking within that cohort, at least when it comes to common interests and intent
Each of these factors is being explored as a potential avenue to help with disambiguating intent. Right now, Google is put their eggs in the past online history basket, feeling that where you have been will provide the best signal to predict where you might want to go.
How the user wins with personalization
So how will personalization help the user experience? Try as I might, I couldn’t think of any online examples where personalization was really well integrated with the user experience currently. The best example I can come up with was Amazon’s personalization engine and that is just scratching the surface of what’s possible. So I had to reach out to a real-world analogy to try to understand this better:
Imagine you had two alternatives when you went shopping. One is a vast department store which carried everything you could possibly imagine. The store is usually well organized and everything is well labeled, but it’s up to you to navigate through the store and find what you’re looking for. There’s no one to really help you, although there are a number of useful signs to keep you pointed in the right direction.
The other option small store ran by a store owner who you deal with all the time; a store owner that knows you as a friend, knows your personal likes and dislikes and always seems to find just the right thing for you. In fact, all you have to do is say what you’re looking for in a few words, and based on how well he knows you, the store owner runs in the back and never fails to bring back exactly the right product, in the right size and the right color.
The first scenario describes where search is today. The second scenario describes where search wants to go in the future. As the scope of the Internet gets larger and larger, the need for personalization to bring it within our scope becomes more and more important. Search had a tough enough job when it was just trying to connect us with websites.
Now it’s trying to connect us with websites, local businesses, news sources, images, audio files, videos, and the list will continue to grow and grow. Search has to move beyond its current paradigm of one query and a list of links to websites. The challenge of disambiguation becomes exponentially harder as there are more potential matches with content. You have to narrow the odds on one side of the equation or the other. Search has no intention of restricting the focus of the content is indexing, so the only choice is to get better at determining intent with the user. That’s why personalization in some form is inevitable.
The user’s sticking points with personalization
Another source of criticism is how personalization may impact the credibility of search results. Once again a comment from Michael Gray:
Let’s take personalized SERP’s a bit farther, let’s imagine we have something like digital books that can rewrite themselves based on user preferences. Instead of Hermione Granger being a brown haired slightly bookish student at Hogwarts, she’s a buxom blonde in a mini-skirt because I’ve demonstrated a preference for that in the past. For someone else she’s a raven haired Gothic princess, for another she’s more of a debutante prom queen.
Here, I’m not sure the point that Michael is trying to get across. We have a totally different level of engagement with the book we’re reading than we do with the search engine. Our interaction with the book can be several hours in duration. Our interaction with any given search experience is probably a few seconds at most. The last thing we wanted to do is get to know the “character” of the search engine results page. Books depend on characters to carry the plot of the story. There is no plot to a search engine results page. It’s simply a presentation of options to match our intent. The overall integrity of the search engine results pages determined solely by its degree of relevancy to the user. Anything that improves that degree of relevancy makes the search results that more useful to the user. There’s no artistic integrity to maintain.
But Michael is not the only critic. Others worry about the potential impact of unknown algorithmic factors on the results that they could be seen on their search results page. This was a comment on my blog from Adam.
The biggest risk I see in personalized search is latency. When you return from your vacation, how long will it take before Google stops biasing results towards travel/Hawaii or perhaps thats a non-issue. I equate this to the impact of buying a gift of a childrens book on Amazon—all your recommendations going forward are tainted with that purchase.
Another concern is the pain with signing in and signing out and how Google knows who is who?
Okay, so I should log into my account, then when my wife wants to surf, she should log into her account. Then, my kids should do the same. Of course, I can just buy every one a computer. The truth is, until this process of identifying users becomes nearly automatic, I don’t see the Google method as adding a great deal of value. Steve Haar, Editor – Think About Search
Am I signed in or not?
With these two concerns, I think we are getting to the heart of where the potential problems from a user’s point of view may lie. What if our past online paths aren’t a good predictor of where we might go in the future? At this point Google is telling us that we always have the opportunity to toggle personalization off and on. The one issue I have is that a number of users don’t even know they’re logged on, despite Google’s claims that it’s easy to see. We just did an in-service session with a large client of ours where a couple members of our team met with a number of their client-side counterparts. While there, they talked a little bit about the possible impact of personalization on the client’s search visibility. The client frankly didn’t see that it was going to be that much of an impact because, in their words, “no one really signs into Google anyway, do they?” Our team pointed out to them that they were, in fact, signed in and that their personal search history was being compiled with each search they did. They were shocked to learn this.
In my conversation with Marissa Mayer she felt strongly that the indications on whether or not you were signed in were very clear to the user. I do disagree here. I know for a fact that once users become engaged with the task of searching, a look up in the upper right-hand corner to see if or not their e-mail address is showing is just not something that happens very often. Every heat map we’ve ever produced of the search results page shows virtually zero scanning up in this area.
So the fact is, for the majority of users who are signed in, they’ll forget that they are signed in. I’m acutely aware of personalization and I often have to double check to see whether or not I’m logged in when the results look little strange to me.
And if personalization does significantly improve the search experience, do we start developing split personalities where our at- home personas differ significantly from our at-work personas? Do we have to start keeping multiple profiles for Google? On multi-user computers, how does Google keep everyone straight? The last thing Google wants is to make signing in and signing a pain. They would much prefer that everyone gets signed in and then forgets about it.
How fast do we go down the personalization path?
Ironically, despite all the negative feedback, the biggest problem that Google is probably facing with personalization is not that it’s going too far, too fast, but rather that it’s not going far enough. Google is addressing the above user concerns by treading very carefully down the path of personalization. Right now, only one in five searches will be impacted by personalization and it will only lift two results into the top 10. In this way, it’s really not a major impact for most of us on the majority of our searches. Google’s cautious approach is definitely showing here. But for personalization to deliver the win that Google is banking on, it’s going to have to play a much bigger role on the search results page. Right now, personalization is only “turned on” by Google when there’s existing search history. So it’s the searches where you tend to be in familiar territory where personalization will have its greatest impact.
This has a somewhat humorous side effect for the SEOs of the world, because we all tend to do those impromptu searches where we see either how we or our clients rank for certain phrases. Of course, we also tend to click through whenever we see our site or our client site show up in the results. This creates a pattern in our search history which will naturally boost the position of those listings we clicked on. This results in the heady but temporary thrill of thinking that our clients have jumped dramatically in the rankings and are now showing in the top 10. Then we realize that we’re logged into personalized search and it’s our past search history that’s given these sites the boost, not the Google algorithm.
But doesn’t it make sense that personalization is most needed when we’re not in familiar territory? Isn’t the challenge of disambiguating intent tougher when the concepts in our heads are fuzzier and not as well defined? To me that’s where personalization could deliver the big win. When it can help me explore territory that I’m unfamiliar with more confidently and helps connect me to the sites I’m looking for, personalization becomes a huge user win. Personalization as its being currently implemented from Google will help most of us in the short term primarily with our navigational searches. These are the searches where we can use Google as a shortcut to get to sites quickly. In the past we may have had to scroll through a few pages of results to find the sites we’re looking for, but now personalization will help lift those sites closer to the top.
Finally, probably my favorite dissenting posted comment was one where I was accused of the complete ignorance of Chaos Theory. Guilty as charged.
One wonders where in the world people like Mr.Hotchkiss are coming from. Chaos theory which Mr. Hotchkiss apparently sees as something unique to “dissenters” really is rather beside the point.
(Funny, I don’t remember even mentioning Chaos Theory in any of my writings about search personalization)
It merely discloses that if Mr. Hotchkiss had a scintilla of knowledge about chaos theory as applied to information per se, he would understand that there is an inverse correlation between the uniformity and complexity of any system, including information systems, and the ability of such systems to produce reliable results.
Hmmm, I’m not sure what he’s saying but I don’t think he’s agreeing with me.
There’s a lot of unrealized potential still to be found in personalization and Google freely admits this. They realize that these are the first few steps on a long path towards a better user experience. And with any implementation of change, there is always a degree of nervousness around that change. Although the current paradigm of iterative searching where we use a query and refine it in order to get closer to the content we are looking for can be cumbersome and time-consuming in some cases, it does leave us in control. And it’s this lack of control that appears to be the biggest push back point against personalization. Many of us are just not at a point where we can trust technology to be able to interpret us as individuals. Be that as it may, I’m not sure we’ll have a choice.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.