Communicating about Collaboration: IntentWednesday, March 18th, 2009 by Jeremy Pickens
This is Part 2 of (at least) 5 in a series of posts about Collaborative Information Seeking. Part 1 is found here.
The most important dimension for distinguishing between various types of collaboration is the intent with which users of a system approach each other for the collaboration.
A bit of history
In early 2006 when we first began exploring our ideas around collaborative information seeking, we used the terms “intentional” versus “non-intentional” collaboration. Our 2007 summer intern, C. Shah, did some intensive reading in the CSCW literature and found that a more common set of terms in the computer-supported collaborative work community is “explicit” versus “implicit” collaboration; we switched to that terminology.
Collaboration and shared information need
Regardless of the terminology used, we approach this problem from the Information Retrieval (information seeking) perspective, rather than from the CSCW perspective. We therefore defined collaborative intentionality in terms of the core concept of the information seeking domain, i.e. the User Information Need. A user information need is a desire, whether fully articulated or not, to locate or obtain information.
When we view collaboration from the IR perspective, it becomes clear that collaboration does not merely refer to a decision to work together. Rather, it refers to a decision to work together on an information need. Users who have actively and consciously decided to work with each other on a specific information need are explicitly (intentionally) engaged in collaborative information seeking. A user whose information seeking activities are influenced (for the better, of course) by others, but without the user’s active involvement, is implicitly or non-intentionally collaborating with others.
Merrie Morris at MSR has created a system called SearchTogether. This is an example of explicitly collaborative search. You work with a known partner to find information that is jointly relevant to your shared need, such as planning a vacation, finding restaurants, and so on. Colum Foley has also developed numerous systems in which users actively work together on a shared information need, and our own work has explored this as well.
On the other hand, systems such as Amazon’s product recommendations create an environment based on implicit collaboration. If I buy flowers and chocolates, and you buy flowers and chocolates, and Sam buys flowers and chocolates, then when Tina buys flowers, the system presents Tina with an option to buy chocolates. Tina does not (necessarily) share the same information need as you, Sam, or I do, or even know who we are. But because there is a similar pattern of information seeking activity between Tina and the rest of us, the system connects her activity to our activity, and Tina finds herself affected by our actions.
Even systems such as Google’s standard web search are (to an extent) implicitly collaborative. If you, Sam and I all type the query [recyclable batteries] and then click the 3rd link in the results set, Google’s learning algorithms will start to realize that maybe the first two links really are not that relevant to the majority of its users, and will start to boost that 3rd link to a higher position in the list. So when Tina comes along and queries for [recyclable batteries], the same link that the rest of us clicked will have moved to the 2nd or even 1st position. While one typically does not think of Google standard web search as “social” in any way, it nevertheless exhibits properties of implicit collaboration.
There is a trickier example, a system where a more subtle distinction is required. Barry Smythe at University College Dublin has a number of papers on a system called I-SPY. The idea is to place a box at the center of a similar community of users, for example at the firewall of a small- to medium-sized business. This box then intercepts search queries and results, and does a sort of “deep packet inspection” in the information flowing between a user and an external search engine. It keeps a history of the other queries and clicks done by other users inside the firewall, and starts to “personalize” rankings based on the shared history and activities of this community.
So is I-SPY an example of explicit or implicit collaboration?
If we take the CSCW perspective, we might say that it is explicit collaboration. The community of users sharing this I-SPY box have actively and intentionally decided to work together. They’ve explicitly joined the same community. However, from an Information Retrieval or information seeking perspective, they do not necessarily share the same information need. Rather, even in a small company, community members have hundreds and even thousands of different information needs over the course of a day or week. One person at the company might be looking for documentation on an old product, another might be planning a trip to a conference, another might be using work time to search for fruit trees to buy from a local nursery. So while the users might share a lot of commonalities in the information that they are seeking, there is not an intentional, explicit action taken to collaborate on any particular information need. When needs do overlap, for example when more than one person is looking for that old product manual, the I-SPY system works better than, say, a general web search engine like Google, because the community’s needs are tighter and more homogeneous, so it is easier for the backend system to make the correct inferences to boost the proper information in the search results. However, at the end of the day, the system as a whole is still implicitly collaborative.
Task vs. trait
I am reminded of the distinction that Merrie Morris makes between task-based collaboration and trait-based collaboration. Task-based collaboration is, for the most part, very similar to our notion of (information need-oriented) explicit collaboration, whereas trait-based collaboration is similar to I-SPY and Amazon and Google implicit collaboration. Some of those communities are tighter than others, some share more traits that others. But all are, ultimately, implicit. (Merrie may correct me on this, but I think both of our ways of describing this particular dimension seem quite similar.)
Why does this matter?
So the real question is: Why does this matter? Why even draw a distinction between explicit and implicit collaboration? How does it help us solve user information need problems?
We will explore this question in greater detail in upcoming posts, but the short answer is that in the explicit case where the users declare their shared information need, you don’t need to infer it as you would in the implicit situation.