Star Search: NASA's Relaunch

Not surprisingly, has a staggering amount of information for students, space enthusiasts and plain old Capt. Kirk wannabes like us. With 25 million unique visitors last year across 600 million page views, the site's relaunch in December was a major undertaking. Brian Dunbar, Internet Services Manager, NASA, worked with eTouch system and its managing principal, public services sector, Eashwer Srinivason. In order to build tag clouds and more audience-driven search results, they relied on the Baynote system of using crowd behavior to improve relevance. Here, Dunbar and Srinivason discuss how a site as vast as the cosmos itself is made manageable to a variety of audiences.

Behavioral Insider: How did you go about creating information architecture for this redesign?

Brian Dunbar:
We look at the traffic stats, where people are going on the site, click paths, and combine that with our customer satisfaction survey. We did a major restructuring of the information architecture to where you go to the main page now and see ten buttons on the top right. And then what we wanted from Baynote is some of that crowd behavior -- not just tracking where people go, but try to put some weight behind it, whether they are going here and it seems to be useful -- or that they are going here and finding it a dead end and leaving. That is what we're experimenting with.

BI: Where is that most visible on the site?

The tag cloud up front is generated by the Baynote software. You see what are the most popular search terms. Perennially the most popular ones are NASA TV, Moon, and Mars. When you click on those, you get the Baynote search results that tell you this is what most people seem to be clicking on for this particular search term.

BI: What have you divined by watching this in action?

We are seeing a very large increase in traffic to the site, about the same level of customer satisfaction, and also a real significant drop in number of page views. We are still trying to formulate a theory for what explains all of that.

BI: How exactly is crowd behavior being applied, perhaps to streamline the experience?

Eashwar Srinivason:
Let's say I am searching for the word Shuttle on the NASA site. The search engine you are using will by default have its own relevancy, and it's going to come back with those results. Typically the user might say he is most interested in, say, the fifth result, and if more and more people always go to the fifth result, then Baynote will automatically make that fifth result the first result in the list.

We are using two primary content guides, one as the tag cloud on the home page and the other for "most watched" videos on the video player. We are also using Baynote Observer to observe as users move from one page to the other. These observations tie back into the searches that get performed, which we call social search. Google is our search engine, and results are fed to Baynote, which then applies its crowd wisdom. Social search comes up with the results that most people have been looking for.

BI: On the front page, then, you have two different navigational schemes, one traditional and one crowd driven.

They actually serve two different sets of users. The information architecture and the navigation structure from my perspective serves the largest chunk of the users, the people who are coming for that commonly sought content. But we have very spiky traffic driven by when we are in the news, and so it is an effort to handle that huge base of people coming for the latest cool pictures and multimedia.

We want search especially to address folks looking for information from the long tail. We have been seeing those long-tail pages going up as an overall percentage of our entrance pages. More people are clicking onto them as a first page than they were in the previous design. The homepage as an entrance page is down. Search is becoming more important as a primary navigation, especially among younger users, so it is going to become more important for people who are looking for the most common content, and we will have to address that.

BI: How effective is the tag cloud as a mode of navigation versus the architecture?

We are looking at how the tag cloud is a snapshot starting from the time Baynote began its observations, so we knew that those terms then would become fairly fixed. Mars is always going to be big, and Space Shuttle.

We need to understand better, how does the newness factor play into it? If we have a spike today on new Moon or black hole images, where does that play into it and how do we take care of that? If everyone is looking for something but they are only looking for it for two or three days, shouldn't it still pop up on the tag cloud after [traffic] goes away? There is a lot we are still evaluating.

BI: Any speculation about why you see page views go down?

If I have my way, it would be because people are finding the information very quickly and getting off the site. If that is the case, I would expect to see an increase in our customer satisfaction ratings, and we haven't. They are fairly steady.

BI: But you are not ad-supported so fewer page views and broader audience could be seen as a good thing.

I would think it would be a good outcome. If it were a negative because people don't like the new design, I would expect a drop in customer satisfaction. I certainly would not expect to see a near doubling in the number of visits to the site, which is what we are approaching. Also, the number of search queries per user is actually down. But the hypothesizing is that people have really started to bookmark some of those pages or they are getting them through RSS feeds.

BI: What's next then?

We have talked internally about focusing a search strategy across NASA's very decentralized Web space. There are something like 3,000 domains. On there are 60,000 Web pages. But NASA has over 4 million Web pages and there is no organizational strategy. So I am pleased with how we started out, but there is still a lot of work that has to be done on our end as content managers to really improve the search results for people.



Next story loading loading..