Social attention analysis and management in online communities
The objective of this research was to investigate the use of Information and Communication Technologies to support people in analyzing and managing their attention at a social level. People are being overwhelmed by solicitations and opportunities to engage into a social exchange but they have little means about how to deal effectively with this new level of interaction. I present several direction of research I investigated at INSEAD during an european research project Atgentive (http://www.atgentive.com) .
Most of the following work have not been finished and published due to a lack of funding / end of the EU project. However I think there are some interesting concepts/direction to consolidate.
An attention-based Ranking Model for social media
Internet increases people’s social capital through increasing connections among friends/professional people. Associated to this change, a new behavior described by L.Stone as continual partial attention comes up: people attempt to stay partially but continuously aware about the activity within their networks. A typical situation for a user in a rich social environment is to decide from a list of incoming messages which ones are the most important for his limited attention capacity (e.g. limited time to read), keeping him aware of unexpected events and avoiding attentional demands according to his or her current interests. In this paper we propose an attention aware system inspired by the visual attention model of J.M Wolfe where visual stimuli/signals are changed in social stimuli (solicitations of other persons) for our concern.
Maisonneuve, N. (2007). Application of a simple visual attention model to the communication overload problem. In Workshop at UBICOMP 2007 , 9th International Conference on Ubiquitous Computing (Vol. 33)
Analysis of the Attention Alignment
I also started to investigate new indicators to observe collective and individual attention and their underlying relationships.
The objective this work was to 1/ find new indicators to understand the orientation of the community’s attention in a given interval of time in 3 different spaces:
- Resources space (users’ profile, messages): Which resources got the most attention? A list of resources that attracted the most the community’s attention (viewed or created) during an interval of time.
- Semantic space: which tag got the most attention? (=the sum of audiences for each resources related to a given tag for an interval of time).
- Social space: Which member got the most attention?(the sum of the audiences for each resources related to a member for an interval of time)
2/For each space, we also would like find an indicator to evaluate the alignment of any user’s attention with her/his community’s one i.e. the similarity of the orientation between his/her attention and the community’s one. Then This evaluation has two subgoals: a) to provide to the user a clear picture of his/her orientation according to the global community’s orientation (Has the user also seen the most popular resources?) (meta-cognition). b) to provide to the user a set of resources to read if he/she desires to improve an alignment about a given resource , tag or user.
Social Attention Profile
I also investigated new indicators to formalize a user’s attention profile according to 2 types of attention:
- the attention the network has for a member (who pays attention to me)
- the attention the member have for his or her network (the people to whom I pay attention
we can quantifiy the attention intensity a person A has for a person (or a topic or a task) B by observing the user’s A activity. the intensity of the A’s attention for B , for a given interval of time, is the numbers of actions done by A related to B during this interval (e.g. time interval = 1 week in our simulation).In a virtual community context, the actions are: “reading”, “posting”. (NB: we can define more precisely these actions e.g. “posting” could be “creating a new thread” or “responding to a previous message”). When A reads a message of B , or about B, A pays implicitly or explicitly attention to B.
Attention Attraction Intensity
We call every action dedicated to attract the other’s attention “attention attraction signal” (nimbus): the userA’s attention has been attracted by the attention attraction signal of userB.For instance, in a virtual community context, If A reads a message of B , it’s probably because B has just posted a message in the community platform. So B sent a kind of signal (the message) willing to attract the attention of the other members about something. Thus we can quantify the intensity of the user’s attention attraction for an given interval of time as the number of attraction signals (e.g. posted messages) sent during this interval.
Analyzing Attention received vs attention given
1/ This attention received to the user from the whole community could be defined as the user’s impact or influence on the community/network. This notion of impact can quantify by two complementary factors:
- Attention Intensity (strengh): for each member, we measure the intensity of the attention dedicated to the user. The cumulative intensity might be interpreted as the global impact i.e. the global intensity of the attention receveid from the whole community.
- Attention Focus (audience): while the intensity measures the strength of the impact, it doesn’t measure the audience level i.e. the number of people paying attention to the user. In fact the user could receive a high level of attention only by few members who seem to be very interested by the user. To better understand the notion of impact, we measure also the evolution of the user’s audience level i.e. the number of people paying attention to him or her. NB: we could also represent this two factors in a 3D graph (axis: time, members, attention intensity).
- charisma or social isolation When we observe the attention received by the community to the user we can’t precisely know if its intensity is normal i.e. related to the amount of posted message by the user (i.e. the strengh of the attraction signal). We then computer a ratio attention intensity/attraction intensity to remove this biais and get a better perception of the impact of the user, revealing respectively an “inattentive” trend of the community for this user (less attention has been paid despite of lots of posted msgs) or a growing interest for the user (more attention has been paid to the user despite of a low effort in attention attraction).
2/ We also measure the intensity of the user’s attention to each member over time. The attention dedicated to members is also quantify by the same two factors:
- Attention intensity (interest): we measure the intensity of the user’s attention for each member of his or her community. In a community / network context, The intenser is the attention to a specific member , the more interested the user seems to be to him or her. (and vice versa, the more interested the user is for a topic or a person , the intenser the user’s attention should be to it.)
- Attention focus (dispersion/openness): We measure also the focus, or the reverse, the dispersion (or diversity it’s depends on how we see this behavior) , i.e. the number of people to whom the user pay attention over time, to understand the management of his attention allocation. e.g last week, I tightly collaborated with only few people on a specific issue, while 2 weeks ago, I discussed about a general issue and pay attention to much more people
- social diversity (according to the expertise of the members: are the people who paid attention to me the same background)
In this Applet, each colored line represents the evolution of the attention’s intensity of a given member over time (unit = number of messages [posted by the user] read by the member). There is one color for each member. To know which member is associated with a given line, you can move your mouse over that line, it will be hightlighted and the member’s name will appear.
Furthermore a cumulative view (the sum of the intensity) is possible as shown in the picture below. The user can select this view by clicking the checkbox “Cumulative view” (label 6)
At the top , we see the graphic about the user’s impact. In adding the attention attraction intensity (bottom figure, in gray) we can know if this impact is normal or not according to the attraction intensity.
The attention focus graphic represents the amount of people associated to the user’s attention (attention given or received). i.e. the number of people paying attention to the user over time (focus unit= number of people) or for the graphic 5 , the number of people to whom the user paid attention over time.
Agents Supporting Attention Management
The objective of this chapter it consists in the adaptation to a social context, of a general model for supporting attention that was proposed by Roda and Nabeth (2008) and that relies on supporting attention at four levels:perception, deliberation, operation and meta-cognition. This chapter also presents how the support of social attention has been operationalised with the design of an attention aware social platform AtGentNet, and tested for supporting interactions of communities of learners and professionals.
- T. Nabeth, N. Maisonneuve (2009) “Managing Attention in the Social Web: The AtGentNet Approach”,In C. Roda, human attention in digital environments. Cambridge University Press.
- T. Nabeth, H. Karlsson, A.A. Angehrn, N. Maisonneuve (2008); A Social Network Platform for Vocational Learning in the ITM Worldwide Network; IST Africa 2008, Windhoek, Namibia