The following is the transcript of an interview of Tim Mayer, who is responsible for the product direction of Yahoo! Search Technology.
Tim brings over 10 years of search experience to Yahoo! with previous roles at Overture, FAST Search & Transfer, Inktomi and Thomson Corp. Tim was previously Vice President of Web Search Products at Overture where he led the product direction of the web search product as well as managing the major web search affiliate partner relationships. Prior to Overture, Tim was Vice President and General Manager of the FAST Web Search division including the core search technology and the alltheweb.com search destination.
Before FAST, Tim spent two years leading product management and product marketing in the web search division at Inktomi. Before entering the web search area, Mayer worked at Thomson & Thomson as the project manager of the SAEGIS and NameStake search platforms.
Mayer received an MBA from Babson College, Wellesley, Mass., and a BA in English Literature from Hobart College, Geneva, New York.
Eric Enge: One of the big things going on in search these days is the move towards personalization. Can you talk a little bit about Yahoo’s basic viewpoint and approach with personalization?
Tim Mayer: The angle that we’ve taken has been focused on social search. We have acquired some of the best social media companies, such as del.icio.us, and Flickr, with del.icio.us being a social bookmarking service, and Flickr being a community where you share very high quality photos. We have a new service called Upcoming which is local event service that allows you to know about events that are happening in your area, or events that your friends are attending, and it allows people to connect.
We have also been built market leading services such as Yahoo! Answers which enables people to explicitly share content and knowledge with others. We see these social services as visibly improving the quality of the search experience, and there is a lot of metadata being generated by these services, about what’s good, or what’s interesting to particular groups of people. We feel that by using this data we can provide a better search experience to people.
Eric Enge: Can you expand upon the notion of looking at trends or preferences of groups of people?
Tim Mayer: Yes, there are subjective queries for example where I want to find an Italian restaurant to go to tonight in San Francisco. If I do a search, I am going to get a listing of a bunch of sites that have lists of restaurants or I might get a few restaurants in there. But, that’s just giving you the information right now. What you really need is you need the information, but you also need some guidance or some trusted source which maybe a friend that you trust, that knows a lot about cuisine or it could be an expert on restaurants.
Some people look at Zagat obviously as an expert on restaurants, but there are other people out there that would be glad to provide you their input as well. We are looking at marrying the information that comes back with the insights of people to give people the answer to those types of questions, such as a good restaurant that matches your preferences or your style. And, if you have that personal connection with someone and you know this person has similar taste to you, and likes things that you like, then you are more likely to give people what they want.
For this example, it’s very hard to look in the review mirror, because looking at your past searches is not really going to help you. But, looking at what other people have thought were good Italian restaurants would really help you discover some interesting possibilities to try that evening.
Eric Enge: Right. You could pull the answer from Yahoo Answers for example.
Tim Mayer: Right. Or you could use the Italian restaurant, San Francisco tag in del.icio.us. You could search on that tag and get some pretty interesting results.
In addition, a lot of people take pictures of food at restaurants on Flickr, so having people taking pictures of the dishes at these particular restaurants would give you a pretty good idea if you would want to go there or not to.
Eric Enge: You would also get some idea of the atmosphere at the same time.
Tim Mayer: Right. For that particular example a lot of these services have some value to add in helping people make the decision of which place they should go. If you see one of your friends is taking a lot of pictures of this restaurant, and has been to it numerous times, and you think they have similar tastes as you do, you would probably give that a shot.
Eric Enge: Have you all begun to integrate these kinds of results into Web Search already?
Tim Mayer: We have had a few experiments in the past in integrating this into the search results, but we are continuing to try and strike the right balance. Right now there is a lot of content from Yahoo! Answers, all the search engines are finding Yahoo! Answers’ content very useful to users, and they are showing up in Yahoo! results as well as the results of our competition. We are also using del.icio.us and Flickr information. There is a lot of rich data now being provided on the search result pages, and I think this sort of data provides a lot of opportunities for some of that integration in the future.
In addition, a couple of years ago we launched a product called Mindset which is really about allowing you to alter the commercial intent of your query.
Eric Enge: Right. That was a slider bar that allowed you to set how commercial you wanted your results to be.
Tim Mayer: It’s very much of an experiment, but if it’s any type of commercial thing that you are looking for, you can increase that emphasis using the slider. You might want something that’s not completely biased by SEO, or marketers. So, if I am buying a new kitchen table, a desk, a lamp or some furniture for my house, I am able to tap in to either people that had a strong reputation in furniture or friends that are going to save pictures or tag it on del.icio.us.
That tends to be a lot more valuable then the results right now which are really based on commercial factors, or SEO, and is SEO the best for desks? You’ve got marketers telling you what’s cool rather than actual people that you trust. By searching on these types of sites, I get much more interesting types of items than doing web search.
Eric Enge: What about other kinds of inputs like recent search history, or geo-locating the user by IP address?
Tim Mayer: The location of the user is something that’s been done for a long time. When you look at Canada versus the US versus the UK, you are going to get all different results. That broad scale personalization has been going on for a long time, but that’s not getting different results for that same person. A different person in the same zip code would probably get the same results.
If you search on “Italian restaurants” and we know that you are in zip code 94025, you will get the same result as someone else in that zip code. Note that in our shortcuts we recently launched movies, a shortcut which is fairly interesting. It takes into account all the different types of intent a user might have, it allows you to view the trailer. It allows you to see local show times, so it’s giving you show times from your location as well as reviews and other things. These get personalized for you assuming we know your IP address or zip code in the Yahoo system.
Eric Enge: I just did a search on Italian restaurants and I get the local data of course. What about things like recent search history?
Tim Mayer: That’s not something we are currently using. In general, we are going towards the social route and I think if we do experimentation and find the other things on significantly improving search results, then we would probably move forward with them. But, I think from Yahoo’s perspective it’s very important to be very transparent to the users about what information we are collecting and what we are using it for, and allowing users to get in and out that experience.
You can turn your personalization on and off, so it doesn’t feel like a spooky experience how did the search engine know that I was looking for that?
Eric Enge: Right. What about the scope of the changes?
Tim Mayer: del.icio.us and Flickr and Yahoo! Answers are all growing extremely quickly. The more the data grows the more it can actually improve web search.
Eric Enge: Not too long ago the world was ten blue links, right. And, they all came from web search more or less, and that’s begun to evolve obviously, and shortcuts’ is one example of that. But, as you personalize what I see versus what you see may well be different. Is it possible that half the things I’ll see are different than what you see?
Tim Mayer: I think it’s really going to come down to measuring quality, and understanding if it’s really improving quality. You tune it to where it’s optimal. Some of our competition that’s been going towards the personalization angle has been limiting the personalized results to just two. One of the challenges is that when you are in a specific session, if you are doing personalization based which is session based, is understanding the task that someone is doing. When a person is searching for a Jaguar, they may have been looking for information about the football team. When I know they are in a sports mode, I might show them all football information, for all 10 results, but it’s very difficult to tell when someone switches mode and moves to a new mode.
And so if you, someone did another query and they put it in “Chiefs”, you would assume that they were still looking for football information, but they could have changed their mode of searching, and they may now be looking for something else. Therefore, if you returned ten results for football you would be completely wrong.
That’s one of the reasons people are being careful with personalization because if you are too heavy handed, you can get 10 results that are completely off base, which is obviously not a great experience to the user.
Eric Enge: The need for diversified results didn’t go away just because you had some information.
Tim Mayer: Yes. If you are a hundred percent sure that someone is in the mode of searching football, then you should show 10 results related to football, but there is always going to be that edge case where the user has changed modes. It’s actually larger than an edge case these days, because of the amount of times people just do searches on lots of different things.
They might be looking for information for a particular product, such as a new stereo, or a new amplifier, and then they realize that the Red Sox are, ant want do a search on the score. Then they might go back to searching for the amplifier. I think people move from one task to the other very quickly.
The other common thing with personalization is if you are doing it based on what people previously purchased it sounds great, but there are a lot of times when people are searching for a gift. If you buy your uncle a bunch of country music CDs, sites like Amazon will assume that you like it and offer you country music products over and over again, but it’s not really something you want.
Our goal is to help people discover new information and new content, which people will find delightful, and having people vouch or recommend these things on our social properties.
Eric Enge: Right. So, there is no need to turn off personalization in Yahoo then.
Tim Mayer: That’s correct. We are not doing any session based personalization based on the tracking of their behavior. When people ask how do they turn personalization off, I always say that an easy way to do that is to come search at Yahoo.
Eric Enge: It keeps Yahoo free of the problems relation to personally identifiable information problems. Are the things that you are doing with social search ultimately going to make the job of Spammers more difficult?
Tim Mayer: The important thing around social content is understanding reputation. Understanding how much to trust a given contributor is critical, and if reputations are linked correctly, it’s going to be hard for Spammers. But in any situation around search the stakes are very high, and it’s always been an arms race, where people come up with new techniques to try and artificially boost the ranking of specific URL’s or sites, and search engines work very hard to neutralize these techniques.
Eric Enge: It was Tomi Poutanen of Yahoo! who explained to me is that if you are on del.icio.us or in Yahoo Answers, and you succeed Spamming one particular thing, your gain isn’t quite the same as it is in web search. This by itself is a deterrent to Spamming.
Let’s talk a little bit about universal search, what’s the Yahoo approach to that type of search?
Tim Mayer: I look at universal search as more of framework than a lot of people who see universal search as the integration of images, news and video. I think it’s really a framework of integrating different types of vertical content be it multimedia content like images, or music, or video, or other verticals like local news, or it could be blog search, or any vertical that might best answer a particular question.
I think if you look at user behavior, most people come to the Yahoo homepage or the toolbar, and see a search box, and they type in a query. The user does not do any preprocessing and select a particular search engines, such as image search, or video search. They just type it in and, it it’s an image search these days, they expect images to show up at the top of the search results. A lot of people would call that image search, even though people in the industry would say an image search is when you go to the image tab, and you do a search. So, universal search gives access to a lot of information to all these people that they didn’t previously have access to.
Eric Enge: Do you have some examples of things that you are doing with this now?
Tim Mayer: I think of the universal search presentation in three stages. One was about optimizing content which is really about just optimizing the best algorithmic result and that was the main focus for search for a long time. In the past few years a lot of people have focused on shortcuts, which are really high confidence matches, when you think you have the answer to a specific query and you put the answer right up top.
I think the third stage is really understanding the user intent better, and integrating vertical content into the main line search experience, where you need to understand which verticals are relevant to that specific core of the user’s intent. What is the relative importance of each vertical as well as how much of each vertical should I show?
We have been doing that for paid results in terms of understanding how commercial is this query, and how many paid links should we be showing to the user. We have been doing this for a while.
Eric Enge: Do you have an example of such an area that I can look at?
Tim Mayer: We have recently a music artists shortcut where you type in a band name, and you get videos you can watch, and you can listen to their music. For example, type in the Grateful Dead, and you can listen to the music, or watch a music video, right there.
We have movies too, where you can watch the movie. We also have a lot of local ones with maps that are very good. We also have shortcuts for consumer electronics, the NFL, and medical search queries as well. For example, try a search for “Alzheimer’s”?
Eric Enge: I looked at the shortcuts for NFL players, and I remember you can type in a player’s name and you will get their picture and key data right in the search results.
Tim Mayer: That worked, and we’ve launched a bunch of things around this, and we will be launching a lot more things in the future.
Eric Enge: How big in an impact do you think these things have on relevance?
Tim Mayer: Previously search engines were focused on getting the user off the search page as quickly as possible to a website to get their answer. Now, search engines, for certain queries, will provide the answer right on the search page. One example I like to use is that of re-gripping golf clubs. That’s something that textual search results just wouldn’t provide the answer, because you really need to see someone doing it.
This is the only way you would get a good idea of how you would actually re-grip them, so a video is the perfect answer to that. If you can provide that on the search page, that’s going to be far more relevant. If someone is looking for a song, being able to listen to the song right there is what the user wants.
I think that’s really the goal is getting people the answer and the content they want as quickly as possible. Sometimes that will be on the search page, and sometimes that will be off on a particular site. I think that’s one of the big changes.
Eric Enge: Right. In terms of the approach you are using, when you have results that you think might be the best match for the query like with the band example, you are presenting that data right up top on the presumption that it’s the most relevant result, versus this notion that Google has of interspersing the stuff through their results.
Tim Mayer: I think it’s just an evolution. I think that there is a time and a place for both of these. To put it up at the top, you have to be extremely confident, and I think when you are not confident, it’s much better to be interleaving it within the results.
It’s not a typical Yahoo shortcut. We give you the official site, and then allow you to listen to music and watch videos. We have high thresholds for interaction, and we test these with users where we have launched them. We definitely try and understand when people are searching for artists. Then we try and provide all the different main activities that they would want from that particular shortcut.
Eric Enge: Right. Expanding on the band example, do you think that the fact that people see a good high quality video match for their query will cause more people to be aware of the ability to access videos online?
Tim Mayer: Yes. A lot of people still don’t know about a lot of the features in the search engine. What’s nice about the multimedia content is it looks very different and the search page looks a lot more engaging and it really changes the user behavior and how people engage with the search page. It really changes the way it looks.
Eric Enge: We might imagine an eye tracking analysis which shows a very different sort of behavior on that kind of a page.
Tim Mayer: That’s right.
Eric Enge: Do you think that the presence of this stuff will cause more people to explore going to the vertical properties, such as image search and video search?
Tim Mayer: Yes, it’s going to drive more awareness of different search verticals, because you are typically going to get one item or a few items and you’ll be able to dive in and click on a link that will send you into the vertical experience.
Where you can see more, but I think the access point, the main access point will continue to be the main search box. I don’t see more users wanting to take extra time and preprocess their queries to decide what vertical search tool to use.
Eric Enge: They want you to figure that out for them.
Tim Mayer: I think so in most cases.
Eric Enge: Right. What about the paid link debates? Google takes a lot of heat on this topic, what’s Yahoo’s stand?
Tim Mayer: I think as a search engine we don’t endorse techniques to manipulate how sites rank. Paid links are one of these techniques, and there are a lot of techniques people use to try to influence their rankings. I don’t single out this one in particular versus others, but paid links are an example of a technique that is artificial. What we want is that when you link to a site, you are vouching for it’s quality or that it’s interesting for a specific topic relevant to your website or from the article that you are linking from. If there is some payment makes this completely changes the value of the link for any search engine algorithm.
So, like the many other techniques that people spamming search engines use, we seek to neutralize it.
Eric Enge: One of the things that is a new theme from the people who are involved in this debate is the search engine should come up with another algorithm so that you are not so dependent on link analysis. I think Danny Sullivan made that statement in the search engine Q&A about links session, which turned into another paid link debate. My take on this is that if it were easy to find another mechanism that made paid links a non issue, the search engines have probably already would have done it. Am I wrong about that?
Tim Mayer: No, I think we are constantly evolving our algorithm. We constantly make changes to improve the quality of the experience, and there are quality signals that worked very well for a certain time, and then people start abusing them. Then, we have to move to new quality signals. We are constantly evolving, and the people that are spamming are constantly evolving too, that’s just how the industry is working. We are constantly evolving, but search engines have been using links for a long time.
Eric Enge: Right. While other signals are emerging, links remain an important part of the picture. As far as I can tell, it’s not something likely to change in near future. What role they play may change, but it’s not like a new algorithm is going to drop in out of the sky.
Tim Mayer: Well, you never know, but we don’t talk about how we use links. We are constantly working on addressing quality signals that are being over optimized, such as paid links. We try to find them, discover them and neutralize them.
Eric Enge: Thanks for taking the time, it was great.
Tim Mayer: Thanks Eric, I enjoyed it.