1
00:00:00,960 --> 00:00:02,840
hi everyone and thanks for attending

2
00:00:02,840 --> 00:00:05,440
this talk I'm going to talk about our

3
00:00:05,440 --> 00:00:07,040
research which is on tracking and

4
00:00:07,040 --> 00:00:09,519
advertising on children website I'm

5
00:00:09,519 --> 00:00:11,840
Zahra a phda student from redb

6
00:00:11,840 --> 00:00:14,160
University and this is a joint work with

7
00:00:14,160 --> 00:00:17,080
my colleague from redb University Ken

8
00:00:17,080 --> 00:00:20,080
and University

9
00:00:20,080 --> 00:00:22,760
B on the modern web trackers and

10
00:00:22,760 --> 00:00:26,439
advertisers often exploit online user um

11
00:00:26,439 --> 00:00:28,359
with online user Behavior without their

12
00:00:28,359 --> 00:00:30,840
content and uh they create a detailed

13
00:00:30,840 --> 00:00:32,719
profile of a user including their

14
00:00:32,719 --> 00:00:34,719
preference and interest for monetizing

15
00:00:34,719 --> 00:00:38,559
users actually for targeted

16
00:00:38,559 --> 00:00:41,200
advertising also there are other kind of

17
00:00:41,200 --> 00:00:43,600
ads that are disturbing in their content

18
00:00:43,600 --> 00:00:46,280
inappropriate and disgusting to to

19
00:00:46,280 --> 00:00:48,960
attract attention a recent research from

20
00:00:48,960 --> 00:00:51,000
University of Washington focused on the

21
00:00:51,000 --> 00:00:52,800
content of the advertisement they

22
00:00:52,800 --> 00:00:54,920
conducted a survey to determine the type

23
00:00:54,920 --> 00:00:58,160
of ads uh that uh people do not like

24
00:00:58,160 --> 00:01:01,640
them and in a later study they uh find a

25
00:01:01,640 --> 00:01:03,600
lot of this inappropriate advertisement

26
00:01:03,600 --> 00:01:07,840
on news and misinformation

27
00:01:08,439 --> 00:01:11,439
websites so while online tracking and

28
00:01:11,439 --> 00:01:13,759
targeted advertising can pose a threat

29
00:01:13,759 --> 00:01:16,759
to the user of all age children are much

30
00:01:16,759 --> 00:01:19,680
more vulnerable uh prior research has

31
00:01:19,680 --> 00:01:21,720
shown that children tend to Mis

32
00:01:21,720 --> 00:01:23,040
misunderstand the message of

33
00:01:23,040 --> 00:01:25,680
advertisement and tend to focus more on

34
00:01:25,680 --> 00:01:29,000
a negative side of the message for for

35
00:01:29,000 --> 00:01:31,280
instance gambling GS and food

36
00:01:31,280 --> 00:01:34,040
advertising but despite the critical

37
00:01:34,040 --> 00:01:36,520
nature of this issue there is uh there

38
00:01:36,520 --> 00:01:39,200
is not a larger scale study focusing on

39
00:01:39,200 --> 00:01:41,799
um website directed at this vulnerable

40
00:01:41,799 --> 00:01:44,680
group so uh to bridge this Gap we

41
00:01:44,680 --> 00:01:46,960
investigated tracking and advertising

42
00:01:46,960 --> 00:01:50,520
practice on child directed

43
00:01:50,520 --> 00:01:53,680
websites of course uh these threats are

44
00:01:53,680 --> 00:01:55,759
are so important that governments around

45
00:01:55,759 --> 00:01:58,439
the world are also trying to take some

46
00:01:58,439 --> 00:02:01,520
serious actions for for instance in the

47
00:02:01,520 --> 00:02:04,520
in the EU uh DSA forbids targeting

48
00:02:04,520 --> 00:02:07,600
children with ads based on profiling or

49
00:02:07,600 --> 00:02:10,840
in the US under Copa data collection for

50
00:02:10,840 --> 00:02:13,720
Target ad advertising on this website is

51
00:02:13,720 --> 00:02:16,360
only allowed after getting uh verifiable

52
00:02:16,360 --> 00:02:19,680
parental consent and actually our study

53
00:02:19,680 --> 00:02:22,040
is timely due to this upcoming and

54
00:02:22,040 --> 00:02:24,599
recent changes in laws that aim to

55
00:02:24,599 --> 00:02:28,640
protect uh children um on the internet

56
00:02:28,640 --> 00:02:31,160
so we wonder what the state of the

57
00:02:31,160 --> 00:02:34,120
targeted and personalized advertising uh

58
00:02:34,120 --> 00:02:36,440
showing to children

59
00:02:36,440 --> 00:02:39,480
now so we conducted this empirical uh

60
00:02:39,480 --> 00:02:42,360
measurement research and we measure

61
00:02:42,360 --> 00:02:45,200
advertise advertisement including uh

62
00:02:45,200 --> 00:02:47,200
targeted and improper

63
00:02:47,200 --> 00:02:50,159
advertisements uh so we focus on U

64
00:02:50,159 --> 00:02:53,000
especially on four categories that PR

65
00:02:53,000 --> 00:02:55,480
research and regulat reports argue that

66
00:02:55,480 --> 00:02:57,840
these categories can be specifically

67
00:02:57,840 --> 00:03:00,640
harmful to Children including dating

68
00:03:00,640 --> 00:03:03,680
Mental Health weight loss and rad we

69
00:03:03,680 --> 00:03:07,120
also uh we also uh detect third party

70
00:03:07,120 --> 00:03:09,959
trackers on children website and also

71
00:03:09,959 --> 00:03:12,519
fingerprinting

72
00:03:12,519 --> 00:03:15,120
attempts in our study we Face some

73
00:03:15,120 --> 00:03:16,720
challenges first of all there is not a

74
00:03:16,720 --> 00:03:18,360
comprehensive and up to- date list of

75
00:03:18,360 --> 00:03:21,760
children website there are uh some uh

76
00:03:21,760 --> 00:03:24,840
list including a few uh a few children

77
00:03:24,840 --> 00:03:27,400
website which are were handpicked by

78
00:03:27,400 --> 00:03:29,799
expert uh but they're not enough for

79
00:03:29,799 --> 00:03:32,599
larg scale study and uh another

80
00:03:32,599 --> 00:03:34,680
challenges that we faced was um

81
00:03:34,680 --> 00:03:36,599
detecting advertisement especially

82
00:03:36,599 --> 00:03:39,439
targeted and improper ads automatically

83
00:03:39,439 --> 00:03:42,439
and to do so we needed to scrape adds

84
00:03:42,439 --> 00:03:45,519
and a disclosure

85
00:03:45,519 --> 00:03:48,480
page so to overcome the first challenge

86
00:03:48,480 --> 00:03:51,040
we built our own repository of children

87
00:03:51,040 --> 00:03:54,439
website um using uh this existing list

88
00:03:54,439 --> 00:03:56,799
and also online categorization Services

89
00:03:56,799 --> 00:03:59,040
we built a machine learning classifier

90
00:03:59,040 --> 00:04:01,799
that detect children website based on

91
00:04:01,799 --> 00:04:04,799
their title and uh metadata such as

92
00:04:04,799 --> 00:04:07,560
description then uh we applied this

93
00:04:07,560 --> 00:04:11,079
classifier to a to a very large data set

94
00:04:11,079 --> 00:04:14,760
of website content uh including several

95
00:04:14,760 --> 00:04:17,959
billions of several billion pages cwed

96
00:04:17,959 --> 00:04:21,720
from the web to identify this website

97
00:04:21,720 --> 00:04:24,160
and since we wanted to have a reliable

98
00:04:24,160 --> 00:04:27,600
and robust list of um list of children

99
00:04:27,600 --> 00:04:29,800
website we conducted a manual review to

100
00:04:29,800 --> 00:04:33,479
remove uh false positive and speaking of

101
00:04:33,479 --> 00:04:35,720
the classifier we use an existing model

102
00:04:35,720 --> 00:04:38,840
a prein model that use um that was

103
00:04:38,840 --> 00:04:41,160
trained on a on a huge data set of texes

104
00:04:41,160 --> 00:04:44,039
with a billions of parameter but there

105
00:04:44,039 --> 00:04:46,360
uh there are some small models called

106
00:04:46,360 --> 00:04:48,840
distill model a tiny and efficient model

107
00:04:48,840 --> 00:04:51,680
that learned that learned from a very

108
00:04:51,680 --> 00:04:54,479
large teacher model and they can distill

109
00:04:54,479 --> 00:04:57,080
the knowledge of bigger model and with a

110
00:04:57,080 --> 00:04:59,840
comparable accuracy

111
00:04:59,840 --> 00:05:02,320
and also uh the model that we picked was

112
00:05:02,320 --> 00:05:05,840
multilingual it support uh around 50 uh

113
00:05:05,840 --> 00:05:08,160
50 different languages then we could

114
00:05:08,160 --> 00:05:12,759
detect uh children website on different

115
00:05:13,520 --> 00:05:16,120
languages after preparing the data set

116
00:05:16,120 --> 00:05:18,400
for crawling process we utilize an

117
00:05:18,400 --> 00:05:21,039
existing open source crawler but uh we

118
00:05:21,039 --> 00:05:24,280
modified it heavily uh our crawler is an

119
00:05:24,280 --> 00:05:26,160
extended version of tracker Rider

120
00:05:26,160 --> 00:05:27,800
collector which is a web crawler

121
00:05:27,800 --> 00:05:31,560
developed by uh Dr go uh the scrawler

122
00:05:31,560 --> 00:05:33,840
interacts with the web page and records

123
00:05:33,840 --> 00:05:37,680
different type of um data during the CW

124
00:05:37,680 --> 00:05:40,600
specifically add related uh and add

125
00:05:40,600 --> 00:05:43,039
disclosure data uh fingerprinting

126
00:05:43,039 --> 00:05:46,520
related function called um HTP request

127
00:05:46,520 --> 00:05:49,400
respond and Etc and it also can record a

128
00:05:49,400 --> 00:05:52,560
video of the

129
00:05:54,080 --> 00:05:56,880
recording uh with this study we measured

130
00:05:56,880 --> 00:05:59,479
whether at targeting is enabled or Not

131
00:05:59,479 --> 00:06:02,039
by by scraping at disclosure page to

132
00:06:02,039 --> 00:06:04,639
accomplish this our interactive C our

133
00:06:04,639 --> 00:06:06,880
interactive crawler first detect and

134
00:06:06,880 --> 00:06:10,039
scrape advertisement and then it clicked

135
00:06:10,039 --> 00:06:13,120
on the add disclosure button and scraped

136
00:06:13,120 --> 00:06:15,720
the information from a disclosure page

137
00:06:15,720 --> 00:06:18,160
and later uh we uh we use this

138
00:06:18,160 --> 00:06:21,039
information to determine if um ad

139
00:06:21,039 --> 00:06:23,080
targeting or ad personalization is

140
00:06:23,080 --> 00:06:25,479
enabled for that that specific website

141
00:06:25,479 --> 00:06:27,680
or

142
00:06:27,680 --> 00:06:31,479
not uh we then of crawler uh to craw

143
00:06:31,479 --> 00:06:33,479
children website from different Vantage

144
00:06:33,479 --> 00:06:37,919
points to EU City uh like uh frankurt

145
00:06:37,919 --> 00:06:41,520
and Amsterdam one non City like London

146
00:06:41,520 --> 00:06:44,440
and two City from from us New York City

147
00:06:44,440 --> 00:06:47,919
and San Francisco we also did a limited

148
00:06:47,919 --> 00:06:50,440
measurement on mobile to see how ads

149
00:06:50,440 --> 00:06:53,639
change when using mobile

150
00:06:53,639 --> 00:06:57,000
um so this data are uh this data is

151
00:06:57,000 --> 00:06:59,599
based on more than uh more than 70

152
00:06:59,599 --> 00:07:03,479
20,000 Pages uh collected from 2,000 uh

153
00:07:03,479 --> 00:07:07,720
websites and we also take five inner

154
00:07:07,720 --> 00:07:10,199
pages from each website uh it was the

155
00:07:10,199 --> 00:07:15,199
scroll is not are not just based on uh

156
00:07:15,360 --> 00:07:18,960
homepages okay now let's uh let's turn

157
00:07:18,960 --> 00:07:21,720
to the finding of our

158
00:07:21,720 --> 00:07:24,960
study here is an overview of our result

159
00:07:24,960 --> 00:07:28,919
our CER scrape more than 70 uh 70,000

160
00:07:28,919 --> 00:07:33,120
advertisement from 8004 different uh

161
00:07:33,120 --> 00:07:36,639
children websites and we found the

162
00:07:36,639 --> 00:07:40,919
advertisement on 36% of this we websites

163
00:07:40,919 --> 00:07:44,319
and more importantly on average 27% of

164
00:07:44,319 --> 00:07:46,080
this website contain targeted

165
00:07:46,080 --> 00:07:48,080
advertisement a practice which is

166
00:07:48,080 --> 00:07:51,120
restricted or Bann based on U regulation

167
00:07:51,120 --> 00:07:55,199
or policy and over 70% of this um

168
00:07:55,199 --> 00:07:57,199
advertisement that we capture are

169
00:07:57,199 --> 00:07:59,560
targeted uh meaning that they enable

170
00:07:59,560 --> 00:08:02,319
build ad targeting or ad personalization

171
00:08:02,319 --> 00:08:05,400
so non-targeted ad or minority even for

172
00:08:05,400 --> 00:08:09,080
for children website uh we also did a

173
00:08:09,080 --> 00:08:11,639
exploratory analysis of whether ads on

174
00:08:11,639 --> 00:08:13,360
child directed website link to a

175
00:08:13,360 --> 00:08:15,840
malicious page or not uh we took a

176
00:08:15,840 --> 00:08:20,240
sample and um we found uh 150 out of

177
00:08:20,240 --> 00:08:23,120
4,000 scan link uh that they were

178
00:08:23,120 --> 00:08:25,360
flagged as a malicious or fishing by at

179
00:08:25,360 --> 00:08:28,599
least one um scan engine uh from uh

180
00:08:28,599 --> 00:08:31,120
virus total

181
00:08:31,120 --> 00:08:32,958
you might wonder who are this website

182
00:08:32,958 --> 00:08:35,279
that are showing Target advertisement to

183
00:08:35,279 --> 00:08:37,839
Children um here in this plot you can

184
00:08:37,839 --> 00:08:40,839
see the popularity rank of this uh this

185
00:08:40,839 --> 00:08:43,320
website and actually this is one of the

186
00:08:43,320 --> 00:08:46,760
positive finding of our studies and um

187
00:08:46,760 --> 00:08:48,480
as you can see in the pinkish bar the

188
00:08:48,480 --> 00:08:50,680
more popular the website is the more

189
00:08:50,680 --> 00:08:54,399
likely to show non-targeted ads to

190
00:08:54,399 --> 00:08:58,600
children so uh top uh top websites the

191
00:08:58,600 --> 00:09:00,800
more popular we website turn of AD

192
00:09:00,800 --> 00:09:03,200
personalization uh more often than low

193
00:09:03,200 --> 00:09:05,120
traffic

194
00:09:05,120 --> 00:09:07,480
websites uh we also looked into the

195
00:09:07,480 --> 00:09:09,680
content of the advertisement including

196
00:09:09,680 --> 00:09:13,360
text and images to detect improper ads

197
00:09:13,360 --> 00:09:15,120
there are some specific category that

198
00:09:15,120 --> 00:09:17,440
should not occur on children website per

199
00:09:17,440 --> 00:09:20,720
policy or regulation like um dating

200
00:09:20,720 --> 00:09:24,240
racing Etc to identify ads containing

201
00:09:24,240 --> 00:09:27,200
clickback Ry content uh we use the

202
00:09:27,200 --> 00:09:29,399
Google Cloud Vision API Ser search

203
00:09:29,399 --> 00:09:31,959
detection and in addition uh using the

204
00:09:31,959 --> 00:09:34,959
multilingual model that allowed us to

205
00:09:34,959 --> 00:09:37,640
query for some Search terms like uh

206
00:09:37,640 --> 00:09:41,560
dating Etc uh and it Returns the related

207
00:09:41,560 --> 00:09:44,480
um related um related ads to those

208
00:09:44,480 --> 00:09:46,680
Search terms the actually the model

209
00:09:46,680 --> 00:09:48,680
captur semantic similarity between the

210
00:09:48,680 --> 00:09:51,760
Search terms and uh Tes extracted from

211
00:09:51,760 --> 00:09:54,000
advertisements and it can also be used

212
00:09:54,000 --> 00:09:57,839
for other for other categories and at

213
00:09:57,839 --> 00:10:00,200
the end we did a manual you to remove

214
00:10:00,200 --> 00:10:02,880
the false

215
00:10:03,360 --> 00:10:05,839
positives uh this table shows the number

216
00:10:05,839 --> 00:10:08,360
of number of improper advertisement we

217
00:10:08,360 --> 00:10:12,320
identified in uh in each CW uh amounting

218
00:10:12,320 --> 00:10:15,720
to more than thousand improper

219
00:10:15,720 --> 00:10:18,680
advertisement across 311 distinct

220
00:10:18,680 --> 00:10:21,880
websites children websites and uh we

221
00:10:21,880 --> 00:10:24,440
stop manual review after reviewing

222
00:10:24,440 --> 00:10:27,680
thousand ads for each uh for each

223
00:10:27,680 --> 00:10:30,839
categories so these uh numbers are lower

224
00:10:30,839 --> 00:10:33,440
but uh just to to have a sense of

225
00:10:33,440 --> 00:10:36,360
whether uh this improper ads exist on

226
00:10:36,360 --> 00:10:37,880
children website or not and

227
00:10:37,880 --> 00:10:40,880
unfortunately they

228
00:10:42,519 --> 00:10:45,880
do now I'm going to show you some images

229
00:10:45,880 --> 00:10:49,240
uh we uh that we found in our study that

230
00:10:49,240 --> 00:10:51,760
are not safe for

231
00:10:51,760 --> 00:10:54,680
work these are some samples of improper

232
00:10:54,680 --> 00:10:57,920
ads we detected on children website uh

233
00:10:57,920 --> 00:11:00,079
we found ads related to weight loss

234
00:11:00,079 --> 00:11:01,560
mental

235
00:11:01,560 --> 00:11:05,639
health race as and etc for instance uh

236
00:11:05,639 --> 00:11:08,720
figure B features an image um image of

237
00:11:08,720 --> 00:11:10,800
ice cream that uh that could be

238
00:11:10,800 --> 00:11:13,440
appealing to children

239
00:11:13,440 --> 00:11:16,959
but uh this is an advertisement about

240
00:11:16,959 --> 00:11:20,160
largest online uh sex toy shop in

241
00:11:20,160 --> 00:11:23,760
Germany uh the other one in figure a uh

242
00:11:23,760 --> 00:11:26,800
is an advertisement from Alibaba uh

243
00:11:26,800 --> 00:11:29,320
display adds featuring racing and this

244
00:11:29,320 --> 00:11:32,519
there being and the stateful images to

245
00:11:32,519 --> 00:11:36,560
IM um images of product sold on Alibaba

246
00:11:36,560 --> 00:11:39,279
alibaba.com and uh these are mostly for

247
00:11:39,279 --> 00:11:41,800
grabbing

248
00:11:44,240 --> 00:11:47,320
attention we uh manually visited some of

249
00:11:47,320 --> 00:11:49,120
these websites showing targeted

250
00:11:49,120 --> 00:11:51,839
advertisement from our computer to see

251
00:11:51,839 --> 00:11:54,760
if uh still we can we can find this

252
00:11:54,760 --> 00:11:57,560
improper ads or not for example here in

253
00:11:57,560 --> 00:12:01,839
this website um like uh which is a uh Vu

254
00:12:01,839 --> 00:12:03,880
uh junior which is a kids activity

255
00:12:03,880 --> 00:12:06,600
obviously for children you can see this

256
00:12:06,600 --> 00:12:08,880
advertisement here in the the bottom of

257
00:12:08,880 --> 00:12:11,600
the page uh which is written in Dutch go

258
00:12:11,600 --> 00:12:14,639
never go to bed alone but it's a flirt

259
00:12:14,639 --> 00:12:18,320
finder it's scams and AI

260
00:12:18,320 --> 00:12:22,480
chatbot or in another example um that we

261
00:12:22,480 --> 00:12:25,720
found in a website uh which is for for

262
00:12:25,720 --> 00:12:29,680
for children like uh to to uh to learn

263
00:12:29,680 --> 00:12:32,360
count numbers to 20 and this kind of a

264
00:12:32,360 --> 00:12:35,240
stuff we found this amig Gates which uh

265
00:12:35,240 --> 00:12:37,480
which are most of the time

266
00:12:37,480 --> 00:12:40,880
scammy or in another website including

267
00:12:40,880 --> 00:12:44,839
Kinder Garten material uh we detect this

268
00:12:44,839 --> 00:12:49,399
um this improp advertisement here

269
00:12:50,360 --> 00:12:53,959
again uh you might wonder who are this

270
00:12:53,959 --> 00:12:56,519
advertisers uh in that disclosure

271
00:12:56,519 --> 00:12:59,440
interface you can also see advertisers

272
00:12:59,440 --> 00:13:02,880
identity actually Google tells you uh

273
00:13:02,880 --> 00:13:05,440
this ad is run by company but but this

274
00:13:05,440 --> 00:13:07,920
is not not for all ad this is just for

275
00:13:07,920 --> 00:13:10,519
some of the ads and as you can see here

276
00:13:10,519 --> 00:13:13,320
in this table advertisers from uh other

277
00:13:13,320 --> 00:13:16,120
part of the world uh like from Cypress

278
00:13:16,120 --> 00:13:19,199
or United Arabic Emir or is like isra

279
00:13:19,199 --> 00:13:22,160
are showing ads uh to children in the EU

280
00:13:22,160 --> 00:13:25,560
or in the US visiting this

281
00:13:25,560 --> 00:13:28,360
website uh we also compare trackers and

282
00:13:28,360 --> 00:13:30,240
website with with advertisement and

283
00:13:30,240 --> 00:13:32,959
without advertisement if a website has

284
00:13:32,959 --> 00:13:36,279
ads uh there is a lot of tracking going

285
00:13:36,279 --> 00:13:40,040
on and also we we compare trackers

286
00:13:40,040 --> 00:13:43,920
between um the EU and US CR and as you

287
00:13:43,920 --> 00:13:45,920
can see in this figure there are more

288
00:13:45,920 --> 00:13:49,600
tracker on on us on us visit so more

289
00:13:49,600 --> 00:13:52,800
tracking on a US visit more tracking on

290
00:13:52,800 --> 00:13:56,320
um on website uh with

291
00:13:56,320 --> 00:13:59,360
advertisements uh this table shows uh it

292
00:13:59,360 --> 00:14:01,160
with the most number of distinct

293
00:14:01,160 --> 00:14:04,240
trackers uh for example in the New York

294
00:14:04,240 --> 00:14:07,120
City Crow uh visiting a website like

295
00:14:07,120 --> 00:14:11,360
Masson work.com we found 161 distinct

296
00:14:11,360 --> 00:14:14,160
third party trackers in just one in a

297
00:14:14,160 --> 00:14:17,480
single website uh whereas in the EU uh

298
00:14:17,480 --> 00:14:20,959
visiting W space.com it trigger 95

299
00:14:20,959 --> 00:14:23,519
distinct trackers which is uh less than

300
00:14:23,519 --> 00:14:26,759
the US SC but still it is a

301
00:14:26,759 --> 00:14:29,959
lot so we reached out to five companies

302
00:14:29,959 --> 00:14:33,079
that we found uh to serve RAC as uh one

303
00:14:33,079 --> 00:14:35,600
of these companies invited us to a call

304
00:14:35,600 --> 00:14:37,959
and explain the likely reason for

305
00:14:37,959 --> 00:14:40,959
showing inappropriate advertisement and

306
00:14:40,959 --> 00:14:42,920
uh indicate that they will uh they will

307
00:14:42,920 --> 00:14:44,399
make further

308
00:14:44,399 --> 00:14:47,959
investigations uh we also disclosed 34

309
00:14:47,959 --> 00:14:50,480
racy ads to Google by manually visiting

310
00:14:50,480 --> 00:14:53,959
the ad disclosure URL of each RAC ad uh

311
00:14:53,959 --> 00:14:57,279
using the report this uh report this ad

312
00:14:57,279 --> 00:15:00,040
button uh in addition we shared our

313
00:15:00,040 --> 00:15:02,399
results with the with uh European data

314
00:15:02,399 --> 00:15:05,959
protection agencies uh consumer data uh

315
00:15:05,959 --> 00:15:08,920
consumer protection agencies and a UK

316
00:15:08,920 --> 00:15:11,639
based NGO focusing on uh children

317
00:15:11,639 --> 00:15:14,440
digital rights and lastly our paper will

318
00:15:14,440 --> 00:15:17,800
be present on U Kel's uh privacy

319
00:15:17,800 --> 00:15:20,040
research Day on on

320
00:15:20,040 --> 00:15:23,600
Paris so to sum up uh our study reveals

321
00:15:23,600 --> 00:15:25,519
the extensive tracking and frequent

322
00:15:25,519 --> 00:15:27,639
targeted and improper advertising on

323
00:15:27,639 --> 00:15:30,399
children website we found that popular

324
00:15:30,399 --> 00:15:33,680
websit are less likely to use targeted

325
00:15:33,680 --> 00:15:37,600
ads and uh we call for more research

326
00:15:37,600 --> 00:15:40,319
regulation and enforcement uh to protect

327
00:15:40,319 --> 00:15:44,720
children on privacy online um and lastly

328
00:15:44,720 --> 00:15:46,839
our source code and data is open source

329
00:15:46,839 --> 00:15:49,880
on GitHub uh make sure to check it out

330
00:15:49,880 --> 00:15:51,560
and thanks for attending this St and

331
00:15:51,560 --> 00:15:53,440
feel free to reach out if you have any

332
00:15:53,440 --> 00:15:56,440
questions