Is Your Group Seeking Information? SearchTogether!

Publié juin 4, 2008

Partagez cette page

By Rob Knies

That long-awaited vacation is almost here—just you, your spouse, and your teenage son, all eager to embark on a leisurely adventure to sunny climes for some much-needed R&R.

But your planning has not kept pace with your anticipation. There is work to be done: hotels to book, flights to schedule, activities to consider. You need to research and collaborate to reach a decision. Time is growing short, though. How will it all get done?

Well, you could try SearchTogether, a research project from the Adaptive Systems and Interaction group at Microsoft Research Redmond. The team recently released a beta of SearchTogether, a free Internet Explorer plug-in that enables groups of people to collaborate on Web searches.

“SearchTogether is part of a larger project about collaborative search,” explains Meredith Ringel Morris, a researcher spearheading the project. “We’ve done surveys and interviews to learn more about situations in which people are collaborating on Web search, what types of tasks drive them to do this, and what their current practices are.

“We’ve been developing and testing prototypes for a bunch of those scenarios, and SearchTogether is for the scenario we call remote collaboration, where each participant has access to a computer.”

Search, of course, has become ubiquitous for enabling users to find Web content, but existing search engines have been designed for use by an individual. Search interfaces don’t support collaborative search. Collaborating on search generally means one person at a keyboard while another makes suggestions, or two people using instant messaging or a phone while each is viewing a Web browser. It can work, but it’s not optimal.

While using SearchTogether, though, users can collaborate locally or at different locations, working in tandem or at different times. As Morris notes, her research has demonstrated that the need to do so certainly exists.

Meredith Ringel Morris

“The big surprise has been how much people are trying to collaborate on Web search, given that it’s not currently supported,” she says. “In my survey, I asked people straight out, ‘Have you ever collaborated on a Web search?’ Over 50 percent of people said they had, which I found very surprising. Another 10 or 11 percent said they wanted to, they’d tried to, but they couldn’t really do it successfully because the tools weren’t right for it.

“If you asked people specific questions, like, ‘Have you ever stood behind someone while they searched the Web and suggested keywords that they should try?’ which I would consider an example of collaborating on search, nearly 90 percent of people said they had done specific things like that. People really want to do these collaborative behaviors. I think it’s good to think of ways to help them.”

In actuality, Morris has found herself needing to perform such collaborative searches.

“I’ve experienced a lot of situations where I’ve wanted to be able to work with other people when I was looking for things online,” she says. “For example, when I joined Microsoft last year, and my husband and I were moving to the [Redmond] area, we were searching online for things like furniture for our new home or purchasing a car, and those were complicated searches. There were a lot of different queries we had to issue and Web pages we had to look at to make those decisions.

“They were complex decisions that we both wanted to be involved in, and it was really difficult to do that. If I did some searching on my own—let’s say, for cars—and he did some on his own, there was a lot of redundant work. We would each end up finding the same things. It was hard for me to know what he had already found. That was part of the inspiration for getting started on this project.”

SearchTogether is designed for just such scenarios, although it is hardly limited to them. A rich set of integrated tools, it focuses on active collaboration among a small group of acquaintances who are working toward a shared goal, whether that be planning travel, making purchases, planning social events, researching medical conditions, or working together on a joint project. The technology captures and shares the collaborators’ combined knowledge, reduces duplicated effort, and removes the need to hover over somebody’s shoulder to be effective.

“We’re trying to think not just about browsing,” Morris says, “but about search as a first-class collaborative activity.”

It’s easy to try for yourself. You’ll need a Windows Live ID; if you don’t already have one, it’s fast and easy to obtain. The tool uses browser tabs, so you’ll need Internet Explorer 7, and you need to install the plug-in. From there, things are pretty intuitive, but a tutorial exists to walk you through all the functionality. You can invite anybody in your Live Messenger buddy list to collaborate, and, if need be, you’ll want to add them to your buddy list before you get started.

SearchTogether works both synchronously or asynchronously, meaning you can work together in real time, or you can work at different times and let the tool communicate the steps you have taken to your search partners.

“You could imagine a group of three people,” Morris says, “and two of us are online at the same time. We have a chat, an IM, about the task, and later, the third person logs in. It’s nice that the chat transcript is preserved along with the search results, because when the third person logs in, he or she can catch up on what discussions other people in the group already have had. The third person comes up to speed and keeps working.”

SearchTogether, featured in a paper presented in 2007 during the Association for Computing Machinery’s 20th Symposium on User Interface Software and Technology, offers users the ability to work in the search engine of their choice. A discussion forum is available to provide support and to accept suggestions.

Morris has a background in research on collaboration. Her Ph.D. thesis at Stanford was on interfaces for table-top computers, similar to Microsoft Surface.

“In general,” she recalls, “I was working on software and interaction techniques that would help groups of people work collaboratively around these types of devices. Collaboration is something that I’ve been very interested in. I moved from thinking about collaboration on tables to thinking about collaboration in a very different domain, which is fun.”

She got her first taste of the search domain in 2002, during an internship at Microsoft Research, working with Eric Horvitz, principal researcher and research-area manger of the Adaptive Systems and Interaction group, and Susan Dumais, a principal researcher in the same group.

That experience came in handy when she started thinking about collaborative search. Previous research had identified a common activity, informational search, in which a user is seeking information rather than seeking a single target site. This, Morris deduced, is the sort of rich search task that would benefit from collaboration.

In addition, earlier studies had observed that people frequently have to re-enter previously used queries to retrieve useful Web pages. That demonstrated the value of capturing search activity and making it available later for multiple users. And an examination of sense making—processing, organizing, and analyzing information—sparked thoughts of providing summary views of previous collaborative work, ratings and comments on discovered content, and the ability to discuss those discoveries via instant messaging.

“We’re really trying,” Morris says, “to go beyond existing work on collaborative browsing.”

In scoping such a project, Morris and Horvitz identified three challenges to surmount in order to make SearchTogether useful:

Awareness: The most valuable of the three. This is enabled by sharing group query histories and making them interactive, thereby reducing duplication. This process also can promote search learning, in which someone with rudimentary search skills can learn how to formulate more effective search queries by working with a person with more expertise. The ability to see other group members’ comments and ratings on specific search results also provides awareness about the group’s progress toward accomplishing their shared task.
Division of labor: SearchTogether integrates session-specific instant messaging into the search experience so that those working together on a search project don’t have to switch back and forth between a search client and Live Messenger to negotiate division of labor. Additionally, special collaborative query styles, such as “split searching,” provide automatic division of labor by partitioning a single set of search results among all team members. Sharing of queries and recommendations helps make the search effort productive.
Persistence: The group’s queries, comments, ratings, and chats are stored in a persistent search session, facilitating asynchronous communication and helping collaborators get up to speed.

“We made the decision,” Morris says, “that those were the important design criteria, based on the surveys we had done. Those emerged as the key challenges that people were facing in collaborating with current technology.”

Two significant enhancements provided by SearchTogether are the ability to rate and comment upon search results and the ability to split a search among all the members of a group.

Users can apply a “thumbs up” or “thumbs down” rating to any search result and share the rating among the group. The tool also enables comments to be appended to a rating. Comments also can be shared by themselves.

“I think that’s really the most important part of the tool,” Morris says. “When you give ratings or comments to a Web page, those get added to the summary of your collaborative search session. That summary is the product of the group’s collaboration. It’s a list of all the best content that people in the group have found, who found it, and what they thought about it. That’s the sort of thing you would synthesize to create your report or make your purchasing decision. The comments and ratings are fundamental to the creation of that summary.”

The ability to rate search results also has led to an interesting collaborative phenomenon.

“It’s interesting to see how people use those,” Morris says. “In our initial evaluations, people used the good ratings a lot. They wanted to mark stuff that’s good. They don’t really use the bad rating. But in a group context, the bad rating is valuable, because it helps other people in the group avoid wasting time going down a bad path. But because that’s very different from the way we’re used to searching, it’s hard to get used to the idea of warning people off as a way to save the group time, even though it doesn’t save you time. It’s a more experimental feature. I’m curious to see how people really use it.”

The Split Search functionality provides automatic division of labor. If a user selects that option, a subsequent search query’s top results are distributed evenly among members in the search session, reducing duplication of effort.

“Each has a non-overlapping subset of the results,” Morris explains. “That could be useful for several reasons. If the group’s goal is to find something quickly, it’s a divide-and-conquer approach. Everyone can work in parallel, and they know they’re not stepping on each other’s toes. There’s no redundancy.”

SearchTogether also features Multi-Engine Search, in which a query is sent to multiple search engines, with each participant in the session receiving a set of results from a particular engine. This increases the scope of the results and also enables their examination and analysis to be conducted in parallel.

“That’s a strategy we learned from our survey when we talked to people who did group searches,” Morris says. “We want to automate that process.”

All these techniques are an attempt to marry the process of collaborative search—query formulation and result exploration—with the products of the effort—comments, ratings, and shared summaries. Other features, such as the ability to peek at a colleague’s search results, the ability to follow a colleague’s search process, and the ability to participate in multiple collaborative efforts, make the tool even more valuable.

“One of the important next steps,” Morris says, “is in sense making. Once groups of people do collaborative Web search, they need to make sense of what they find, make their decision about what car to buy or what information to include in their final report. The ratings and comments feature is a first step toward that, but it doesn’t quite provide all the support people really need.”

Sharoda Paul, an intern from Penn State, will be helping to develop an extension to SearchTogether to add additional support.

“You could imagine,” Morris says, “that you could use the tool from start to finish, to gather your information and to help you produce your final outline that your group would use to make its report.”

Paul will be joining others who have worked with Morris on SearchTogether. In addition to Horvitz and Dumais, Piali Choudhury, Matt Maddin, and ThuVan Pham all helped create the Internet Explorer plug-in. And Jaime Teevan and Steve Bush have worked with Morris on “groupization,” which could be an advanced means to re-rank or to distribute search results among a set of collaborators.

“There are changes to underlying search-engine algorithms that could take advantage of the knowledge of a group,” Morris says. “Groupization takes the idea of personalization techniques for customizing a list of search results to an individual and customizes a list of search results to a group, based on information you know about each member of the group. You can either exploit similarities among all the group members, to bubble up search results that would be most relevant to the group as a whole. Or you could exploit differences among the group members, send portions of the search results to different people in the group based on what you can automatically determine is their specific area of expertise.”

CoSearch is another, related project, for co-located collaboration in which a number of people gather around one computer. That scenario occurs a lot in school settings, and the use of additional devices, such as cellphones, could augment the use of the shared computer. Intern Saleema Amershi, from the University of Washington, worked with Morris last summer to design, develop, and evaluate CoSearch.

That project, though, like SearchTogether, will require getting enough people to use it to enable the technology to achieve critical mass.

“You need a certain size of user base for the system to become valuable,” Morris notes. “The collaborative search tool is only valuable if your friends are also using it so you have people to collaborate with. That’s the biggest challenge to evaluate the success of SearchTogether’s features.

“A lot of individual people have been exploring it and were individually excited about it, but they have to convince their colleagues or classmates to try it with them. That’s the tricky part.”

Hence the beta. But regardless of how the project evolves, it’s clear that Morris is exploring a space that could prove enormously beneficial to groups of users who have a shared need for information.

“I think people haven’t really been thinking about collaboration as part of the search process,” Morris concludes. “People have been thinking about search as an individual activity. This work has helped change that kind of thought.

“I think we’ve shown some thought leadership, getting other academics and researchers to realize that this could be a fruitful area.”