Embed Size (px)
Citation preview
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
QuestionsHow to facilitate learning/teaching through language (knowledge building, feedback, etc.)
Computer is good at storing-recalling-matching. More difficult to assess higher-lever language uses, understanding
Which Natural Language Processing techniques to provide just-in-time feedback as freely as possible?
31
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
How to Get More with ICT?
Three main levels of analysis [Dessus et al. ’00]
multiple-choice exams (term by term matching);
free-text assessment based on shallow features (e.g., readability, frequency count)
free-text assessment based on course content
32
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Latent Semantic Analysis: Intuitive Presentation
[Lemaire & Denhière ’05]
33
“The pilot parks the plane”
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
1. MCQs [Dessus ’00]
Les principes de la causalité naturelle sont…
A. Des principes causaux qui existent réellement dans la nature. (0,37)
B. Des principes que tous les scientifiques utilisent. (0,32)
C. Des principes des scientifiques modernes. (0,27)
D. Des principes qui ont fait leur preuve mais qu’on n’utilise plus beaucoup. (0,37)
E. Des principes inexacts qu’on a naturellement tendance à appliquer. (0,47)
LSA-based comparisons predict answers. Overal grade: 12/27 (random 5,4/27)
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
2. Note Taking [Mandin et al. ’05]
35
!"#$"%&'()"*+,-.%
//%
%
%!"#$%&'(')'*+%,-'./$-&'0,#&'.&'+1$%23'.$'!""#$%&'(&4'.&'5/&20,+&'.&'0%"2&'.&'-14&2''
!"#$%&'&(%)*&+)*,((-&+.)0"$1% '23"&%4"%4,556"&%&,5'% -")$"(77("&%8% 7"&%&),-"&%9$1%:$"&'(,559(-"&%;%)<,(1%
=$7'(37"&% "'% 7"&% 5,'"&% :$(% &,5'% 95972&6"&% 9#")% >?@+% @*(5% 4A,3'(=(&"-% 7"&% )97)$7&% -697(&6&% 9#")% )"%
3-,.-9=="B% 5,$&% $'(7(&,5&% $5% "&39)"% =$7'(4(="5&(,55"7% :$"% 5,$&% "&'(=,5&% 9))"3'9C7"B% )A"&'D;D4(-"%
&$**(&9=="5'% -"3-6&"5'9'(*% 4"&% ),559(&&95)"&% &6=95'(:$"&%4"&% 6'$4(95'&+%E,$&%5A9#,5&% )"-'"&% 39&% '"&'6%
4(**6-"5'&%),-3$&%3,$-%),5&"-#"-%7"%37$&%,3'(=97B%=9(&%5,$&%"5%9#,5&%),5&'-$('%$5%"5%-"&3")'95'%9$%=("$1%
7"&% :$97('6&% 3-"&)-('"&% 495&% 79% 7(''6-9'$-"+% F7% "&'% 4,5)% #,7$=(5"$1% G0"&&$&HHI% 39-% 7A(5'6.-9'(,5% 4"%
7A"5&"=C7"%4"&%9-'()7"&%39-$&%495&%7"%J,$-597%/&)0,(*&)"5%/KKKB%"'%$5%=(5(=$=%&36)(97(&6%GL$-C2@7HMI%
39-% 79%),5)9'659'(,5%4"&%9-'()7"&%;% 7A(5'6.-97('6%4$%),$-&%4,5'%"&'%"1'-9('% 7"% '"1'"% 7$%39-% 7"&%6'$4(95'&+%>9%
=9'-()"%),5&'-$('"%&$-%)"%),-3$&%9%6'6%-64$('"%;%MHH%4(="5&(,5&N+%O-,(&%#97"$-&%&,5'%9(5&(%46.9.6"&%8%%
!! 12$(*$3&)*&) 3,4-"&(3&)*&+)(,%&+)5) 79% ),<6-"5)"% '"1'$"77"B% '"77"%:$A"77"% "&'%="&$-6"B% 3"-="'%4A,C&"-#"-
7A"5)<9P5"="5'% 4"&% (46"&% '-95&)-('"&B% 3<-9&"% 93-Q&% 3<-9&"B% 39-% 7"&% 6'$4(95'&+% R77"% ),--"&3,54% ;% 79
=,2"55"% 4"&% (54()"&% 4"% 3-,1(=('6% &6=95'(:$"% )97)$76&% 39-% >?@% "5'-"% )<9:$"% 3<-9&"% ),5&6)$'(#"
S3<-9&"%/%"'%NB%3<-9&"%N%"'%MB+++T%4"&%<$('%39."&%4"%5,'"&+%U"''"%=6'<,4"%4"%)97)$7%-"3-"54%)"77"%#97(46"
39-%V,7'WB%L(5'&)<%"'%>9549$"-%GV,7'W@7KXI%"'%39-%#95%0(JY%"'%L(5'&)<%S/KXMT%G>9549$"-0$=9(&KZI+%%
!! 12$(*$3&) *&) +$'$1$%6*&) *&) 3,(%&(6) 7#"%$&1)5) 79% &(=(7('$4"% 4"% ),5'"5$% 39-'("7% 3"-="'% 4"% #6-(*("-% &(% 7"&
6'$4(95'&%&"%*,)97(&"5'B%7,-&%4"%7"$-%3-(&"%4"%5,'"&B%"**")'(#"="5'%49#95'9."%&$-%7"&%3,(5'&%),--"&3,5495'
;%7"$-&%79)$5"&+%R77"%"&'%933-6<"546"%39-%79%=,2"55"%4"&%(54()"&%4"%3-,1(=('6%&6=95'(:$"%(&&$&%4"%79
),=39-9(&,5%4"&%5,'"&%;%)<9:$"%&")'(,5%4$%),$-&%5,5%9):$(&"+%>"&%&")'(,5&%4$%),$-&%&,5'%),5&(46-6"&
),=="% 5,5% 9):$(&"&% 4Q&% 7,-&% :$"% 7"&% -63,5&"&% (5('(97"&% 9$1% :$"&'(,5&% ),--"&3,5495'"&% 3,&6"&
),--"&3,54"5'%;%$5%8&&*9#3:%4"%#97"$-%=9$#9(&%,$%'-Q&%=9$#9(&B%)A"&'D;D4(-"%:$954%7"%5,=C-"%4"%)9&"&
),--")'"="5'%),)<6"&%"&'%(5*6-("$-%;%79%=,('(6%4$%5,=C-"%4"%C,55"&%-63,5&"&%9''"54$"&+%%
Note-taking with feedback: Turtle visionweak cohesion of notes, higher similarity with source text
Note-taking without feedback: Eagle visionhigher cohesion of notes, lower similarity
Bachelor students in educational sciences (N=44). Knowledge Pretest + Course reading + Knowledge PosttestFactors: Feedback from pretest + Note-taking
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
3. Essay Writing [Lemaire & Dessus ’01]
36
Summary Pane:Written essay from a course
Overall Grade
How well each topic is covered
Possible followed outline
Feedback Pane:
21 real-settings essays processed:teacher vs. Apex 1 grades r = .59
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
4. Summary Writing2 intertwined cognitive processes in summarizing:
selection of the most important ideas of the source text (ST)macrorule application on some sentences of ST to compose the summary (Sum) [Kintsch & van Dijk ’78]
copy: a Sum sentence is very close to a ST onedeletion: a ST sentence is very far to all Sum onesgeneralization: a Sum sentence is close to several ST sentencesconstruction: a Sum sentence is not very close to all ST one, but pretty close to some of themSee the Demo @ http://webu2.upmf-grenoble.fr/sciedu/smandin/demos/resumwebdemo.swf
37
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
4. Summary Writing. Effect on student’s activity
38
0
7,5
15,0
22,5
30,0
Sum 1 Sum 5 Sum 10
Frequency of Copies
0
12,5
25,0
37,5
50,0
Sum 1 Sum 5 Sum 10
Frequency of Generalizations
Resum’Web Control
p < .05p < .05
Frequency of “off-the-subject” and constructions were equalSummary grades by teachers were equal
p < .05
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
5. Adding SRL Functionalities 1/2 [Dessus & Lemaire ’02]
39
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
5. Adding SRL functionalities 2/2 [Mandin et al. ’07]
40
I.2 C
om
m
I.3 Human-Computer Interaction
Fourth Pillar
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Human-Computer Interaction [after Spector 08]
Psy
cholo
gy o
f Le
arnin
g
Com
munic
atio
n T
heo
ry
HC
I
Inst
ruct
ional
Des
ign
42
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Focus on HCI“Googleization” of interfaces and feedback: “Be brief and they shall learn” [Di Eugenio & Fossati ’08]
Not only simple interface, but ones that embed psychological/instructional models (see 1st & 2nd pillars)
Personalize the interaction (agents, feedback) or the content (graphs, word clouds)
Warning: Author of none of the following examples
43
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Personalizing Feedback [Wiemer-Hastings & Graesser ’00]
Cohesion-centered feedback from an agent (Cow-boy)
44
location. The student can then dismiss the window or ask for further explan-
ation about that topic.
The personality of the kibitzers is exhibited only by the type of feedback
that they make and the picture on this window. Ideally, this association
between character and text aspect would be augmented by other agent attrib-
utes. The inclusion of a distinctive voice for each agent and perhaps some ani-
mation could make the agents much more distinguishable. These feature will
be implemented in a future version of the system.
Another type of interaction that is supported is the “push” (Graves, 1983).
For certain types of problems, students can rewrite parts of their composi-
tions “off-line’’. Figure 3 shows feedback from Marvin which presents the
student with a portion of her original text and asks her to modify it. She can
then edit the original text, make one or more completely new versions, or
maintain the original. Then she can have the new text spliced back in to the
full composition.
In summary, the student’s interaction with the tool is fairly simple. The stu-
dent enters the composition and asks for advice. The active critics offer their
feedback, giving further explanation when necessary. The complexity of the
system lies behind the scenes, as is described in the next sections.
SELECT-A-KIBITZER 157
Fig. 2. Advice from the cowboy.
Advice from Cowboy
Thank you, bye I don’t understand
I couldn’t quite understand the connection between the first sentence and the second sentence. Could make it a bit clearer? Or maybe make a new paragraph.
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Graphs & Word Clouds
Stored content easy to compute and grasp with vector-based representation
Graphs can represent semantic distances between pieces of content, persons etc.
betweenness, centrality, cohesion, density [Wild ’08]
Word clouds (e.g., Wordle.net) allow immediate perception of language-based corpora
45
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Research Trends in ED-MEDIA Conf. [Wild et al. ip]
46
!"#$%&'"()*+,-.,"/%)&0'*1&)'")10,).&#(,)*+,-.,"/%)2*)10,)1,+3&)'")10,&,)4'/1'2"#+',&)/#")5+26
7'4,) '"&'(01) '") 10,) 1,+3'"2$2('/#$8) $,9'/#$6&,3#"1'/) /0#"(,&:) ;2) 4'&/27,+) <.+&1&8) #) =.&16
"21'*'#<$,64'**,+,"/,) 10+,&02$4) 2*) :>>?) @#&) .1'$'&,4:) ;0'&) 10+,&02$4) ,-.#$&) #") #<&2$.1,) '"6
/+,#&,)<%)A:B).&,&)2+)#)+,$#1'7,)'"/+,#&,)<%)CD:E)F)2*)10,)3,#")*+,-.,"/%)'")C>>>:)G&'"()10'&)
10+,&02$48)?B)1,+3&)&,,3)12)0#7,)#)4'3'"'&0,4)+2$,)@0,+,#&)HA),952&,)#"),"*2+/,4)+2$,:))
!)@2+4) /$2.4) I.&'"()@2+4$,:",1J) 12) 7'&.#$'&,),#/0)2*) 10,) *2.+)(+2.5&) I4'3'"'&0,4) +2$,8) ,"6
*2+/,4)+2$,8)",@8)4'7,#+,4J)'&)4,5'/1,4)<,$2@8)+,*$,/1'"()10,)#/1.#$)*+,-.,"/%)'")10,)$,16
1,+)&'K,:))
)
!"#$%&$%L,@);,+3&)IC>>>)12)C>>MJ:%
)
!"#$%'$%N'7,#+,4);,+3&)I*+23)C>>>)12)C>>MJ:%
)
)!"#$%($%N'3'"'&0,4)O2$,)I'")C>>MJ:%
)!"#$%)$%P"*2+,4)O2$,)I'")C>>MJ:%
)
Q.#$'1#1'7,$%8)10+2.(0)10,)/2&'",)4'&1#"/,&)2*)10,)1,+3)7,/12+&)'")10,)$#1,"16&,3#"1'/)&5#/,8)
#)(+#50)&1+./1.+,)/#")<,),&1#<$'&0,4)10#1)R)&'3'$#+)12)#)&2/'#$)",1@2+S)R)/2"",/1&)1,+3&)#&)
10,)"24,&) '")10'&)(+#50)@'10),#/0)210,+)10+2.(0) $'"S&8) 10.&)+,*$,/1'"()10,'+)&,3#"1'/)/$2&,6
",&&:)G&'"()4'7'&'7,)/$.&1,+'"()I5+,/'&,$%T)10,)4'#"#)4'7'&'7,)#"#$%&'&)/$.&1,+'"()#$(2+'103J8)10'&)
",1@2+S) /#") <,) 5#+1'1'2",4) 0',+#+/0'/#$$%) '"12) /2352","1&:)U0,") $22S'"() #1) 10,) 4,"4+26
(+#33,&)4,5'/1,4)<,$2@8)#).&,*.$) $,7,$)2*) #"#$%&'&) *2+) '"&5,/1'"()#) +,#&2"#<$,)".3<,+)2*)
/2352","1&)2")10,),)$,7,$)/#")<,)&,1)#1)#)0,'(01)2*):?:)V2+)10,)(+#50)2*)10,)4'/1'2"#+%)'")
%,#+) C>>>8) 10'&) ,**,/1'7,$%) +,&.$1&) '") 1,") /2352","1&8) @0,+,#&) C>>M) $'&1&) ?B) /$.&1,+&:)
)
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Relations between Persons [Wild ’08]
47
Red nodes: reviewersGreen nodes: editorsOther nodes: contributorsLinks: interactions
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Knowledge from Papers [Landauer et al. ‘04]
48
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Learner Positioning from Interviews [Berlanga et al. ’09]
Generation of expert and student concept maps with Leximancer
49
ExpertLearner
23
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Representing Topic Flow in Essays
[O’Rouke & Calvo ’09]
50
4. Analysis of the 2-Dimensional Visualisation
A well structured and developed essay answer should have a clear and logical flow of topics throughout its paragraphs. To support feedback on this aspect of essay writing, we use the 2-dimensional representation to provide insight into an essay’s topic flow. In a paragraph ‘map’, such as those in Figure 1, an essay’s paragraphs are plotted on a circular grid with the diameter of the grid equal to the maximum possible distance between any two paragraphs (i.e. no topic overlap). The paragraphs are represented using a node-link diagram with text labels and arrows used to indicate the sequence of paragraph.
Consider, for example, how the clear sequence of topics in the five paragraph essay paradigm [16] would appear in a paragraph map. In this paradigm, the content of the ‘introduction’ and ‘conclusion’ paragraphs is expected to be similar, so these paragraphs should appear close in a map. The ‘body’ paragraphs address different subtopics and should ideally be linked through transitions, so they should be sequentially positioned in the map. Thus, the map of an ideal five paragraph essay would have a circular layout of sequential paragraphs, indicating a natural change in topic over the essay, with the introduction and conclusion paragraphs starting and finishing on similar points. In contrast, we would expect a poorly structured essay to have many rough shifts in topic, with paragraphs positioned almost randomly around the map.
Figure 1 illustrates the paragraph maps of two short essays. The essay on the left was given a low grade while the essay on the right was given a high grade. The topic flow of the high grade essays clearly resembles that of the prototypical five paragraph essay described above, while topic flow of the low grade essay appears disorganised. The low grade essay shows possible signs of disconnectedness through rough topic shifts as well as repetition through paragraphs of near identical topic mixtures.
Figure 1. The paragraph maps of an essay with a low grade (left) and an essay with a high grade (right).
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Social Proxies[Erickson & Kellogg ’04]
51
On-Line Lecture
Erickson and Kellogg 7 Social Navigation Chapter
their views of a social proxy. This is important because it is what supports mutual awareness and
accountability: I know that if I see something in the social proxy, that all other viewers can see it as
well.
• Third Person Perspective. Social proxies are represented from a third-person point of view. When
I look at a social proxy, I see myself represented in it in the same way that other participants are
represented. This opens an important avenue for learning. As I act within system, I can see how my
actions are reflected in my personal representation, and thus I can begin to make inferences about the
activities of others.
This is rather abstract, so let’s take a look at an example. Figure 1.1 shows
a social proxy from the Babble system (discussed in section 1.4), that we
refer to as “the Cookie.” The purpose of the Cookie is to reflect the real
time presence and activities of participants in a multi-channel chat-like
system. The large circle represents the ‘current’ conversation (i.e. the one
being viewed by the user). The small colored dots (called “marbles”)
represent people who are logged onto the Babble system, including the user
from whose viewpoint we are seeing things. Marbles that appear inside the
circle depict participants who are looking at the current conversation;
marbles outside the periphery represent people who are logged on to the
chat system but in a different chat. Finally, when participants are
active—meaning they either ‘speak’ (i.e. type), or ‘listen’ (i.e. click or
scroll)—their marbles move to the inner part of the circle, and then, when
activity ceases, drift back out to the edge over the course of about twenty
minutes.
Thus, the Cookie shown in Figure 1.1 shows that ‘something is happening.’ The tight cluster of five
marbles around the center core of the circles shows that those participants are engaged (they are either
typing, or clicking and scrolling as one often does when participating in the chat). Two other users,
depicted by marbles at six and seven o’clock, are also viewing the same conversation, but have not been
active. Possibly they are away from their computers, or possibly they are working on other things and
ignoring the conversation. The eighth marble (at four o’clock) depicts a user who is logged onto the
system but viewing a different conversation; that user may or may not be active—all we can tell is that
he or she is connected to the system. Thus, the Cookie shows people are either here or ‘around’, and if
here, how recently they have done something; it also shows when people arrive or depart, in that new
marbles appear or disappear (if logging on or off) or move into or out of the bounds of the circle.
We will discuss the Babble Cookie and how people make use of the cues it provides in the next section.
However, before we do that, we will present some other examples of social proxies to provide a glimpse
of the power and generality of this technique.
1.3.3 The Lecture Proxy
A common situation in the face to face world is that one person will speak to an audience which, by and
large, remains quiet. Class room lectures, professional talks, and business presentations are all examples
of this. Suppose that we have an online analog of this situation. Perhaps the talk or lecture is being
delivered via audio as part of a large conference call; we can imagine that members of the distributed
Figure 1.1. The Babble
‘Cookie’
Chat Discussion
Color Circles: PeopleWhite Circle: Conversation FlowBlack-line Circle: Chat Room
I.3 H
CI
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Research QuestionsThe design of cognitive tools wrt. HCI:
Interface not too cognitively demanding
Underlying metaphor?
Representing intentions/beliefs/desires, which are crucial in collaborative learning and teaching?
52
I.3 H
CI
II. Research Projects
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Rationale1. Embed teaching and learning models into ILEs
2. Integrate them in a communication way or technique, which can help teaching/learning
3. Design the interface and fix HCI concerns, according to
54
II. Pro
jects
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Activity Integration [Dascalu ’09]
55
Chatting
Reading Summar-izing
Prov
ide pr
ior k
nowledg
e
Iden
tify un
know
n to
pics
and
retu
rn
appr
opria
te re
adin
g su
gges
tions
Generate chat sum
maries autom
atically
Present information & sum
mary
Obtain summary/synthesis
Integrate in the current lecture
II. Pro
jects
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Three mainactivity loops [Dessus et al. 09]
56
Understanding/Knowledge Building
Chat
Write Utterances
Chat LoopChatChat
Manuels Read Texts
Reading Loop
TextbooksPortfolio
Write Texts
Writing Loop
Portfolio
II. Pro
jects
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Future ConcernsCombine cognitive-based and narrative-based points
Multiple-source syntheses
Embed individual comprehension-focused processing to predict student’s own understanding of course and utterances
57
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Questions for Discussion
All these activities are already existing in current classroom settings
Is ICT the good way to assist teacher activity? Why not to use pencil/paper-based ones?
What kind of improvements, experiments or new cognitive tools to devise, accounting for the 4 pillars
psychology of learning & Instructional Design
communication theory
HCI?
II. Pro
jects
Thanks for your attention [email protected]
Get the refs: http://www.citeulike.org/user/pdessus/tag/ictresearch
Get these Slides: http://webu2.upmf-grenoble.fr/sciedu/pdessus/
© veryveryfun.com
Additional Material
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Analysing Content with LSA [Landauer ‘02; Lemaire & Denhière ‘05]LSA determines the statistical context in which each word occurs; semantically compares words; serves as semantic memory
two words are similar if they occur in same paragraphs
two paragraphs are similar if they contain common words
two words are similar if they occur in similar paragraphs
two paragraphs are similar if they contain similar words
61
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Latent Semantic Analysis 2/2 [Lemaire & Denhière ’05]Given a corpus processed beforehand
split in paragraphs
words are projected in a n-dimension space so that
• words can be compared to each other by computing the cosine of their corresponding vectors
• paragraphs can be compared to each other by computing the cosine of the sum vector of the words they are composed of
62
I.2 C
om
m
Des
sus
- Ed
uca
tion &
Tec
hnolo
gy S
um
mer
Sch
ool -
Stra
sbourg
2009
Polyphony : Feedback tool for Chat [Trausan-Matu & Rebedea 09]
63
Chat with explicit links (blue) and implicit (red). Red bars = contribution’s importance
Automatic detection of themes and discussion threads (token analysis)
Visualization of chat threads and feedback
II. Pro
jects
