Social Search


Recently, a new class of search applications that support collaborative information seeking has emerged. In these systems, users work in small groups with a shared information need, rather than relying on large numbers of anonymous users with potentially diverging information needs. One clear way to distinguish different social search activities has been proposed by Colum Foley. In his PhD thesis, he characterizes search systems on two dimensions, “Sharing of Knowledge” and “Division of labor.”  Sharing of knowledge separates all social search systems from traditional single-user approaches, while division of labor separates social search from collaborative search.

In social search systems such as those described by Ed Chi in his blog post, knowledge is shared implicitly by aggregating the behaviors or opinions of many people. There isn’t much division of labor because each new user has to establish patterns of behavior similar to others’ to benefit from the others’ activities. Furthermore, each person has to vet the recommendations to make sure they are appropriate to that person’s information need. Collaborative search systems, on the other hand, can be characterized by high division of labor. Information found by one team member is made available to other team members without requiring additional work. Thus social search systems tend to be effective at improving retrieval of common documents (the “big head”), while collaborative search systems tend to be more effective at identifying novel or unusual information (the “long tail”).


  1. There are some interesting things in here that I hadn’t thought of before: that division of labor separates social search from collaborative search, for example. However, I think that “big head” is only one model of social search. Question-answering models (or social answering models, as Ed Chi calls them) are another. I think it’s worthwhile to distinguish “collaborative search” from “social search” since they do accomplish different things, but it’s still hard to wrap my head around the difference.

    One way I’ve been thinking about it recently is that your social search systems (above) are suited for individual users with personal requests for information. This would make aggregating social behaviors or collecting social answers fair game. Collaborative search systems are for groups of people with a shared goal. Would you agree with this?

  2. […] of Aardvark, but in the mean time wanted to highlight a post today by Gene Golovchinsky about social search. He in turn points to work  by Colum Foley and Ed Chi. Check it […]

  3. Brynn, the way I like to talk about it is in terms of “re-covery” vs. “dis-covery”. Social search is about re-covering something that someone else in your social network has already found — you personally just don’t know about it yet. Collaborative search is about dis-covering information that no one in your social network has yet found — but that is nevertheless still relevant to your information need.

    In social search, all the fancy algorithmics go toward the propogation of already-discovered relevant information. In collaborative search, all the fancy algorithmics go toward pointing out pathways that no one has yet examined yet, but that appear to be relevant.

    That’s why you have division of labor as a fundamental tenet of collaborative search. If someone in your social circle has already found something, that information should not appear in the search results, because it has already been found. It might flow to you through a secondary channel, such as a list of shared results or shared bookmarks. But there is no need to clutter one’s search results with information that the other person has already found. In social search, on the other hand, the whole goal is to clutter your search results with the items that your friends have already found or think are interesting. Social search is about getting everyone onto the same page, so to speak, by using those positive social connections to re-introduce the same pieces of information, over and over, to the various people in a social network.

    imho. :-)

  4. By the way, we’ve got a series of posts on the various dimensions of collaboration in search.. implicit vs. explicit, synchronous (symmetric) vs. asynchronous (asymmetric), etc. Once we finish describing all those dimensions, it should be more clear where and how we think social and collaborative search differ.

    Personally, this is not about either one being better than the other. It’s about understanding that each serves different purposes and has different advantages and disadvantages. Understanding where and how each is applicable is my goal.

  5. @Brynn Using one’s social network to answer questions (a la Aardvark, for example) is an interesting complementary mechanism to automated search. But unless there is a back-and-forth between the person asking the question and the person answering the question, you have delegation rather than collaboration.

  6. Another interesting aspect of social search is the role of network topology. I have some vested interest in this topic because this is what my Ph.D. thesis was going to be on, but it seems to me that different models of propagation and divisions of labor would be appropriate for different network situations. In other words, what might work well for one person in a particular network position, may not work for other people in other network positions.

  7. It sounds like different social network situations may reflect different roles in a collaboration, and therefore, as you suggest, may require different mechanisms to mediate collaboration.

  8. @Gene I agree that Aardvark may have limited usefulness without a back-and-forth, but I’m not sure I’d call it delegation. It’s social because you get an answer from another person. It is no collaborative (like you point out). But where would you rank it on your scale of social–collaborative?

  9. @Brynn My concern is with the breadth of the “social” term. Since almost everything we do is in some way connected to other people, we should be careful about scoping our terms. We should look at the entire human-computer system (that includes multiple humans and potentially multiple unrelated computer systems), and try to figure out which behaviors we are interested in modeling.

    If I ask a question through Aardvark or Twitter or some other network, and you send me answer, have we collaborated? In the sense that you took on my information need, yes; in the sense that we worked together on the problem, no. The “work together” aspect of collaboration would be met if we had some sort of dialog (whether computer-mediated or not) about the information need and the found information. Saying “Thanks” doesn’t seem to be sufficient.

    So I am taking the position that in the absence of true bi-directional communication, we don’t have collaboration. Do we have a “social” system? Broadly-speaking, yes, since more than one person is involved. But is that definition useful? Does it reveal something non-obvious about the system? I am not sure. That’s why I chose the term “delegation” to represent that class of human-human activity that allows one person to ask another one to do something in a more open-loop manner. But I agree that there is a gray area between the two.

  10. @Brynn @Gene

    As I pointed out in another post earlier this week, even Google’s standard web search is, in a sense, “collaborative”: If enough people type the same query, and then click the 3rd result, then by the time I issue that query, that 3rd result will have crept up in the rankings to 2nd or 1st. Google uses the historical mass actions of lots of people to influence future searches. By getting together and using the same search engine, all these people are, it could be said, collaborating with each other.

    However, the influence in that scenario is uni-directional — past to current. There is no bi-directionality in the system.

    Gene should have another post on this idea of bi-directionality next week, and then I will try and tie together these two dimensions (intent and synchronicity) shortly after that.

  11. @Brynn

    Regarding your first post, I think you have touched on what I think is the fundamental difference between these “social” search and the full collaborative search systems: The move from satisfying a single user’s information need to satisfying a group information need.

    I tend to describe these social search systems as asynchronous collaborative IR (or IS) systems. They are often characterised by an implicit collaboration over a long-ish time frame and involving a large group of people. Although they are collaborative in the sense that the IR has a fancy algorithm attached to it which incorporates other users’ actions into the ranking, the purpose of these systems is still the same as any other IR system – satisfying a single user’s information need.

    What makes the collaborative search systems we talk about different is that they represent a significant paradigmatic shift in how we approach the IR problem – we have moved from satisfying a single user’s information need to satisfying a group information need. These searches are often characterised by a more explicit collaboration, often over a shorter period and among a smaller, focused group of individuals. Therefore I think we have to examine how the IR process operates at a much lower/fundamental level than is needed for these “social” search systems.

    I really like the way you describe the collaborative search systems as finding new pathways. It’s a very interesting way of looking at the “whole being greater than the sum of its parts” phenomenon which we would like to exploit in collaborative search systems.

  12. […] week or so ago, we wrote a post on Social Search, and how (we believe) it is different from Collaborative Search.  We have also begun laying out a […]

  13. […] Gene Golovchinsky: Social Search […]

  14. […] is a newish entry into the social search arena that allows searchers to share search results with their peers through Twitter.  Rather than […]

  15. […] be held a few days later at CSCW.  For an interesting thread on the distinction between the two, please see another FXPAL post from March of last year. January 29th, 2010 | Category: General, Information Retrieval Foundations | Leave a […]

  16. […] a bunch of time on definitions, and we thought we’d jump in early. We’ve talked about social search before, but that was without reference to social […]

  17. […] off working alone, begrudging others scraps of information we uncover. Far from it! As we’ve argued in the past, working together to find information can be more effective and more efficient than […]

Comments are closed.