Interactive Powerpoint_How to Master effective communication
Network Learning: AI-driven Connectivist Framework for E-Learning 3.0
1. 国立大学法人
電気通信大学
振興調整費
Unique and Exciting Campus
ネットワーク ラーニング
Network Learning
AI-driven Connectivist Framework for E-Learning 3.0
Neil Rubens
Active Intelligence Group
Knowledge Systems Laboratory
University of Electro-Communications
Tokyo, Japan
2. Evolution of eLearning: eLearning 1.0
eLearning uses technology to enhance Learning
---------
‣ eLearning 1.0:
---------
---------
---------
---------
‣ reading: content became easily accessible
‣ logging: user’s activities could be logged and analyzed
‣ Learning Theories:
‣ Behaviorism: learning is manifested by a change in behavior,
environment shapes behavior, contiguity
‣ Cognitivism: how human memory works to promote learning
3. Evolution of eLearning: eLearning 2.0
‣ eLearning 2.0:
---------
--------- ---------
--------- ---------
--------- ---------
--------- ---------
---------
‣ writing: anybody can easily create content (e.g. blogs, wiki, etc.)
‣ socializing: interaction is easy (e.g. facebook, twitter, etc.)
‣ Learning Theories:
‣ Constructivism: constructing one's own knowledge from one's
own experiences
(enabled through writing)
‣ Social Learning: people learn from one another
(enabled through socializing)
4.
5.
6. Broken Knowledge Cycle
‣ Problem: The current cycle of knowledge creation/utilization is inefficient !
‣ large portion of created content is never utilized by others*
only 0.05% of twitter messages attracts attention (Wu et. al., 2011)
only 3% of users look beyond top 3 search results (Infolosopher, 2011)
‣ large parts of created contents are redundant (Drost, 2011)
‣ Peak Social – the point at which we can gain no new advantage from social
activity (Siemens 2011)
utilize
U0lized
dge
no wle
K
is ting
Ex
Redundant
create
Knowledge Novel
*there are some personal benefits e.g. externalization, crystallization, etc.
7.
8. Information Overload
for Computers
not a Problem
but an Opportunity
Tan,
Steinbach,
Kumar;
2004
500
Points
2,000
points
8,000
points
9. Messaging Networks
How can we use computers to
learn in these settings?
http://datamining.typepad.com/photos/uncategorized/2007/04/08/twitter20070405.png
Citation Network
Social Network
hp://wiki.ubc.ca/images/f/ff/SocialWeb.jpg9
hp://www.kieranhealy.org/files/misc/SocCoreCites.jpg:
11. Connectivism (Learning Theory)
Connec0vism:
Knowledge
is
distributed
across
a
network
of
connecTons,
and
therefore
learning
consists
of
the
ability
to
construct
and
traverse
these
networks
(Siemens
Downes,
2008)
Property Behaviourism Cognitivism Constructivism Humanism Connectivism
Learning Thorndike, Pavlov, Watson, Koffka, Kohler, Lewin, Piaget, Maslow, Siemens, Downes
theorists Guthrie, Hull, Tolman, Skinner Piaget, Ausubel, Vygotsky Rogers
Bruner, Gagne
How learning Black box—observable Structured, Social, meaning Reflection on Distributed within a network, social,
occurs behaviour main focus computational created by each personal technologically enhanced,
learner experience recognizing and interpreting
(personal) patterns
Influencing Nature of reward, punishment, Existing schema, Engagement, Motivation, Diversity of network, strength of
factors stimuli previous experiences participation, experiences, ties, context of occurrence
social, cultural relationships
Role of Memory is the hardwiring of Encoding, storage, Prior knowledge Holds changing Adaptive patterns, representative
memory repeated experiences—where retrieval remixed to concept of self of current state, existing in
reward and punishment are current context networks
most influential
How transfer Stimulus, response Duplicating knowledge Socialization Facilitation, Connecting to (adding) nodes and
occurs constructs of “knower” openness growing the network (social/
conceptual/biological)
Types of Task-based learning Reasoning, clear Social, vague Self-directed Complex learning, rapid changing
learning best objectives, problem (“ill defined”) core, diverse knowledge sources
explained solving
12. ConnecTvism:
Nice
Theory
Need:
Tools
Frameworks
To
make
it
Prac0cal
hFp://imgs.sfgate.com/c/pictures/2011/12/19/ba-‐BRIDGE20_SFC0105724887.jpg
17. Sequence Diagram (Example)
user system user system user system
I want to know about term t_i I think t_i and t_k are similar ..
I want to know about term t_i and t_k
social
semantics semantics
t_i
t_i u_i: I think t_i is same as t_j …
t_i
t_i u_j: no t_is is more like t_p ...
u_j: no t_is is more like t_p ...
t_i t_i
t_i t_i u_m: I think t_i and t_k are similar ..
u_i: you are right
t_i
t_i
t_i
t_i
t_i semantics t_i
t_i
contents contents
t_i t_i
t_i
t_i
contents t_i
social social
social
u_i: I think t_i is same as t_j … u_i: I think t_i is same as t_j …
u_j: no t_is is more like t_p ... u_j: no t_is is more like t_p ...
u_k: you are both wrong t_i is ... u_j: no t_is is more like t_p ...
u_i: I think t_i is same as t_j …
u_j: no t_is is more like t_p ...
u_j: no t_is is more like t_p ...
u_m: I think t_i and t_k are similar ..
u_i: you are right
18. Connecting Representations ⇤
V (d1 )
Semantic + Network
⇤
V (d2 )
⇤
Type! Network (N)! Content (C)! Proposed Hybrid!
V (d1 )
⇤
V (d3 )
⇤
V (d2 ) Characteristics!
This is a Title of a ! tfi,j = P
ni,j
Research Paper ⇤ (d
V (d )1 )
V
k nk,j
Joe Fakeman Jane Noman
! 3
Analysis!
V (d1 )
Nowhere University
{fakeman, noman}@nowhereuni.edu
|D|
Abstract idfi = log
~~~~~~~~~~~~~~~~~~~~~ ni,j |{d : ti ⌅ d}|
tfi,j = P
!#$%$'()%*+(,( !#$%$'()%*+(-(
~~~~~~~~~~~~~~~~~~~~~
!
~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~
k nk,j
Introduction
~~~~~~~~~~~~~~~~-
~~~~~~~~~~~~~~~~-
~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~ V (d2 ) tf-idf i,j = tfi,j ⇥ idfi
same as C part of N
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
|D| ⇤
! V (d1 )
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
idfi = log
!
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
V (d2 )
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
|{d : ti ⌅ d}|
w (d1 )
V = tf-idf
i,j i,j User Interface!
! ⇤
V (d2 ) ./+$(/.(,(
V (d3 )
This is a Title of a tf-idf i,j = tfi,j ⇥ idfi 2 3
Research Paper w1,j
6 w2,j 7
Joe Fakeman Jane Noman
! ⇤
! 6 7
V (d3 ) V (dj ) = 6 . 7
Nowhere University
wi,j = tf-idf i,j
V (d ) 4 . 5 ⇤
{fakeman, noman}@nowhereuni.edu
Abstract . V (d3 )
~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~
2 wt,j
~~~~~~~~~~~~~~~~~~~~~ 2 3
Introduction
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
6
w1,j
7 ni,j
Deployment!
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
⇤ 6 w2,j 7 tfi,j = P
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
V (dj ) = 6 . 7
4 . 5 ./+$(/.(,(
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
.
! k nk,j
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
wt,j
V (d3 )
This is a Title of a ⇤ ⇤ |D|
Research Paper V (di ) · V (dj ) idfi = log
sim(di , dj ) = ⇤ ⇤ |{d : ti ⌅ d}|
Joe Fakeman Jane Noman
Nowhere University
{fakeman, noman}@nowhereuni.edu
V (di ) V (dj ) Execution
Abstract
~~~~~~~~~~~~~~~~~~~~~
⇤ Speed! ./+$(/.(,(
V (d1 ) tf-idf i,j = tfi,j ⇥ idfi
~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~
Introduction
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
⇤
V (d2 ) wi,j = tf-idf i,j
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~- ~~~~~~~~~~~~~~~~~~
Reverse could also be done: i.e. converting textual representation to a network one.
⇤ 2 3
term-space V (d3 )
6
w1,j
7
1 ⇤ 6 w2,j 7
V (dj ) = 6 . 7
.
tfi,j =
n
P i,j
4 . 5 ⇥
! k nk,j w|T |,j w1,j
V (d1 ) ⇧ ⌃
1
! ⇤ ⇤ ⇧ w2,j ⌃
V (d1 ) |D|
idfi = log sim(di , dj ) =
V (di ) · V (dj ) ⇧ .
. ⌃ !
|{d : ti ⌅ d}| ⇤ ⇤
V (di ) V (dj ) ⇧ . ⌃ V (d1 )
! ⇧ ⌃ !
— V (d2 ) ⇧ w|T |,j ⌃ V (d1 )
—-
! Network Content based ⇥ idfi
tf-idf i,j = tfi,j ⇥ ⇧ ⌃
V (d2 ) V (dj ) = ⇧
⇧
⌃
⌃ !
wi,j i,j = (i, j) j)
w = dist(i,
⇧ w1,j ⌃ V (d2 )
!=
wV (d tf-idf i,j ⇧ ⌃ !
—- i,j
3 ) 2 3 ⇧ w2,j ⌃ V (d2 )
Generalized Assignment Problem
! w1,j ⇧ ⌃
V (d3 ) 2
Objective: given a paper p, 1,j ⇤
3 6
choose a 6 w2,jof 7
group
7
experts ⇧ .
.
⌃ ! !
V (d1 )
w V (d ) = 6
M ⇥ M (of a fixed size6s) 2,j 7jcollectively (d )
⇤ w that 4
! possesses
.
.
V 5
.
7 ⇤ . ⌅ V (d3 )
V (d ) = 6
6 7 1 !
the most expertise about p: . 7
j
4 . 5.
w|N |,j w|N |,j V (d3 )
wt,j P
maximize R(M, p) = m2M r (m, p) (1) !
subject to |M | =!(d )
s (2)
V (d2 )
⇤ ⇤
V (di ) · V (dj )
V 2
Challenge sim(di , dj ) = ⇤ ⇤
V (di ) V (dj )
Expertise is not additive: 1 !
X ! V (d3 )
R(M, p) ⇤= V (d3 .
r (m, p)) (3)
m2M
–
Group Expertise Estimation
Assumptions
Mp are athors of paper p; so R(Mp , p) = 1, and
⇤ ⇤
network-space
R(M, p) = 0, where M ⌅ Mp = Ø.
⇤ term-network-space
M ⌅ M⇤