Player is loading...
Présentation du projet FABULA : Fake news detection in social media (en anglais)
Michael Bronstein, Université de Lugano
Wednesday, Nov. 21, 2018 · 5:51 p.m. · 12m 04s
Michael Bronstein, Federico Monti et Davide Eynard de l'Université de Lugano Le projet FABULA est un projet de recherche qui vise à détecter les fake news (fausses nouvelles ou infox) dans les médias sociaux. Les travaux réalisés jusqu'à présent s'appuient tous sur une analyse du contenu des messages et butent sur la difficulté de l'analyse sémantique du langage naturel. Les chercheurs tessinois ont choisi une approche différente qui s'intéresse aux modèles de diffusion dans les réseaux sociaux qu'ils analysent avec des méthodes d'apprentissage profond, une technologie qui a démontré son efficacité dans de nombreux domaines.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
everybody has a beginning yeah well alone or are they
00:00:10
uh_huh big everyone and then
00:00:15
uh_huh uh_huh oh cool
00:00:20
uh how would you once in a while giving me all that that they use
00:00:27
and that's it you know i mean he's uh it's in all the classes
00:00:33
are mean uh all the usual or when you're or it's not it's
00:00:43
now it will all that i'm using these things
00:00:50
basically you go and uh uh he's the now our one of a number of
00:00:57
age i mean what's the what's the or maybe it's that you see that
00:01:05
or or the i. phone but does it look like the the main
00:01:10
successes where are you learning uh because for many years uh
00:01:16
what we call you get started the message is the bridges or tunnels with these
00:01:21
and underline it's touch but the only other time so they weren't weasels
00:01:27
oh ah obligation or for example of a start let's assess or ross
00:01:34
uh and basically these uh the the the existing these it makes
00:01:38
that are now raising the low you know either by
00:01:43
and uh basically says one of our our way of sort of e.
00:01:48
we all uh actually learning and what are we are all in all a good lines
00:01:57
well
00:01:58
just very briefly because it's not just you know
00:02:01
oh ah with one of the problems is how we do uh and also feel clean all uh
00:02:09
so you're just for ah ah one of the basic operation is used when i see
00:02:17
i noticed that
00:02:20
users basically we attach ah ah the features of his own
00:02:27
wrap information to present this country does this with the values of
00:02:31
numbers and basically we design your little papers with the information
00:02:39
social leaders in
00:02:43
of the times
00:02:50
to view the name of a fast
00:02:57
if you've used ah was mentioned in the introduction of our release form
00:03:07
open analysis excel usually will be will try to do that some kind of someone working
00:03:14
but the rest of the people
00:03:17
maybe five they use
00:03:20
uh if they are awful real solution look at these are all well there might be
00:03:24
some words that will or something was you know it's like that it's they use
00:03:31
a little bit another article nothing in the world who actually i'll
00:03:35
be able to tell for example the view of the news
00:03:39
from yeah yeah it's you know that he's been on for us as a humorous it's only you need all the
00:03:46
slogan when he sees one comes on the conspiracy theory website
00:03:49
that's very uh over during the also sixteen hours
00:03:54
it's the it's another one is your oversight or but i have the survey the purpose
00:04:00
without the explicit and then all of a basic to those a holes from age
00:04:07
so they are all this is that ah it is almost
00:04:11
impossible to do is uh entirely based on them
00:04:16
own linguistic or visit some of them would come to visit for visits
00:04:20
the bottles all over food it's a little bit different information anybody gonna release it when you know
00:04:27
evidence white piece of paper that was published five below that there might be in in science
00:04:33
but shows that they use and we'll use it also yeah in particular analyst our use i don't yeah
00:04:42
and we've seen that a little gauge markets all state and we don't use our
00:04:48
the problem is that it's very difficult to uh somehow actual
00:04:51
manipulating five what uh what was taken what was cool
00:04:55
that's where our make that comes with it in to the point where you can roll that such has fallen down
00:05:01
characteristics that l. they all real and what we needed example let's say one of the violent means well
00:05:10
yeah eventually be that that became very what were we gonna
00:05:14
basically we lose the speaker or at least it
00:05:18
is stable is awful the in kind of see these big because i've different users that will supplement or
00:05:25
now oh there are many organisations uh with the package has a special journalists with uh
00:05:33
no is one of the revision be trying to free five a source that would be
00:05:39
so nice easy story was very five we journalists and was found to be not
00:05:44
basically once you obviously will that is associated with each story or with
00:05:49
i stayed or number seasons yeah that the purpose of this right
00:05:53
we can transfer these legal and identify with it 'cause it's
00:05:57
so we have a wonderful examples also uh estes get basically the way that you spend and this is a little with
00:06:05
the the the storms we're not we're this to the best of our knowledge that that those are a five story
00:06:14
so is it basically machinery in terms of what we call the trainings based on these examples we write general is fine
00:06:22
they uh they basically grow you will go use that well as well
00:06:26
whether a new story that doesn't mean c. or will be who will fall
00:06:32
well we show you how it looks like and what you can see a properly on this
00:06:35
thing here but this you will be shown here's a tiny portion of the social network
00:06:40
each a circle here represents the uh user this license over the sense of the word
00:06:47
you better be more muscle is a different colours represent whether the user is a how ah whether the user is
00:06:54
a liar or is this the the beatles they don't actually user preferences this is maybe in war falls
00:07:01
and you can see an example of a useful each so so what we
00:07:05
see for example from your well maybe you don't see your but
00:07:09
uh basically they use all using you know these every war was even if we consider
00:07:15
just ah our forwards on your uh basically we can reach almost i'm right
00:07:23
it's hold on him as a with just a few balls as little network
00:07:28
and everyone was also interested to see the the result of realisation this
00:07:33
is not a problem or testing of them before maybe different context
00:07:38
but somehow very different communities these days are the we that
00:07:42
probably not exactly on the other so it is a often comment was
00:07:48
they use with certain we will start to sell uh it is like i mean you can see
00:07:55
it it will be to use it does is fine in the run a uses of this device
00:08:02
but i think there is a be a sociological uh then managed to get to the study
00:08:09
or just you know trying to uh is it on the same note that actually use but
00:08:14
important or was it more than age it's not only be accurate but also accurate results
00:08:23
you're right right he's not only so it'll cost five or or does it say
00:08:29
uh it was a little snow white
00:08:33
it sort of summarise basically says uh uh the quality that were or the lighting up an overview of the way i
00:08:41
so what is a unit always produce holes of additional courses that it's something else
00:08:47
and the menu and just you know because instances when we don't need to or designer special linguistic what'll forage
00:08:55
oops on the the way that they disliked this awful but ah
00:08:59
it is robust servicer x. because we've gotten based approaches
00:09:04
ah basically stopped lose visitation of that first writes its
00:09:08
use and unfortunately these actions are having more remorse
00:09:13
it'll all human understanding in b. c. sits at least leave it's a very difficult if not impossible
00:09:20
to ah money leave here with that and there is always some kind of collective
00:09:26
wisdom the what the paper oh no it's better uh going to germany but
00:09:33
the stories tool and finally we also all that basically talking about this image
00:09:39
this little girl you pages we won't be able also
00:09:44
the uh basically while we classify the stories
00:09:50
and basically it's about the uh the interest you want
00:09:56
i would like to think ah me obviously it is is that most of it but it would now also has a very much
00:10:14
for one or two very short question
00:10:19
i think they can also be a ah i guess one uh
00:10:28
ah yeah uh now uh
00:10:35
i have thought about that and uh what would be the questions
00:10:40
when you use when we ah we will also be used both uh
00:10:47
uh how a source yeah work and how information is fine
00:10:54
one of ways of doing it for the whole weekend oh why don't we believe that uh i stories for fate
00:11:00
we're gonna basis you're examples what was probably even so what
00:11:05
are the buttons that we believe to be suspicion
00:11:10
it's like a cruise that windows will look at the end i feel just a little of the urge
00:11:17
that's exactly what we want uh uh you just more aware of a different style but
00:11:26
of everything i okay thank you live actually yeah
00:11:36
why there's a little problem with two to three or four
00:11:43
um
00:11:46
uh_huh yeah um
00:11:49
um
00:11:56
um hum
Conference Program
Mot de bienvenue
Marc-Henri Jobin, Directeur du CFJM
Nov. 21, 2018 · 5:35 p.m.
207 views
Présentation de la Fondation Dalle Molle
Jean-Pierre Rausis, Président de la Fondation Dalle Molle
Nov. 21, 2018 · 5:38 p.m.
108 views
Présentation du concours Label 2018
Rolf Ingold, Président du Jury
Nov. 21, 2018 · 5:46 p.m.
Présentation du projet FABULA : Fake news detection in social media (en anglais)
Michael Bronstein, Université de Lugano
Nov. 21, 2018 · 5:51 p.m.
161 views
Présentation d'AgroSustain
Jean-Pascal Aribot, AgroSustain
Nov. 21, 2018 · 6:05 p.m.
522 views
Table ronde «L’infox : menace ou chance pour la démocratie?»
Nov. 21, 2018 · 6:31 p.m.
139 views
