What email address or phone number would you like to use to sign in to Docs.com?
If you already have an account that you use with Office or other Microsoft services, enter it here.
Or sign in with:
Signing in allows you to download and like content, and it provides the authors analytical data about your interactions with their content.
Embed code for: CHI2004 Point to Point.docx
Select a size
Point to Point: Exploring Overlapping Social Networks
Redmond, WA, USA
UW CSE, Microsoft Research
Seattle, WA, USA
Point to Point provides an automated interactive social map that allows users to find out how they are connected to other people or projects within a closed corporate social network. Point to Point was developed on the assumption that knowledge management is a social activity, where relational ties between people serve as primary conduits for the transfer of information. We expected that people would be more likely to seek out or share information with others near them in their social network. We found in a user study of Point to Point that the network visualization had a meaningful impact on users' ratings of others' similarity to self, and that when using Point to Point to make decisions about meeting potential contacts distance in social network, social status, and informal social connections had the most impact on contact choices.
social networks, visualization, information seeking, expertise location, CSCW, computer-supported cooperative work, information interfaces and presentation
Knowledge management is a social activity. Within organizations individuals are a prime source of knowledge [1, 20], and knowledge management has as much to do with locating who knows what as with managing the knowledge itself [2, 17]. Research has shown that people actively use direct word of mouth to acquire the information they seek [6, 14]. As much as people care about the information, they care about developing collaborative relationships with individuals throughout the corporation .
In the current paper we describe a project, Point to Point, which provides an interactive social map that allows users to find out how they are connected to other people or projects within a closed social network, such as a corporate environment. Point to Point facilitates the process of using people to help find or meet others by showing users intermediary people between themselves and others they might want to talk to or know more about. Point to Point provides a network visualization that highlights the overlapping social context between two people. In a study of Point to Point we explored how the similarity information encapsulated in the network visualization has a meaningful impact on people’s likelihood of seeking out or sharing information with others.
People commonly use the Internet and email to seek out people who are knowledge experts or project contacts . In addition several knowledge management systems have explored how to support people’s tendency to seek out knowledge through people. Referral Web, for example, uses co-occurrence of names in documents on the web to develop connections between people and then makes referrals through a chain of such connections . The Expertise Recommender  allows users to employ a social network filter which sorts recommendations based on social distance in a network.
Social network visualizations allow users to infer information about social structures, such as the centrality of a person in a network, the similarity between any two people, and clusters of people . People, their relationships, network clusters and network patterns may be visualized for the user with a variety of methods [4, 5, 11, 19], the most common of which have been graph visualizations where people are represented by nodes in a web of interconnecting lines. In the field of computer mediated-communication, network visualizations have been used in the domain of representing semantic similarity  and message adjacency [5, 7, 19] to allow people to navigate through information spaces.
Although people are increasingly exploiting relationships in social networks to provide structure and navigation paths through information spaces, online systems that provide visual overviews of social networks and tools for navigating through them have had limited success. Not only do they tend to be overly complex and difficult to interpret, when users are exploring an unknown social space the relational information about who’s connected to whom loses much of its meaning.
We expect that rather than wanting to see people in their own social context, users care most about how others relate to the self. Relative distance from each other in the social network, relative social status in the network, and the presence of overlapping people in their social contexts will have a meaningful impact on the likelihood of seeking out or sharing information with someone. For example a person is probably more likely to seek out information from a friend of a friend than from a stranger. The key goal of the Point to Point visualization is to highlight for users how other people’s social contexts relate with their own.
McDonald  conducted a qualitative study of whether expertise recommendations that incorporate social network information such as social distance from the user could improve the quality of recommendations. He found that users reported some reservations about using social network similarity for expertise recommendations, because the best experts may be suppressed. We expect, however, that people’s explicit goals of finding the most expert person may contradict people’s implicit goals of finding someone they are comfortable interacting with. For example people may be more comfortable seeking out information from a lower status person, while more likely to provide information for a higher status person, irrespective of level of expertise. Thus an important goal of our study of Point to Point is to explore what features of social networks have a meaningful impact on people’s decision-making about whom they are likely to share information with.
Point to Point
Point to Point builds on past work, Connections  and the Personal Map , which provide automated interactive social maps that allow people to navigate for information about people using social ties. In Connections, people are placed in the center of the map in the context of their social networks, where relationships between people are indicated by lines. Data for measuring similarities between people is collected from the company’s employee directory service which includes information about internal email distribution list memberships and management relationships. It is important to note that people may easily create and join distribution lists, so they are a reasonable approximation of informal, dynamic groups .
Point to Point adds to the Connections project the ability for users to answer the questions “how am I connected to person X”? Point to Point derives connections between people by integrating standard hierarchical information found in organizational charts with the more informal group information found in email distribution lists. It determines the most important overlapping people by sorting the product of network similarity values of the two points. Point to Point then provides a unique graphical visualization of the two points and their overlapping social networks (including both management structure and informal distribution list groups).
As in standard graph visualizations Point to Point represents people as nodes in a map of interconnecting lines. However Point to Point imposes additional constraints. See Figure 1 for an overview of the underlying structure. The visualization places the user in the upper left corner of the screen (Point A, Figure 1), and places the people most similar to the user around him or her. The person of interest is placed in the bottom right of the screen (Point B, Figure 1), similarly embedded in his or her own social context. The management chain connecting the two points is rendered in the upper right hand corner of the screen, and the overlapping social context is placed in between the two points.
Figure 1: Underlying structure of the Point to Point visualization. Each person is an end point, connected to each other formally through the management chain, and connected to each other informally through email groups.
Point to Point has five main components: collecting data, measuring similarity from the data, selecting people to place on the map, visualizing the network, and providing features that allow the user to interact with the network. Our prototype is built using C# on .NET, SQL Server, and Active Directory.
Public email distribution lists and management relationship data are collected from the corporate member directory and stored in SQL Server.
Point to Point measures the similarity between people based on how often they co-occur in email distribution lists of fifty or less people. Conceptually, a matrix is constructed that contains the similarity values for all pairs of people. The connection, or similarity, between two people is measured by the extent to which they tend to occur in the same distribution lists. If two people are very similar, their similarity will approach one. If two people are very dissimilar, their similarity will be zero. The impact of each distribution list on the similarity measure is adjusted depending on the size of the distribution list, such that co-memberships in smaller distribution lists would lead to people being more similar. We normalize the similarity value between each pair of people by the number of distribution lists in which each person is a member separately, so that their similarity to other people is smaller if they are each in a large number of groups overall.
Point to Point integrates hierarchical information found in organizational charts with the more informal social group information found in email distribution lists. Toward this end, the algorithm builds on the similarity measure and selects three specific sets of people to place on the map.
First, the map contains the sets of people most similar to both end points separately (the context people). This provides the social context of each endpoint person and is usually representative of the end point’s working groups. We select a fixed number (10) of top ranked people according to their similarity to the end point person. The inclusion of this group can be turned off by the user.
Second, the map contains the set of all people in the management chains of both end points.
Finally, the map contains the set of people most similar to both end points (the shared people). Using the terminology of shortest paths, we define the length of a path between two people as the product of the similarity measures of the connections between each pair of people in the path. We make all of our shortest paths involve just one person other than the end points (a friend of a friend connection). The connecting person is placed on the map, and we select a fixed number (10) of top ranked people according to the path length.
Visualizing Overlapping Social Networks
Point to Point uses the spring model of Kamada and Kawai  to minimize the error between the similarity between every pair of people and the distances between that pair within the visualization. The two end points are constrained to opposite corners of the screen (See Figure 2), the top of the organization chart is constrained to the upper right hand corner, and the spring layout algorithm places every other person such that the screen distance appropriately represents the similarity among all pairs of people.
The shared people naturally move to the center of the screen between the two endpoints, and the context people naturally move closer to the endpoint to which they are most similar. The names of representative email distribution lists are drawn in the background for the two end point social contexts (see “group1” and group2” in Figure 2) and the overlapping people (see “group3”) to explain the relationships among those smaller groups of entities. The representative email distribution list for a group of people is the smallest distribution list that contains most of the members of the group of people. The context groups for the end point person tend to have a representative email distribution list that follows their most direct work group. The shared group tends to have a representative email distribution list that helps clarify to the user why the shared people were selected as the connection between the endpoints.
Figure 2: The Point to Point visualization. PointA and PointB are the two end points. The management chain is drawn with a darker green line.
A line is drawn between a pair of people if the similarity meets a particular tolerance, if there is a management relationship between the two people, or if the one of the pair of people is one of the top three strongest connections between the end point people.
If the two end points have no connection other than through their management chain the map is empty in the middle. See Figure 3A. If two people have a few informal connections through their email distribution lists, the middle of the map becomes populated. See Figure 3B. If two people have entirely overlapping social contexts, usually because they belong to the same work group, all of the context people cluster in the middle of the map. See Figure 3C. Thus it only takes a glance at the Point to Point visualization to gain a sense of the similarity between any two people.
Interaction with the network
Users may learn more about people by interacting with the network. When the user hovers over a person, a tool tip presents name, title, group, number of direct reports, and a size-sorted list of email distribution lists. When the user hovers over a line between two people, a tool tip shows management relationships and shared email distribution lists. See Figure 4. Right clicking on a person presents the user with options to search for the user on the Internet and inspect the user’s data from other organizational databases within the corporation.
Figure 3: The Point to Point visualization with varying levels of overlap between the two end points.
The primary scenario for Point to Point is helping the user understand the corporate social network in the context of a particular person or project. Toward this end, the application provides several methods for exploring the social network, many of which leverage popular web browsing idioms. The navigation transitions are smoothly animated allowing the user to visually comprehend the dynamic nature of the network.
There is a text entry box for each end point to search for people or projects. If the search succeeds with a specific person, that person is placed on the appropriate corner of the map. If the user searches for a distribution list, the distribution list member that is highest within the organizational chart is used to connect the group to the social network.
The user may activate the “Point to Me” command on any query person. As its name suggests, this option allows the user to switch the map such that they are one end point person and the query person is the other end point person. The user may navigate through the history of their exploration through back, forward, and home buttons. The user may also drag and drop any name from the map to generate a new map involving that person.
Figure 4: Mousing over a person activates a tool tip that shows information about that person (left). Mouse over connecting lines shows how two people are connected (right).
We conducted a user study to test whether the social network information in the Point to Point visualization would provide meaningful information to the user. We expected that viewing a person in his or her social context, and seeing how that person relates to the self, would have an impact on the impressions formed about that person. Seeing people in their social context should help people form more positive and decisive impressions of that person’s trustworthiness, competence, and so forth. We further expected that when people use Point to Point to explore their social networks, they will make faster decisions about whom they might want to talk to about a project, and they will be more confident that they found the right person.
The social network information found in the visualization should have an impact on who people are likely to meet to seek out or share information when given a choice. We predict that people will be more likely seek out meetings or accept meetings from others who are closer in the network, have higher management status, and have some connection to the user through the email distribution lists. We further expected that status information would interact with who initiated the meeting, such that people will be more likely to request a meeting from a person who has lower organizational status but accept a meeting from someone with a higher organizational status.
17 company employees (7 female and 10 male) completed the study in exchange for a coffee coupon. Participants were on average 33 years of age, had worked at the company for an average of 5.4 years, and had either program management positions (8 people) or software development and testing positions (7 people). 9 of the participants were not managers and 8 were managers of at least two people (M = 3.8).
Participants completed the study within their own offices, following a series of structured tasks at the direction of the experimenter. The first task was to complete a questionnaire that asked for basic demographic information, and then how participants currently learned about other people in their organization. Participants then installed the Point to Point application on their primary desktop computer. They were instructed in its use, and given an opportunity to explore the application.
In order to explore the impact the Point to Point map had on impression formation, the first structured task had participants look up six people using Point to Point, and then rate each person on a number of dimensions. For half of the participants, the map connections were turned on, and for half the participants, the map connections were turned off. In both conditions the participant could view the basic profile information available.
Participants then completed a series of timed tasks. For the first task, they were instructed to look up five people, imagining that they had all sent requests for a meeting but they only had time to accept one meeting request. They were instructed to select the one person out of the five they would most prefer to meet. For the second task, participants looked up three groups, and were instructed to find for each group the one person they would talk to. Participants completed each of these tasks both using Point to Point, and their standard company directory system. The stimuli names and groups were randomly selected from all names within the corporation. Both order of method and set of stimuli names and groups were counterbalanced.
In order to assess whether variability in the Point to Point map had an impact on whom people might decide to meet, we had participants complete a series of 16 choices, deciding between two people whom they would most likely want to meet. For eight of the choices they were instructed to imagine they would be sending a meeting request because they wanted to learn more about that person’s project. For eight of the choices they were instructed to imagine they had received meeting requests from two people. The 32 stimuli names were randomly selected from two sets of names within the corporation—managers and individual contributors—to ensure some choice pairs had differences in management status. We also ensured that they were co-located with the study participants. The order of the two sets of eight pairs (sending vs. receiving) and the stimuli names for each set were counterbalanced.
Participants then completed a questionnaire in which they indicated what features impacted their choices, and then finally they provided general feedback for the application.
In the preliminary questionnaire, we first asked people about their use of email distribution lists. People reported on average actively using 10.3 email distribution lists, through which they interacted with on average 65 people. Thus distribution lists should provide an approximation of informal connections in the corporate network.
We then asked participants how they currently seek out information about unknown people. Over 50% of participants reported meeting someone at the company they did not already know once a week or more. Participants listed the three most common methods they used, and rated each method by how often they used them. 100% of the people said they used the company’s email address book to find a person, which has a profile with contact information for each employee, with job title and group, and manager and direct reports. See Table 1.
Table 1: How people currently find information about people and groups within their corporation.
People may explore management structure around a person in the address book by opening the profiles of that person’s manager and direct reports. Secondary sources for learning about unknown people were searches of company internal web pages, and word of mouth.
When trying to learn more about projects or groups, people primarily used web searches or word of mouth. Although people reported having somewhat adequate knowledge of people and projects within the company for their job (M = 4.7, SD = .86, where 1 = not at all and 7 = extremely so), they also reported on average being somewhat frustrated when trying to learn more about people and projects (M = 4.7, SD = 1.5).
Impression Formation Task
To explore the impact of the social network visualization on impression formation we had people look up six people within Point to Point. Half of the people had the map on, and half had the map off. Each person was rated on a number of dimensions, including likeability, competence, expertise, and similarity to the self. We also asked to what extent they would accept a meeting request from the person or hire the person. Responses for names already known by the participant were filtered out, and then ratings were aggregated for each item across stimuli people for each participant. We expected that seeing people in their social context would increase ratings of trustworthiness and likeability, and improve the likelihood of agreeing to a meeting with that person. However between subjects analyses of variance show no main effect differences of presence of network visualization across all of the dimensions except similarity. Participants on average reported that people were less similar in the presence of the network (M = 2.3, SD = .70) than in the absence of the network (M = 3.4, SD = .70), F(1, 15) = 9.6, p < .01.
Timed Search Tasks
For the timed task, people were instructed to select one person out of five they might care to meet. We expected that people would be faster at making their decisions when using Point to Point to compare people than when using the standard email address book profiles to compare people. However we found no difference in time across the conditions (Ms = 216s and 218s, F(1, 13) = .11, ns). We then asked how difficult was the task to complete and how confident were they that they made the right choice. We found no differences depending on condition. People reported finding the task to be not difficult for both Point to Point (M = 3.5) and the email address condition (M = 3.6) across both conditions and were somewhat neutral in confidence in both the Point to Point (M = 4.3) and email address book conditions (M = 4.1). For the second timed task, participants were instructed to find the appropriate contact for each of three groups. They completed the task once using Point to Point, and once using their regular methods. Again, we did not find a difference in time to find contact people between the Point to Point (M = 190s) and email address book (191s) conditions, F(1, 13) = .05, ns. Neither did we find a main effect difference in self-reported task difficulty or confidence across conditions. However we did find a confidence by order interaction (F(1, 15) = 4.5, p = .05), such that people were much more confident when using Point to Point after using the email address book than when using the email address book after using Point to Point. See Figure 5.
Figure 5: People were especially confident they made the right choices when they completed the task using Point to Point after completing the task using their regular method.
We explored how variability in the network information in Point to Point might impact knowledge sharing by having participants indicate for 16 pairs of people whom they would prefer to meet. For each stimuli person we calculated organizational distance to and number of common people with the study participant. Organizational distance between two people is the number of nodes up and down the management chain between them. We also coded people’s job type and then classified each stimuli person as having the same job type as the participant or not.
On average, the stimuli names had 8.7 (SD = 17.3) reports. They had an organizational distance from the study participants of 12 nodes (SD = 1.4). 30% of the stimuli people had the same job type as the study participants. Stimuli people had on average 1.7 (SD = 1.96) people in common with the study participants, with 39% having at least one person in common in their social networks. Organizational distance was negatively correlated with the number of common people (r = -.46, p < .01) and job similarity was positively correlated with number of common people (r = .35, p < .06). Organizational distance was also negatively correlated with number of reports (r = -.45, p < .01) such that the further apart stimuli are from participants in the management chain, the more likely they are to not have any reports.
To test the impact of each similarity measure on participant choices, we performed a multivariate analysis of variance on the four measures, with stimuli pairs’ choice status (chosen vs. not chosen) entered as a repeated measure and the choice task instructions (send vs. receive) entered as a repeated measure. We found that the chosen people had on average a higher relative status (F(1, 16) = 18.59, p < .001), were closer in the organizational management chain (F(1, 16) = 44.20, p < .001), and more overlapping people (F(1, 16) = 4.20, p < .06). See Figure 6.
Figure 6: The people chosen by participants tended to have higher status, be closer in the organizational management chain, and have more overlapping people in their social networks.
Chosen people did not overall tend to be more similar in job type. In examining the impact of task type on choices, we found that whether participants were imagining they were sending a meeting request or accepting a meeting request had a meaningful impact, such that people were especially more likely to accept a meeting request from a high status person (F(1, 16) = 4.65, p < .05) (see Figure 7), and were especially likely to accept a meeting request from someone who had a similar job type (F(1, 16) = 5.34, p < .04) (see Figure 8).
Figure 7: People were especially likely to accept a meeting request from a high status person.
Figure 8: People were more likely to accept a meeting request from someone with a similar job type.
Once participants completed the choice task, we asked them to indicate what features about a person were important to them when making their decisions. People reported that job type and job status were most important. Familiarity with the person or the person’s team was not as important, and number of reports and nearness in the corporation were rated as the least important features.
Figure 9: Self-reported importance of various features on choices about whom participants would meet (where 1 = not at all and 7 = extremely so).
These self reported responses somewhat contradict our findings from the choice task, in that number of reports, distance in the organization, and overlapping people were stronger indicators of choice than similarity of job type.
Once participants completed the structured tasks, we asked them to provide more general feedback on the Point to Point application. They on average reported liking it (M = 5.0, SD = 1.15), but found it somewhat confusing (M = 4.4, SD = 1.18). Even so, they reported that they would actually use Point to Point (M = 4.6, SD = 1.97), with 53% saying they preferred it to the standard email address book. When asked why they liked Point to Point, people said they found it valuable to see the relationships between people.
having a visual map is important to helping me understand how I relate to others
visual, clearly see distance relationships and find connections
People thought however that the user interface could be simplified. The lines and names tended to obscure each other and were hard to read. Several participants requested that information such as job title be on the surface of the user interface. When asked under what circumstances they might actually use Point to Point, people generally said in preparation for meeting someone.
Point to Point was designed to facilitate the process of sharing information in a corporate social network by showing users the overlapping social context people between themselves and others.
We found in a user study of Point to Point that the network visualization had a meaningful impact on users' ratings of others' similarity to self, such that people appeared less similar with the presence of the visualization. Although there were no differences in the ease with which people were able to find others when using Point to Point and using standard employee company directory, they were somewhat more confident in the choices they made.
When using Point to Point to make decisions about meeting potential contacts, distance in social network, social status, and informal social connections had the most impact on contact choices. People generally were more likely to choose to meet people who had higher status, were closer through the management chain, and had people in common. Status and similarity of job types had more of an impact when accepting meeting requests than when sending meeting requests, suggesting people are more likely to help high status people and people with similar jobs. Although people reported using similarity in job type to make their decisions, job similarity had no impact on actual choices. These results suggest that people attend to social contextual information such as relative status more than they realize.
Our findings highlight that it is important to incorporate social network, social status, and job role information into user interfaces that are geared towards helping people find others to share information with in a corporate social network.
Thanks to the members of our research group, study participants, and paper reviewers without whose help this project would never have been completed.
Ackerman, M. (1998). Augmenting Organizational Memory: A Field Study of Answer Garden. In ACM Transactions on Information Systems, 16, 3, 203-224.
Alavi, M. and Leidner D. (1999). Knowledge Management Systems: Issues, Challengs, and Benefits. Communications of AIS, vol 1.
Aldenderfer, M. and Blashfield, R. (1974). Cluster Analysis. Iowa City: Sage Publications.
Card, S., Mackinlay, J., and Scheiderman, B (eds.). (1999). Readings in Information Visualization: Using Vision to Think. San Francisco: Morgan Kaufman Publishers.
Donath, J., Karahalio, K., and Viegas, F. (1999). Visualizing Conversation. In Proceedings of HICSS-32, Hawaii.
Ehrlick, K. and Cash, D. (1994). Turning Information into Knowledge: Information Finding as a Collaborative Activity. In Digital Libraries ’94, 119-125.
Eick, S. and Wills, G. (1993). Navigating Large Networks with Hierarchies. Journal of Systems and Software, 204-210.
Farnham, S., Turski, A., Portnoy, W., & Davis, J. (2002). Connections: Exploring Who Knows Whom through Social Networks. Paper presented at HCIC 2002, Winter Park, Colorado.
Farnham, S., Portnoy, W., Turski, A., Cheng, L., Vronay, D. (2003). Personal Map: Automatically Modeling the User's Online Social Network. In Proceedings of Interact 2003, Zurich.
Erickson, T., and Kellogg, W. A. (2000). Social Translucence: An Approach to Designing Systems that Support Social Processes. ACM Transactions on Computer-Human Interaction, 7, 1, 59-83.
Freeman, L. C. Visualizing Social Networks. Journal of Social Structure, 1, 2000.
Kamada, T., & Kawai, S. (1989). An Algorithm for drawing general unidirected graphs. Information Processing Letters, 31, p. 7-15.
Kautz, H., Selman B., and Shah, M. (1997). Referral Web: Combining Social Networks and Collaborative Filtering. Communications of the ACM, 40, 3.
Kraut, R., Galegher, J., and Edigo, C. (1990). Intellectual Teamwork: Social and Technological Bases for Cooperative Work. Lawrence Erlbaum, Hillsdale, NJ.
McDonald, D.W. (2003).
http://www.ischool.washington.edu/mcdonald/papers/McDonald.CHI03.pdfRecommending Collaboration with Social Networks: A Comparative Evaluation. Proceedings of the ACM 2003 Conference on Human Factors in Computing Systems (CHI'03), April 2003.
McDonald, D. and Ackerman, M. (2000). Expertise Recommender: A Flexible Recommendation System and Architecture. In Proceedings of CSCW, December 2000, Philadelphia.
Nardi, B., Whittaker, S., & Schwarz, H. (2002). It's Not What You Know, It's Who You Know: Work in the Information Age. First Monday, 5, 5.
Sack, W. (2000). Discourse Diagrams: Interface Design for Very Large-Scale Conversations. In Proceedings of HICSS-33, Hawaii.
Smith, M. and Fiore, A. (2001) Visualization Components for Persistent Conversations. In Proceedings of CHI 2001, Seattle.
Stewart, T. (1997). Intellectual Capital: The New Wealth of Organizations. Doubleday, New York, NY.
Wasserman, S., & Faust, K. (1994). Social Network Analysis: Methods and Applications. Cambridge, UK: Cambridge University Press.
1eet people who had higher status, were closer through the management chain, and had people in common. Status and similarity of job types had more of an impact when accepting meeting requests than when sending meeting requests, suggesting people are more likely to help high status people and people with similar jobs. Although people reported using similarity in job type to make their decisions, job similarity had no impact on actual choices. These results suggest that people attend to social contextual information such as relative status more than they realize.
Farnham, S., Portnoy, W., Turski, A., Cheng, L., Vr