Player is loading...

Embed

Embed code

Transcriptions

Note: this content has been automatically generated.
00:00:02
i uh first of all thank you for the stock uh i have a quick question of you develops over here to score
00:00:10
or does it to target i don't really know this is that would be great better event i think at the moment it
00:00:17
does not detect irony but obviously this is um one of the
00:00:20
things that we have a narrator like i said with
00:00:24
um with specialised data um from whatever customary
00:00:30
might have that would be interested
00:00:32
in this week and train the model that would detect irony well
00:00:37
hum
00:00:41
me too and your questions or
00:00:46
oh there are two or correct
00:00:57
a minimum standards uh for questionable chopped so data uh_huh uh uh so what kind of that to you
00:01:05
it from the top converse or some user customer and the uh uh i'm pretty sure
00:01:10
you do it's something um how to use you you mean aborts so if you
00:01:17
hi dennis um so uh things requesting an uh yeah
00:01:23
so we do take a yeah there's it
00:01:26
there's also matter data coming with the chats obviously it we don't have the name of the customer unless
00:01:32
they they provided voluntarily or unless it's in the text
00:01:36
mentioned somewhere um there is location uh um
00:01:42
maybe language things like that um for the chats this
00:01:47
is something that uh obviously that we're looking into
00:01:51
um we're working very closely with e. p. f. l. i'm not sure if you've heard that uh says
00:01:57
come recently opened a lab at the p. f. l. it's got that has come digital that
00:02:02
and for this lab we planned to um invest
00:02:06
a lot of money in in research um
00:02:10
in various areas and and artificial intelligence specifically natural language presses presses thing is one of them
00:02:17
oh a chat part a case it is part of
00:02:21
this research that is that has very recently started
00:02:28
no yeah we'll use this data in order to um
00:02:32
develop uh first of all uh a generic chart
00:02:36
but and then look into how you can personalise it to the to the uh a customer
00:02:46
but we haven't come from for more question order or we take first of all
00:02:58
hello ah how you interested in day sentiment analysis part
00:03:03
of your stolen our application maybe can you explain
00:03:06
a bit more how are you doing and how easy what elements are you conceding yeah sure so
00:03:14
maybe i should mention this uh this on our
00:03:17
project started very small so uh about
00:03:20
a year and a half ago and it was a six month project uh with
00:03:27
to developers me doing the data set and when i bring different
00:03:32
so um yeah not very uh not a lot of
00:03:35
resources and we started a very simple i mean you could
00:03:39
say we didn't even use a highway we had uh
00:03:42
uh i'm dictionary of words that were annotated with sentiment and
00:03:46
we did calculated a score based on that then
00:03:48
i had the same time we started internally gathering uh
00:03:52
data um that would be labelled by by employees
00:03:58
well we had some um feedback from customers that unfortunately we're not able
00:04:03
to to put on crowd surfing platforms because it you know
00:04:08
at my data and they're not allowed to to be published um so we
00:04:12
did this with them plays obviously it's not easy seen since people have
00:04:16
for other things to do it and sit down and labels for you
00:04:21
but we did manage together as some data in this way and then we is this um
00:04:27
mm to feed 'em a classifier from the stanford korean out the library
00:04:32
and so currently the model the is used in stoner is
00:04:36
based on this data and a quarter an l. p.
00:04:42
sore backs well thank you
00:04:46
oh thank you for the presentations i have um high level non technical question
00:04:51
what about privacy because i saw that you actually log user data is
00:04:55
their way where we take that we accept as is or is it
00:04:58
automatic from the fact that it's anonymous because even though it's anonymous
00:05:03
from the data we can collect we can already profile the user somehow with
00:05:06
the location and the the behaviour yes so um at the moment
00:05:13
at the moment the application is not i'm open to the public so there is
00:05:20
you cannot take a box in finland you agree but i'm probably there will be something like this
00:05:25
yeah i mean if if it becomes public and like i said we um uh oh
00:05:31
we sat down with our legal department and um came up with
00:05:35
a list of of conditions that we need to fulfil in
00:05:38
order to be able to um nice this data and this included among the other is the annotation part that i mentioned
00:05:51
next question real sorta recorder works well
00:05:57
probably across the room
00:06:01
i i uh i mean the so um i have a question about the
00:06:07
so it's now it's accretion that this is done and you're right this
00:06:11
is something you lied or you're looking at but not yet
00:06:16
yes you don't have it yeah yes now we don't have but um i know
00:06:20
the p. though is is you automate cold so actually replace the first level
00:06:27
which is mainly the port all of that arise in the incident and
00:06:31
assigning it still sucking the corresponding second level team which has
00:06:36
the deeper insight into it as you call so mainly
00:06:40
um it away this porch from men all in to admit that we
00:06:46
yeah it's not an explicit goal uh i can imagine that it might happen as a side effect
00:06:53
um it hasn't yet happened it's eskimo as i mentioned we're already using this
00:06:59
says and a few months and am to press the permanence has come up
00:07:04
um we'll see end up um i guess say there there
00:07:09
is potential there for the first level to be replaced
00:07:13
thanks
00:07:15
so who do we have to we have to par for wallace question of war
00:07:23
yes there as well
00:07:31
but i think you're presentation rather interesting do learn it's just come to
00:07:35
to open source some project booked maybe like this one other little
00:07:42
oh and version i cannot provide uh and um binding answer
00:07:47
at but um as i've already mentioned to uh
00:07:53
uh i seek name we're we're using cornell b. which is open
00:07:56
stuff source software weary i guess most of our or
00:08:00
oh the components that we're using open source um so this
00:08:04
is definitely something that we're looking into but like i
00:08:08
said we're and the beginning but we're not competing with google
00:08:12
and phase begins on so there's not really any
00:08:15
um intermediate incentive to do this
00:08:23
who are out there was a rubber i'll or or or
00:08:33
uh_huh
00:08:35
it's just to to pick up all you mention but you're not allowed to puzzles custom
00:08:41
what are the rules regulations with slim backing up a little are you what are you operating with him
00:08:50
though when you turkeys all these they're all they're very
00:08:53
very strict in switzerland and we've had to
00:08:59
have some quite heated discussion we legal about it um
00:09:04
i think for us on at the moment
00:09:08
we were able to get away with doing it because it's a internal only
00:09:15
but as soon as it gets out we we will need to do a lot more so
00:09:21
um hum like i can mention we're not allow uh we not
00:09:26
allowed to have any um custom identification in the data
00:09:31
um if someone comes to us and is asking whether
00:09:35
we're processing their data we need to be able
00:09:38
to answer does and we need to be able to stop doing it if they so request
00:09:43
um there are also some some um a
00:09:47
constraint that don't come from switzerland
00:09:50
from just law for example for critter um if you gather tweaked and someone deletes there treat
00:09:57
you need to be able to detect this and delete also in your
00:10:02
system oh there are lots of things and you can imagine what we're doing our best

Share this talk: 


Conference program

Keynote
Jean-Baptiste Clion, Coordinator DevFest Switzerland
26 Nov. 2016 · 9:40 a.m.
197 views
How to convince organization to adopt a new technology
Daria Mühlethaler, Swisscom / Zürich, Switzerland
26 Nov. 2016 · 10:14 a.m.
116 views
Q&A - How to convince organization to adopt a new technology
Daria Mühlethaler, Swisscom / Zürich, Switzerland
26 Nov. 2016 · 10:38 a.m.
Animations for a better user experience
Lorica Claesson, Nordic Usability / Zürich, Switzerland
26 Nov. 2016 · 11:01 a.m.
121 views
Q&A - Animations for a better user experience
Lorica Claesson, Nordic Usability / Zürich, Switzerland
26 Nov. 2016 · 11:27 a.m.
Artificial Intelligence at Swisscom
Andreea Hossmann, Swisscom / Bern, Switzerland
26 Nov. 2016 · 1:01 p.m.
193 views
Q&A - Artificial Intelligence at Swisscom
Andreea Hossmann, Swisscom / Bern, Switzerland
26 Nov. 2016 · 1:29 p.m.
An introduction to TensorFlow
Mihaela Rosca, Google / London, England
26 Nov. 2016 · 2:01 p.m.
305 views
Q&A - An introduction to TensorFlow
Mihaela Rosca, Google
26 Nov. 2016 · 2:35 p.m.
Limbic system using Tensorflow
Gema Parreño Piqueras, Tetuan Valley / Madrid, Spain
26 Nov. 2016 · 3:31 p.m.
331 views
Q&A - Limbic system using Tensorflow
Gema Parreño Piqueras, Tetuan Valley / Madrid, Spain
26 Nov. 2016 · 4:04 p.m.
How Docker revolutionized the IT landscape
Vadim Bauer, 8gears AG / Zürich, Switzerland
26 Nov. 2016 · 4:32 p.m.
Closing Remarks
Jacques Supcik, Professeur, Filière Télécommunications, Institut iSIS, HEFr
26 Nov. 2016 · 5:11 p.m.
Rosie: clean use case framework
Jorge Barroso, Karumi / Madrid, Spain
27 Nov. 2016 · 10:05 a.m.
Q&A - Rosie: clean use case framework
Jorge Barroso, Karumi / Madrid, Spain
27 Nov. 2016 · 10:39 a.m.
The Firebase tier for your app
Matteo Bonifazi, Technogym / Cesena, Italy
27 Nov. 2016 · 10:49 a.m.
Q&A - The Firebase tier for your app
Matteo Bonifazi, Technogym / Cesena, Italy
27 Nov. 2016 · 11:32 a.m.
PERFMATTERS for Android
Hasan Hosgel, ImmobilienScout24 / Berlin, Germany
27 Nov. 2016 · 11:45 a.m.
Q&A - PERFMATTERS for Android
Hasan Hosgel, ImmobilienScout24 / Berlin, Germany
27 Nov. 2016 · 12:22 p.m.
Managing your online presence on Google Search
John Mueller, Google / Zürich, Switzerland
27 Nov. 2016 · 1:29 p.m.
Q&A - Managing your online presence on Google Search
John Mueller, Google / Zürich, Switzerland
27 Nov. 2016 · 2:02 p.m.
Design for Conversation
Henrik Vendelbo, The Digital Gap / Zurich, Switzerland
27 Nov. 2016 · 2:30 p.m.
Q&A - Design for Conversation
Henrik Vendelbo, The Digital Gap / Zurich, Switzerland
27 Nov. 2016 · 3:09 p.m.
Firebase with Angular 2 - the perfect match
Christoffer Noring, OVO Energy / London, England
27 Nov. 2016 · 4:05 p.m.
Q&A - Firebase with Angular 2 - the perfect match
Christoffer Noring, OVO Energy / London, England
27 Nov. 2016 · 4:33 p.m.
Wanna more fire? - Let's try polymerfire!
Sofiya Huts, JustAnswer / Lviv, Ukraine
27 Nov. 2016 · 5 p.m.
Q&A - Wanna more fire? - Let's try polymerfire!
Sofiya Huts, JustAnswer / Lviv, Ukraine
27 Nov. 2016 · 5:38 p.m.
Closing Remarks
Panel
27 Nov. 2016 · 5:44 p.m.

Recommended talks

Business Marketplace de Swisscom: la solution pour commercialiser votre application en Suisse
Joël Tendon, Online Experience Designer, Swisscom (Suisse) SA
11 Oct. 2012 · 6:19 p.m.
Comment être entrepreneur et social impacter
Laura Tocmacov, Co-fondatrice du Social Innovation Fund
29 April 2016 · 4:29 p.m.