Download human networks

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Belongingness wikipedia , lookup

Group development wikipedia , lookup

Human bonding wikipedia , lookup

Impression formation wikipedia , lookup

Albert Bandura wikipedia , lookup

Social perception wikipedia , lookup

Transcript
Automatic mapping and
modeling of human networks
ALEX (SANDY) PENTLAND
THE MEDIA LABORATORY CAMBRIDGE
PHYSIC A: STATISTICAL MECHANICS
AND ITS APPLICATIONS
2007
Outline
2
 1. Introduction
 2. Socioscopes
 3. Reality mining
 4. Social signals
 5. Practical concerns
 6. Conclusions
 Comments
1. Introduction (1/2)
3
 Studies on office interactions [1]



35–80% of work time in conversation,
14–93% of work time in communication
7–82% of work time in meetings
 The properties of human networks :



Location context: work, home, etc.
Social context: with friends, co-workers, boss, family, etc.
Social interaction: are you displaying interest, boredom etc.
 To obtain solid, dynamic estimates of the users’
group membership and the character of their social
relationships.
[1] T. Allen, Architecture and Communication Among Product Development Engineers, MIT Press, Cambridge, MA, 1997, pp. 1–35.
1. Introduction (2/2)
4
 Using this data to model individual behavior as a
stochastic process

allows prediction of future activity.
 The key to automatic inference of information
network parameters is the recognition
 Standard methods, surveys
subjectivity and memory effects, out-of-date.
 Even information is available, need to validate or correct
by automatic method

 we present statistical learning methods
 wearable sensor data to estimates user’s interaction
2. Socioscopes
5
 mapping and modeling human networks
 the conceptual framework used in biological observation,


such as apes in natural surroundings
natural experiments

such as twin studies,
 but replacing expensive and unreliable human
observations with automated, computermediated observations.
accurately and continuously track the behavior
 recording with near perfect accuracy.

Imaginary Socioscope
6
 Using mobile telephones, electronic badges, and
PDAs

Tracking the behavior of 94 people in two divisions of
MIT

the business school and the Media Laboratory
between 23 and 39 years of age
 the business school students a decade older than the
Media Lab students.
 2/3 male and 1/3 female


half were raised in America.
Three main parts of the Socioscope
7
 The first part: ‘smart’ phones
 to observe gross behavior (location, proximity)
continuously over months
 330,000 h of data , the behavior of 94 people, 35 years
 The second part: electronic badges
 record the location, audio, and upper body movement
 to observe for fine-grained behavior (location, proximity,
body motion) over one-day periods
 The third part: a microphone and software
 to analyze vocalization statistics with an accuracy of
tenths of seconds
2. Socioscopes (4/5)
8
 Four main types of analysis:
 characterization of individual and group
distribution and variability

using an Eigenvector or principal components analysis
conditional probability relationships between
individual behaviors known as ‘influence modeling’
 accuracy with which behavior can be predicted



with equal types I and II error rates
comparison of these behavioral measures to standard
human network parameters.
3. Reality mining
9
 Eigenbehaviors provide an efficient method for
learning and classifying user behavior [9].
 Given behaviors Γ1, Γ2, . . . ,Γm for a group of M
people,

the average behavior of the group can be defined by
 To deviate an individual’s behavior from the
mean.

A set of M vectors, Φ = Γi - Ψ,
[9] N. Eagle, A. Pentland, Eigenbehaviors: Identifying Structure in Routine, October 2005, see TR 601
hhttp://hd.media.mit.edui.
Fig. 1
10
 Γi(x,y), 2-D location
information
a low-dimensional
‘behavior space’,
 spanned by their
Eigenbehaviors

3.1. Eigenbehavior modeling
11
 Principle Components Analysis, PCA
 a set M orthonormal vectors, un, which best describes
the distribution of the set of behavior data when linearly
combined with their respective scalar values, λ n.
 Covariance matrix of Φ


Where
The Eigenbehaviors can be ranked by the total amount
of variance in the data for which they account, the
largest associated Eigenvalues.
3.2. Human Eigenbehaviors (1/2)
12
 The main daily pattern, observed
 subjects leaving their sleeping place to spend time in a
small set of locations during the daylight hours
 breaking into small clusters to move to one of a few other
buildings during the early night hours and weekends
 then back to their sleeping place.
 Over 85% of the variance in the behavior of low
entropy subjects can be accounted for by the
mean vector alone.
3.2. Human Eigenbehaviors (2/2)
13
 the top three Eigen behavior components
 the weekend pattern,
 the working late pattern, and
 the socializing pattern.
 The ability to accurately characterize peoples’
behavior with a low-dimensional model means
automatically classify the users’ location context
 the system to request that the user label locations
 can achieve very high accuracies with limited user
input.

3.3. Learning influence (1/2)
14
 Behavioral structure
 Conditional probability to predict the behavior
 Two main sub-networks
during the day
 in the evening

 Critical requirement for automatic mapping
and modeling of human networks
to learn and categorize user behavior
 accurately capture the dynamics of the network.

3.3. Learning influence (2/2)
15
 Coupled Hidden Markov Models, CHMMs [10-12]
 to describe interactions between two people
 the interaction parameters

limited to the inner products of the individual Markov chains.
 The graphical model for influence model
behavior has the same first-order Eigen structure
 it possible to analyze global behavior

[10] A. Pentland, T. Choudhury, N. Eagle, S. Push, Human Dynamics: Computation for Organizations, Pattern
Recognition, vol. 26, 2005, pp. 503–511, see TR 589 hhttp://hd.media.mit.edui.
[11] W. Dong, A. Pentland, Multi-sensor data fusion using the influence model, IEEE Body Sensor Networks, April,
Boston, MA, 2006, see TR 597 hhttp://hd.media.mit.edui.
[12] C. Asavathiratham, The influence model: a tractable representation for the dynamics of networked Markov
chains, in: Department of EECS, 2000, MIT, Cambridge.
3.4. Influence modeling
16
 Using the influence model to analyze the proximity
data from our cell phone experiment
 we find that Clustering the daytime influence
relationships
96% accuracy at identifying workgroup affiliation
 92% accuracy at identifying self-reported ‘close’
friendships.

 Similar findings, using the badge platform.
 the combination of influence and proximity predicted
whether or not two people were affiliated with the same
company with 93% accuracy [6].
4. Social signals
17
 People are able to ‘size up’ other people from a very
short period of observation [13, 14].
linguistic information from observation,
 to accurately judge prospects for friendship, work
relationship, negotiation, marital prospects

 we developed methods for automatically
measuring some of the more important types of
social signaling [7].

Excitement, freeze, determined and accommodating.
Predict human behavior
18
 Can predict human behavior?
 without listening to words or knowing about the people
involved.
 By linear combinations of social signal features to
accurately predict human behaviors.
who would exchange business cards at a meeting;
 which couples would exchange phone numbers at a
bar;
 who would come out ahead in a negotiation;
 who was a connector within their work group;

5. Practical concerns
19
 Continuous analysis interactions within an
organization may seem reasonable and if
misused, could be potentially dangerous.
 Conversation postings:
the data should be shared, private, or permanently
deleted.
 Decided by individuals.

 Demanding environments:
 the environmental demands may supersede privacy
concerns.
6. Conclusions
20
 human behavior is predictable than is generally
thought, and especially predictable from others.
 This suggests that

humans are best thought of social intelligences rather
than independent actors.
 As a consequence
 can analyze behavior using statistical learning tools


such as Eigenvector analysis and influence modeling,
to infer social relationships without to understand the
detailed linguistic or cognitive structures surrounding
social interactions.
Comments
21
 經由human network 找出人與人間的關係及其建立
model的作法
 在我們的運用是找出criminal及找criminal的同伙
 瞭解將行動電話及不同的sensor等如何運用在
human network
 Human network 對 prediction的幫助為何?