Paul Resnick

Bonds and Identity: Navigating the Tension Between Attachment to Topic and Attachment to People in Online Conversation Spaces

Biography
Paul Resnick is a Professor at the University of Michigan School of Information. He previously worked as a researcher at AT&T Labs and AT&T Bell Labs, and as an Assistant Professor at the MIT Sloan School of Management. He received the master’s and Ph.D. degrees in Electrical Engineering and Computer Science from MIT, and a bachelor’s degree in mathematics from the University of Michigan.

Professor Resnick’s research focuses on SocioTechnical Capital, productive social relations that are enabled by the ongoing use of information and communication technology. His current projects include analyzing and designing reputation systems that help maintain trust among strangers on-line, convening technologies, and ride share coordination services.

Resnick was a pioneer in the field of recommender systems (sometimes called collaborative filtering or social filtering). Recommender systems guide people to interesting materials based on recommendations from other people. He chaired the PICS Interest Group at MIT’s World Wide Web Consortium and was one of the main authors of the PICS technical specifications. PICS, the Platform for Internet Content Selection, provides a common infrastructure for the creation of labeling systems, and filtering software based on those labels.

Network Computing Magazine named Resnick one of its 25 Network Technology Drivers for 1996 (September 1996, issue). His articles have appeared in Scientific American, Wired, Communications of the ACM, and many other publications.

Position Paper
There often seems to be a tension in online spaces between staying “on topic” and having a warm, friendly, personal atmosphere. As part of the CommunityLab project, Bob Kraut, Sara Kiesler and I have been working towards articulating why this tension seems to be so pervasive, we’ve been working with John Riedl and Loren Terveen on the design of a new system, called ConversationLens, that may help to reduce that tension.

The analysis draws on two theories from social psychology about how people can feel attached to groups: interpersonal bonds and social identity. The theories translate into the setting of online conversation spaces as follows. First, people can be attracted to the space because they like the people there (bonds). Second, they can be attracted because they like what people talk about (identity). Wide ranging conversation and personally revealing conversation will generally help form bonds with other individuals (both revealing information about yourself to others, and learning more about them increases your liking of them). Thus, off-topic conversation will generally be appreciated when it’s with people you have formed bonds with or when it leads to the formation of bonds. But it’s not possible to form bonds with too many people. So, in a large conversation space, most people would prefer that others who they do not bond with stick to the topic that defines the space.

As long as everyone sees all the conversation, I think that the tension about staying on-topic will be inherent. However, the network model of conversation that is becoming apparent in the blogosphere suggests the possibility that one need not have tidy group boundaries around a conversation, with everyone in the group seeing all the messages. Even when conversation does take place within a bounded community, we were inspired by the idea that not everyone needs to receive all the messages.

The ConversationLens system will attempt to guide people to the conversation that is either on topics they care about, or that involves people they have bonded with. One challenge is to infer a person’s topical interests and friends based on his or her reading (and possibly rating) behavior. A second challenge is to automatically classify messages and message threads based on their topical and friendship relevance to each person. Finally, there is the challenge of presenting relevant messages with sufficient context of surrounding messages so that the conversation can remain coherent.

Back to Social Computing Symposium 2005