Nicolas Previous Works Meeting Turin Mai08

31
Brief presentation of my previous work Nicolas Maisonneuve – Associate Researcher at Sony CSL Blog: http://nico.maisonneuve.free.fr Tagora Project – Mai 2008

description

 

Transcript of Nicolas Previous Works Meeting Turin Mai08

Page 1: Nicolas Previous Works Meeting Turin Mai08

Brief presentation of my previous work

Nicolas Maisonneuve – Associate Researcher at Sony CSLBlog: http://nico.maisonneuve.free.fr

Tagora Project – Mai 2008

Page 2: Nicolas Previous Works Meeting Turin Mai08

My current & new Interest

Human/Organization sciences + computer science

(e.g. the new interdisciplinary initiative: MIT Colllective Intelligence Centre )

Page 3: Nicolas Previous Works Meeting Turin Mai08

My previous project: Atgentive

European project: 2-year STREP 2005-2007

Purpose: “The investigation about the use of artificial agents to support the management of the attention in an online community”

Experimentation: Private platform, small group ( ~20 persons)

(on going) personal research: - Social translucence design- Community orientation & user’s Alignment - Attention based recommendation engine- Simple attention analysis (Other’s attention to me vs. my

attention to the others)- Little experience about attention spamming- Dynamic Feeds for granular problem of the activity

Page 4: Nicolas Previous Works Meeting Turin Mai08

Social translucence design

Page 5: Nicolas Previous Works Meeting Turin Mai08

Social translucence design

Goal: Design knowledge management system to emphasize the social aspect

(Erickson et al. (2002). “Social translucence: designing social infrastructures that make collective activity visible”)

What emphasizing the social aspect important?User’s perception change: From a repository of resources to a

community of peopleEmergence of social mechanisms(reputation, trust , affinity) Increasing the participation (main problem in Knowledge

Management)

Page 6: Nicolas Previous Works Meeting Turin Mai08

Social translucence design1/2 – Displaying social interesting events + Behavior recommendations

Interventions done by external agents that:Input: Observation the community’s activityDetect a given activity pattern:- Basic pattern: (e.g. the creation of a response) - More complex pattern: (a burst of activity)

Output: a Feed/list of personalized recommendations for each user

Interventions done by external agents that:Input: Observation the community’s activityDetect a given activity pattern:- Basic pattern: (e.g. the creation of a response) - More complex pattern: (a burst of activity)

Output: a Feed/list of personalized recommendations for each user

Objective: To help the user to better perceive/understand what happens in his/her community

(c.f FaceBook MiniFeed but before them)

Objective: To help the user to better perceive/understand what happens in his/her community

(c.f FaceBook MiniFeed but before them)

Page 7: Nicolas Previous Works Meeting Turin Mai08

Social translucence design1/2 – Displaying social interesting events + Behavior recommendations

Patterns about the user’s state

1)New user: "I can give you a presentation of the community platform, complete your profile“ + notifying the community about a new user

2) Inattentive user: Emailing inattentive user with an activity report to attract their attention

Patterns about the user’s state

1)New user: "I can give you a presentation of the community platform, complete your profile“ + notifying the community about a new user

2) Inattentive user: Emailing inattentive user with an activity report to attract their attention

Patterns about Activity related to your resources

3) “UserA viewed your profil” or “UserB responded to your message“

4) Burst of collective activity: “Your profile/message was viewed by a high number of people (12 member(s)) [..]compared to the normal audience”

5) Special interest : “2 members UserA, UserB viewed your profile more frequently than the others”

Patterns about Activity related to your resources

3) “UserA viewed your profil” or “UserB responded to your message“

4) Burst of collective activity: “Your profile/message was viewed by a high number of people (12 member(s)) [..]compared to the normal audience”

5) Special interest : “2 members UserA, UserB viewed your profile more frequently than the others”

Page 8: Nicolas Previous Works Meeting Turin Mai08

Social translucence design1/2 - Stream of social interesting events (aka Facebook)

Patterns about resource’s lifecyle

6) Related to the message’s deadlineFor the inattentive members, few days before: “you haven't viewed The message XXX closed to its deadline, maybe you should read it”

Patterns about resource’s lifecyle

6) Related to the message’s deadlineFor the inattentive members, few days before: “you haven't viewed The message XXX closed to its deadline, maybe you should read it”

Patterns about your behavior

7) Attraction power/charisma: “you’re loosing/gaining some audience compared to the last month”

8) Diversity of attention foci: “You don’t enough diversify your interest (focus on same people or same tags)?”

Patterns about your behavior

7) Attraction power/charisma: “you’re loosing/gaining some audience compared to the last month”

8) Diversity of attention foci: “You don’t enough diversify your interest (focus on same people or same tags)?”

Page 9: Nicolas Previous Works Meeting Turin Mai08

Social translucence design2/2 Indicators about the social activity related to the message

Diffusion in the community / audience of a resource - is my message well diffused in the community? - Is everybody aware about these posting?

Diffusion in the community / audience of a resource - is my message well diffused in the community? - Is everybody aware about these posting?

Lifecycle of a resource - “is the resource dead?” - “is there a burst of activity now or these last days?”

Lifecycle of a resource - “is the resource dead?” - “is there a burst of activity now or these last days?”

Social aspect - who were the last readers? - who was interested by this document?

Social aspect - who were the last readers? - who was interested by this document?

Page 10: Nicolas Previous Works Meeting Turin Mai08

Social translucence design2/2 Indicators about the social activity related to the message

Indicators for 2 types of user’s activity (2 levels of engagement):– Awareness (access to a message)– Reaction (response to a message)

These indicators could be used also to described the user’s activity or tag’s activity

Page 11: Nicolas Previous Works Meeting Turin Mai08

orientation of the collective attention&

Alignment of the user’s attention to the community’s one.

Page 12: Nicolas Previous Works Meeting Turin Mai08

Orientation Community & user alignment

Objective 1: To understand the orientation of the community’s attention during the interval [t1, t2] following 3 levels:

Resource: which resources (message, user’s profile) got the most attention? Resource: which resources (message,

user’s profile) got the most attention?

Attention Space ( an attention focus = a resource)

Page 13: Nicolas Previous Works Meeting Turin Mai08

Orientation Community & user alignment

Objective 1: To understand the orientation of the community’s attention during the interval [t1, t2] following 3 levels:

Resource: which resources (message, user’s profile) got the most attention? Resource: which resources (message,

user’s profile) got the most attention?

Concept: Which concepts (tag/keyword) got the most attention?

Concept: Which concepts (tag/keyword) got the most attention?

Attention Space (an attention focus = a concept)

Page 14: Nicolas Previous Works Meeting Turin Mai08

Orientation Community & user alignment

Objective 1: To understand the orientation of the collective attention during the interval [t1, t2] following 3 levels:

Resource: which resources (message, user’s profile) got the most attention? Resource: which resources (message,

user’s profile) got the most attention?

Concept: Which concepts (tag/keyword) got the most attention?

Concept: Which concepts (tag/keyword) got the most attention?

User: which member got the most attention?

User: which member got the most attention? Attention space

(attention focus = a user )

Page 15: Nicolas Previous Works Meeting Turin Mai08

Objective1: Orientation of the community’s attention during [t1, t2]

Output: For each attention space (focus = resource, tag, user)

A ranking of elements sorted by their weights in the orientation of the community’s attention.

Their weights in the collective attention = Attention metrics - Resource Space: Audience of the resource during [t1, t2] - Social/Tag Space: Sum of the audiences of the resources

related to a specific user or / tag during [t1, t2]

Orientation Community & user alignment

Page 16: Nicolas Previous Works Meeting Turin Mai08
Page 17: Nicolas Previous Works Meeting Turin Mai08

Orientation Community & user alignment

Objective 2: for each level (resource, concept, user), to understand and help a user to regulate his attention by aligning it with the community’s one:

Meta cognition level: Is my attention oriented to the same resources ( or same concepts, same users) as the community’s attention?

Meta cognition level: Is my attention oriented to the same resources ( or same concepts, same users) as the community’s attention?

Regulation: “What should I read, or who should I read to improve my alignment?”Regulation: “What should I read, or who should I read to improve my alignment?”

Page 18: Nicolas Previous Works Meeting Turin Mai08

At the resource level:At the resource level:

Outputs for objective 2: For each items, displaying indicators about the user’s alignment + Suggestion to regulate the user’s behavior

Orientation Community & user alignment

Focus on me InattentionSame focus Only me

At the tag /people level: What metrics do we want?“The more the user is aware about popular resources (or active resources related to a popular tag), the more he is aligned with his community during [t1, t2]”

At the tag /people level: What metrics do we want?“The more the user is aware about popular resources (or active resources related to a popular tag), the more he is aligned with his community during [t1, t2]”

Page 19: Nicolas Previous Works Meeting Turin Mai08

Community Orientation & user alignment

TODO (work quickly done: 1 week..)

Participation alignment: “Do I have participated the hot topics?” (Alignment user’s participation/collective participation)Participation alignment: “Do I have participated the hot topics?” (Alignment user’s participation/collective participation)

Temporal alignment: “do I have stable foci of attention (reading always resources related to the same users/tags?” (Alignment past activity / user Present)

Temporal alignment: “do I have stable foci of attention (reading always resources related to the same users/tags?” (Alignment past activity / user Present)

User’s Interest alignment: “do I have a dispersed behavior according to my declared interest ?” (Alignment user’s attention/user’s intention)User’s Interest alignment: “do I have a dispersed behavior according to my declared interest ?” (Alignment user’s attention/user’s intention)

Add the notion of engagement: presence->reading-> participating“which tag /discussion stimulated the most the community (i.e having generated the most resources related to it) during [t1, t2]?”

Add the notion of engagement: presence->reading-> participating“which tag /discussion stimulated the most the community (i.e having generated the most resources related to it) during [t1, t2]?”

Page 20: Nicolas Previous Works Meeting Turin Mai08

Attention based recommendation engine

Page 21: Nicolas Previous Works Meeting Turin Mai08

Attention based recommendation engine

Problem : Is there a way to recommend me the most important messages ? 1) Avoiding uninteresting messages according my interests, 2) … except if it’s about an important issue in the community

Situation• Member of an active community• I’m overwhelmed by the unread messages• I only have 10 minutes to understand the highlights since my last login.

Page 22: Nicolas Previous Works Meeting Turin Mai08

Research problem

Question: In a rich information (and social) environment, How do I choose items (message, blog posting, .. ) due to my limited resources (e.g. time, or people)?

Answer: the notion of attention economy“in a rich information environment, information competes for the user’s attention”

I choose the most attractive items (not only about the user’s interest or what expect the user)

Attention-based Ranking Model to select items

Page 23: Nicolas Previous Works Meeting Turin Mai08

Attention based recommendation engine

Economical aspect of Attention: « Attention economy »

Before: information has value because of its scarcity

Now: Overload of information=> information has no value. What is still a limited resource is the people’s attention

Attention has value New economy information compete for the attention

(Davenport et al, 2001, “The Attention Economy: Understanding the New Currency of Business” Harvard Business School Press)

Page 24: Nicolas Previous Works Meeting Turin Mai08

Attention based recommendation engine

Psychological aspect of « Attention »

“Visual” Attention guiding the 2 types of factors:• Top-down /voluntary factor (User guidance) e.g. user searching a green object

• Bottom-up / unvoluntary factors (Stimuli guidance) e.g. flashy object in a dark scene

Page 25: Nicolas Previous Works Meeting Turin Mai08

How does an item attract the user’s attention?

Similarity of the problem in vision• In a scene (visual rich environment), which area (item) will attract my attention? • how to predict where my attention will be guided? (Visual Search problem)

Approach • Use of a visual search model: “guided Search2.0” (J. Wolfe, 1994)• Turn visual signals into communication signals (Message Reader = eye to perceive the social activity)

Page 26: Nicolas Previous Works Meeting Turin Mai08

How does an item attract the user’s attention?

Attention guiding the 2 types of features:• Top-down features (User guidance)e.g. user searching a green object

• Bottom-up features (Stimuli guidance) e.g. flashy object in a dark scene

Saliency (i.e. attractivity) of an itemThe saliency of a signal is computed as the (weighted) sum of the saliency for each attractive feature of the signal (e.g. color, size, intensity, motion,etc…)

The Visual attention model “Guided Search 2.0” - 1/2

Page 27: Nicolas Previous Works Meeting Turin Mai08

Process 1) For each attractive feature, the signals are computed into a Feature Map (i.e. their levels of saliency according to the feature)2) Mix of the feature Maps into a global Saliency Map

The Visual attention model Guided Search 2.0 - 2/2

How does an item attract the user’s attention?

Page 28: Nicolas Previous Works Meeting Turin Mai08

In your context of communication signals…

Question 1: What are the top-down features (user’s interest profile) ? Question 2: What are the bottom-up features? (i.e. attractive features without knowing the user’s intention)Question 3: How to compute a feature map?Question 4: how to compute the saliency map?

Page 29: Nicolas Previous Works Meeting Turin Mai08

Question 1: What are the top-down features? (User driven attention)

Top-down features• Message’s Topic: focus on specific topics• Message’s User: focus on specific users

Simple Vigilance profile P For a given context K (e.g. a task to do) , P(k) = (C,W) with:

- C = The set of concepts c (user, topic) I want to pay specially attention to in a signal-W = their respective levels of vigilance wc for the user

- + Limited capacity H ( ∑wc<H and wc>w min )

(I can’t want to pay attention to everything)

Vigilance feature map

Page 30: Nicolas Previous Works Meeting Turin Mai08

Question 2: What are attractive bottom-up features?

(i.e. without knowing the user’s intention)

Page 31: Nicolas Previous Works Meeting Turin Mai08

Attention based recommendation engine

Conclusion: features of this ranking model

•Based on a Visual Attention Model Not only what the user expects ( bottom up feature)

• Use of social factors to rank items.

• Try to integrate the notions of limited capacity & vigilance

•Adaptive to the context (possible change of the vigilance profile)