WEBVTT

1
00:00:00.001 --> 00:00:02.910
So as I start this that we good afternoon, everyone.

2
00:00:02.950 --> 00:00:31.699
Thank you for being here. Let me introduce myself. My name is Elaine Casa asthma, and I am a lecturer. I'm a researcher at the confluence, the University of Marie, the Spain. I'm going to cure this session, together with my colleague. Antonio Gavada, from the European University of Madrid to you are very welcome. Thank you for being here this afternoon. I have that you are enjoying these well organized Congress

3
00:00:31.830 --> 00:00:50.979
and Yesterday we had the first session of the big data when it was a complete success, and today we have the session big data, too. It's going to be a hybrid model with some panelists talking through the

4
00:00:51.270 --> 00:01:20.069
through the zoom platform. There is also one media. And there we have 2 panelists here in person. So maybe if you have any question, we will do questions after the talks, after after the panelists. That this position, their expositions are there for one each other. So I need to ask you to be a test of time, more or less between

5
00:01:20.070 --> 00:01:39.270
10 and 12 min as maximum to have a Q&A period of 2 min. So for those who are those penalties that are doing the talks through the show. My colleague. We just sent you some messages about about 3 min left.

6
00:01:39.370 --> 00:02:03.930
then 1 min there, and for for those who are going to do the talk here in version, we will just tell you. Let you know by showing you a paper with 3 or 1 min left, and that will give us the opportunity to have it. Maybe questions. And also I have a small or short debate. It depends on on the time you need to do your your talks.

7
00:02:04.070 --> 00:02:10.810
So let us start. Shall we, our first panelists today?

8
00:02:14.690 --> 00:02:38.630
It's going to be chiro, Clement, the developer of the University of Naples.

9
00:02:38.730 --> 00:02:41.199
so are they ready.

10
00:02:43.550 --> 00:02:56.979
Hi, everybody! I am Ferdinand. Be a threat, a Phd. Candidate from Sapiens and University of Rome, and I present this research in collaboration with the Shiloh, lamented the Falco reserves from

11
00:02:58.570 --> 00:03:00.970
See a little bit longer, please.

12
00:03:01.050 --> 00:03:12.919
for Italic or University of Rome. Gabriela Tunciano, Associate Professor from Federico Segundo University of Naples, and Dominique got it.

13
00:03:13.150 --> 00:03:18.799
Research, fair Love from Federico Segundo University of Maples.

14
00:03:20.070 --> 00:03:39.179
The results of the Zalan election in September the 2,022 Italian political parties showed distinct software demographic standardification indicating a link between voting patterns and specific population characteristics.

15
00:03:39.200 --> 00:03:49.600
At the same time there was a participation of prominent candidates and political leaders in this election on the most popular social media platforms.

16
00:03:50.070 --> 00:04:00.949
In this research we explore the relationship between voting target demographical characteristics and the political communication on the different platforms.

17
00:04:01.070 --> 00:04:19.409
Our goal is to understand, if the candidates use the different social platform to reach specific targets of the electorate and to explore. If there is a discernible relationship between David, communication strategies and the electoral resistance.

18
00:04:19.910 --> 00:04:38.009
the paper aims to present the first part of the analysis of the digital profiles of Italian political readers during the 2,022 actual campaign, and the possible association between the social demographic characteristics of the wood and the the platforms used.

19
00:04:38.390 --> 00:04:49.819
The results of the 2020 sure actually showed the extra sense of an association between the building and the social component of the electorate.

20
00:04:50.760 --> 00:04:53.909
How much did the leaders of social media presents.

21
00:05:06.010 --> 00:05:08.949
and the communication of these aspects

22
00:05:12.620 --> 00:05:33.170
in the scholars of electoral beaver questioned the relevance of the concept of a class voting in Italy, for example, in 2,001, the working class support for Silvio Berlusconi, the leader of the Liberal right, challenged the validity of this concept.

23
00:05:40.840 --> 00:05:52.360
However, with the advent of the Populist movement and the parties in the following years the social demographic issue for voting memory class voting regained attention

24
00:05:53.120 --> 00:06:09.839
according to some hold, also, such as the Crazy and his colleagues voting the photo Populist formation can be understood as a class voting as they are primarily supported by the so-called loss of globalization.

25
00:06:11.540 --> 00:06:26.790
These individualants have experienced the reduction in their, the purchasing power and the position in the lab or market, due to the changes brooked about to by globalization and accelerated by the global crisis of 2036.

26
00:06:26.860 --> 00:06:43.330
In Italy the 5 Star movement has been identified as it Populist movement, that this rapidly the political landscape entering in Parliament in 2,013, and going into Covarnament in 2,018.

27
00:06:45.550 --> 00:07:13.890
 what is it? Focus or marginalizer to people it to promote it? They see you so my political initiative. But in the it does in economic difficulty, including the the citizen income which generated the values so protest and the resistance in the country, despite its presence in other countries as well.

28
00:07:16.190 --> 00:07:27.960
Today social media represented the main environments where a political and electoral communication takes place, and to intercept the most of the electoral target audience.

29
00:07:27.980 --> 00:07:39.780
For example, platforms such as Instagram and Facebook, have become very popular in the recent years. In Italy alone, according to data from, we are searching, and

30
00:07:40.190 --> 00:07:51.350
2,000 to 22 more than half of the population over the age of 18 is registered on one of these 2 social networks.

31
00:07:53.550 --> 00:08:27.720
Today social platforms are emerging that are increasingly different in terms of the type of content, mode of communication, and structure. These specific characteristics attract to different audience, based on agent interest, and even ideology. Again, according to the we are a social report. Considering the main platforms, Facebook, Twitter, Instagram, and the Tik Tok, one finds a very sophon, a grammatically diversity audience.

32
00:08:32.140 --> 00:08:51.290
Therefore the analysis of political issue and the target must take this diversification into account. Indeed, the presence of the platforms with their specific audience could influence about the way political message are a comment, and the message is themselves.

33
00:08:52.600 --> 00:09:09.229
Political content can thus take different forms on a different social media. For example, on Instagram the communication is image-oriented. while on Tiktok there are a short video of this is done in more than streamlined, and the simple language.

34
00:09:10.090 --> 00:09:25.250
or this is an open way for additional research question. More oriented to the digital environments is the communication of politicians the same across the social platforms, or does it be fair

35
00:09:28.050 --> 00:09:54.010
in light of this contemplation it is to explore the extended to which the Italian political landscape has undergone a transformation as the result of a to 5 Star movement, particularly in in terms of theoretical guns of a naturalist, political narratives, and a discernible focus on a specific social strata within politician pronuncience.

36
00:09:57.100 --> 00:10:10.420
Equally not aworthy is the investigation into whether the politician choice of thematical content varies. Depending on the platform employed to engage with the different social media.

37
00:10:10.790 --> 00:10:20.129
The general election of 2023 provided an excellent opportunity to empirically test these foods.

38
00:10:29.430 --> 00:10:45.329
The research question are one of which issued the campaign. Basic, what are the social, economic categories mentioned by politicians? Is the communication of the politicians the same across the social platforms, or does it defer

39
00:10:45.400 --> 00:11:10.920
the politicians of a reference are Maloney, Berlusconi, Salvini, Contact Calendar, and the Leta exploring on a platform stick Tok, Facebook and Instagram number of analyzes the post to 200 to 55 to use it for a data extraction are a triangle and the Api ticking in the into consideration the only

40
00:11:11.010 --> 00:11:22.240
the election campaign in the period and the as an analysis, technically technique, we use it, the a content analysis for the for the type.

41
00:11:23.190 --> 00:11:24.120
How are you?

42
00:11:25.240 --> 00:11:40.580
There are 3 dimensions on emerging from the analysis. The first relates to communication which lacks a specific issue, but is so focused on empathy issues, such as the criticism without argument, supporting ones on coalition, etc.,

43
00:11:40.680 --> 00:11:54.619
does an absence of our augmentation on the real problems of the country members. The second, then, makes a continuous attack on up to it's revealed. It is a demonization of the political opponent.

44
00:11:55.740 --> 00:12:02.290
the, for example, led against the Meloni and the her coalition, but also Salvini is remarked.

45
00:12:02.740 --> 00:12:12.959
Confex Tech Support Martin: It does, in instance, that at least apparently the targeting, the strategy in the political communication of candidates is greater and marginal.

46
00:12:13.700 --> 00:12:28.459
On the other end, the focus on a specific targets or categories is limited, and if there is a little difference between one platform and another, thus paying little attention to specific audience or potential voters.

47
00:12:31.090 --> 00:12:38.799
Confex Tech Support Martin: the communication of the analyzing the politicians is not highly focused on specific issues.

48
00:12:39.000 --> 00:12:50.189
The majority of the analyzes the possible concentrate of the increasing opponents or protesters going as far as demonizing them.

49
00:12:50.880 --> 00:13:01.910
In Salvini's case there is a prevailing attention towards opposing factions and individuals. For example, frequent references to

50
00:13:02.740 --> 00:13:17.239
on the other end. In Melanie's communication there are many references to protesters who interrupt her speech. The campaign of Berlusconi and the

51
00:13:18.120 --> 00:13:41.359
Berlusconi's campaign leverages pensions-related topics, while contents campaign is a more, a century around the issue that have but characterizes that the five-star moment, such as the citizen income, but also includes the new topic, like the super bombs or military service.

52
00:13:41.630 --> 00:13:55.170
Additionally, there is a mention for an environmental issue which is a present in the contents and the calendar's communication with the focus on topic like school orropatra.

53
00:13:58.970 --> 00:14:22.129
Regarding the god's second and resolution only Conte seems to make constant references to specific socioeconomic issue at the targets. In fact, his campaign was based on defending and enhancing that the minimum income policy, and there for his Hagans was mainly the socially disadvantaged.

54
00:14:25.900 --> 00:14:47.980
Another that we mentioned the category is a pensioners. Men shot by Berus Kuni. The others, except that we's a radical dimensions issue affecting the young generation like a climate change, have no specific issue or targets with their respects to the general general electorate

55
00:14:52.660 --> 00:15:07.799
currently enter the Western is the communication of the politicians the same Microsoft social platforms, or does Italy first, as mentioned, that the does not, seems to be any significant difference between one platform and another.

56
00:15:08.250 --> 00:15:20.119
at least apparently, than targeting the anecdote on the Socialist not to be in the prevalent strategy, or at least it so, has not received the success

57
00:15:22.970 --> 00:15:33.610
on the other end. This could be a consequence of the little differentiation of the issue mentioned earlier. There are, of course, if you are exceptionally

58
00:15:33.770 --> 00:15:49.620
among them, a centrally Salvini, who makes a very different use of a tick-tock than the other platforms, while Nevertheless, Targeting an outdoor audience, not really that of the platform.

59
00:15:52.550 --> 00:16:04.269
And. in conclusion, it is interesting to note that the post-analyses do not toberglom in entrepreneurs and materialistic and post-materialistic teams.

60
00:16:05.260 --> 00:16:19.639
Instead, they seems to revolve around the empty teams. Moreover, political communication lacks a significant socioeconomic and demographic targeting strategy on social media platforms.

61
00:16:22.360 --> 00:16:37.089
As for the link between these results and the height, abstention, rating, and the last round of election. It is difficult to to establish a direct casual relationship without further data and analysis.

62
00:16:38.190 --> 00:16:56.700
however it is a plausible to speculate that to the lack of a specific and substantive communication from political leaders. who without they do, but did the not to contribute to the sharp increasing the trend of of both. And this engagement

63
00:16:59.180 --> 00:17:31.200
disconnect between the issue, address it, and the the concerns of the electorate that could potentially contribute to to lower and water turn out to further develop these analysis, it will be useful to conduct more in dep research and analysis, to explore the relationship between political leaders to communication strategies, the prevalence of empathy issue and the impact on voter turnout and the political participation.

64
00:17:31.610 --> 00:17:48.879
One step that to should definitely be token would be to include the post with the Heger engagement in the analysis to understand if the evidence that emerged her affected the overall electoral communication.

65
00:17:51.390 --> 00:17:55.879
I will finish it in my presentation. Thank you for your attention.

66
00:18:00.500 --> 00:18:21.380
Thank you very much for this presentation that if it was a a video and it has been a little bit long. There is no time for questions. I just want just to just pass a question for the audience, maybe to be discussed up there, and after the whole panelists have a disparation, the explanation or their presentations.

67
00:18:21.380 --> 00:18:46.379
But this maybe it's time to know, to cross another kind of data, not only because of, but also demographic as and shell demographic debt data. In order to know what are the how is that identity of the jump voters, the identity, social identity, the herbs, unemployment and employment rates that leads them to to vote for these

68
00:18:46.380 --> 00:18:54.219
new politics. All parties. I have a question there in the air. Maybe we have just time to discuss it later. Thank you very much.

69
00:18:54.220 --> 00:19:20.560
So let's move on. Shall we? Now, with our next? Our next panelists, who is going to be Professor Robles from the Competency University of Madi, and his talk is about the Ital activism, masked the Fridays for future movement and the global day on Plymouth action, testing social function and framing typologies of planes on Twitter.

70
00:19:20.980 --> 00:19:37.440
Thank you. Okay, thank you very much. Berlin for your introduction. It's a real pressure to be here with you. First of all, I would like to highlight that this presentation aims to disseminate a work done with 101,

71
00:19:41.870 --> 00:19:55.810
all of us, all of us. We are all members of our research group from Kropotensa University in Madrid. This research group is called data science and computing for social analytics.

72
00:19:56.290 --> 00:20:16.920
And this work is about a collective action in digital social networks. More specifically, this work is focused on the process of collective action carried out by the movement Friday for future during the the global day of climate action 2 years ago.

73
00:20:17.260 --> 00:20:27.040
But without a further delay. Here you have the milestones of this presentation.

74
00:20:27.230 --> 00:20:28.820
5 points.

75
00:20:29.010 --> 00:20:44.390
first, a brief introduction, and second, a review of the literature about this movement and Friday for future, and the connection between this movement and social network analysis.

76
00:20:44.630 --> 00:20:51.560
the first data methods, some results, and finally conclusions.

77
00:20:52.070 --> 00:20:53.240
Well.

78
00:20:53.300 --> 00:21:19.769
as you know, social media, especially in the Twitter. Are very important sources for analyzing. Several kind of political processes, for example. the climate change! But not only that, thanks to the the social media, the way in which people participate citizens participates are changed.

79
00:21:20.070 --> 00:21:39.580
We have now some general theoretical concepts, like collective action or mass self-communication. We take all these concepts, and from the literature as inspiration for this research.

80
00:21:40.090 --> 00:22:03.639
you have to know that a Friday for future is a very important movement. Organization. Part of the climate. Jestive movement is it's a very peculiar organization, because because it's basically formed by John people. And they are very important power of modernization around the world.

81
00:22:03.770 --> 00:22:17.979
So for this and other reasons, we consider that this organization is very important to take it for, for to take it as

82
00:22:18.170 --> 00:22:26.530
case of a staff. So the second point of my presentation is literature revenue.

83
00:22:26.800 --> 00:22:51.260
We used for this literature. Review English contributions in the literature in the scientific literature. and several data sources, social media data and other sources.

84
00:22:51.440 --> 00:23:08.059
So basically, we we did is to analyze or to take the information about. And this social movement from the the current literature about social movement and social network analysis.

85
00:23:08.210 --> 00:23:23.079
Thanks to this review, we found basically 2 big and focus of interest. First is around the traditional idea of social movement studies.

86
00:23:23.470 --> 00:23:44.500
and the second one is was about a frame. Framing literature. Framing, you know, is basically inspirated in the classical proposal contribution from a it's now bent for late in the in the last century.

87
00:23:44.860 --> 00:24:01.340
when we talk about social movement studies, we found that they, the author in this tradition, are interested in how this new media increase mobilization

88
00:24:02.160 --> 00:24:15.479
in opposition, when we analyze, when we read framing literature, we found that the authors in this tradition are more interested on

89
00:24:15.980 --> 00:24:25.829
measure, different types of frames. For example, the agnostic frame, pornnetic frame, or frame of mobilization.

90
00:24:26.770 --> 00:24:40.480
In some cases we found a comparation between both traditions, but normally the general tendency was 2 very separate

91
00:24:40.520 --> 00:24:47.430
schools of analysis of this movement and the use of social network of Friday for future

92
00:24:49.320 --> 00:25:01.219
thanks to these 2. So, thanks to this review of the literature, we proposed it for research questions. 3 of them

93
00:25:01.420 --> 00:25:05.450
was theoretical questions

94
00:25:05.520 --> 00:25:14.640
per second, and one of them, the last one, was a question for a methodological point of view.

95
00:25:15.100 --> 00:25:24.659
basically we try to understand. Thus the faction of tweets change within a covid context.

96
00:25:25.810 --> 00:25:35.099
Second, we want to know more. We want to have more information about the main frames of tweets.

97
00:25:35.780 --> 00:25:43.529
And what is the relationship between the social faction, the first context of the literature, and the second one, the framing context.

98
00:25:44.150 --> 00:25:59.000
and finally, the last question, the methodological one was around, to what? To what extent automated data classification processes work across and between typologies.

99
00:25:59.450 --> 00:26:08.440
So to what extent these techniques are or not useful for this kind of research.

100
00:26:09.810 --> 00:26:18.769
please take into account. We have here 3 theoretical questions. The first one is about

101
00:26:19.170 --> 00:26:20.510
how

102
00:26:21.000 --> 00:26:31.239
these social network techniques can can help us to understand the main concerns from the first.

103
00:26:31.420 --> 00:26:43.859
theoretical tradition, social movement studies. And the second one is basically to try to understand, to go further, to understand, oh, sorry.

104
00:26:43.930 --> 00:27:02.549
E. A. Framing literature that the first one is about. What of this? Which of this tradition are the best one or more informative, and finally, the last. The last objective for the last question is about a methodology, right? So one.

105
00:27:02.900 --> 00:27:07.629
Now, some keywords about data and methodology.

106
00:27:08.660 --> 00:27:18.619
what data collecting, we, we developed the the data from Twitter's Api.

107
00:27:18.630 --> 00:27:20.740
and in a streaming.

108
00:27:20.920 --> 00:27:26.779
the global day of climate action, basically

109
00:27:27.930 --> 00:27:29.920
3 days, not 4 days.

110
00:27:31.170 --> 00:27:36.290
We focus on the English language tweets.

111
00:27:37.480 --> 00:27:48.090
And here you have the end for this research. Basically the half of the whole data set.

112
00:27:49.060 --> 00:27:56.750
We. We prepared a a code book with the categories of

113
00:27:57.420 --> 00:28:02.490
the 2 traditions, social movement, study, and framing framing literature.

114
00:28:03.290 --> 00:28:06.950
and we made a manual codification

115
00:28:07.000 --> 00:28:14.629
and finally, an an automatic classification process. You have more details in the presentation with Powerpoint

116
00:28:15.990 --> 00:28:30.010
about the methods. We made some different. We applied some different methodologies, descriptive statistics, confusion, matrix tables.

117
00:28:30.020 --> 00:28:34.129
and and so on. You have more information there.

118
00:28:34.800 --> 00:28:35.870
Oh, thank you

119
00:28:36.950 --> 00:28:41.580
a briefly, son.

120
00:28:41.970 --> 00:28:43.949
Words about the results

121
00:28:45.190 --> 00:28:51.659
you have in the in the in the left side

122
00:28:51.800 --> 00:29:04.420
categories for the variable social faction, topology, first division, and in the in the right. table, you have the categories for a framing topology.

123
00:29:04.620 --> 00:29:08.440
As you can see, the most popular.

124
00:29:08.970 --> 00:29:19.879
a category in social faction. Topology is opinion. So this is very informative, because the health of the Tweets.

125
00:29:19.900 --> 00:29:22.590
and during this process

126
00:29:22.840 --> 00:29:24.220
a

127
00:29:24.620 --> 00:29:28.279
from people of this movement.

128
00:29:28.290 --> 00:29:34.669
What about opinions in general terms, especially opinion about the protest?

129
00:29:34.790 --> 00:29:42.229
It's more than it's more a debate than a collective action process.

130
00:29:42.440 --> 00:29:49.730
But at the same time, when you focus on the framing topology.

131
00:29:49.970 --> 00:29:55.310
you can see prognosis especially. and

132
00:29:55.360 --> 00:30:00.519
legislation and policy change is the most popular category.

133
00:30:03.010 --> 00:30:09.340
After that we wanted to know to what extend both categories

134
00:30:09.540 --> 00:30:11.090
or or not.

135
00:30:11.140 --> 00:30:15.890
 are not independent.

136
00:30:16.190 --> 00:30:22.060
And you can see here, thanks to this table that both categories are not in the

137
00:30:22.820 --> 00:30:23.910
so

138
00:30:24.160 --> 00:30:28.069
one tweet can be part of both categories.

139
00:30:29.140 --> 00:30:30.340
So.

140
00:30:31.560 --> 00:30:37.869
in conclusion, we have a 5 general conclusion first.

141
00:30:38.030 --> 00:30:39.330
and

142
00:30:40.570 --> 00:30:41.760
we

143
00:30:42.030 --> 00:30:49.669
didn't find an important change of mobilization fraction tweets

144
00:30:49.730 --> 00:30:52.939
after the COVID-19.

145
00:30:53.050 --> 00:31:04.749
The valence diversity of frames, the diagnosis for noxia mobilization with an popular number of tweets that embassies solution of the hehalation and policy change.

146
00:31:05.110 --> 00:31:10.009
Social faction and mobilization. Typology. Apologies, as I said.

147
00:31:10.050 --> 00:31:13.240
are not independent. But

148
00:31:13.450 --> 00:31:23.810
please, from my point of view, the most important result is automatic data classification model performed well, especially across social faction, topology, and other category.

149
00:31:24.160 --> 00:31:32.329
Finally, general and a specific general essential and a specific word of the most discriminant tweets.

150
00:31:33.050 --> 00:31:46.730
And after finish my presentation I would like to inform you about this special issue we are preparing if you want to send any proposal during the next weeks.

151
00:31:47.160 --> 00:31:50.270
So thank you very much for your attention.

152
00:31:53.210 --> 00:32:06.739
I don't know. Maybe there is a question among the audience today for Professor Broglas.

153
00:32:08.870 --> 00:32:13.079
Yes, please. let me just use the microphone and

154
00:32:13.110 --> 00:32:14.180
thank you.

155
00:32:16.040 --> 00:32:38.110
I I have a quick question. Thank you. First of all, I thought the the coding. And you know, going into that into more detail. We don't probably don't have the time, but that would be very interesting to me. but what I was wondering. Maybe I missed it. How you you had you use the streaming Api on 3 days around one doing the global climate action. Yes. And then

156
00:32:38.220 --> 00:33:02.100
in the end, the conclusion was that the the Covid pandemic did not change the the how. I I mean you don't really. Maybe I missed it. But did you compare now some of previously? Exactly. It is my fault, because I I didn't say that we compared results of this case of a static with a other before the pandemic.

157
00:33:02.200 --> 00:33:13.240
Sorry about that. My phone. No worries, thank you that it clears it up. Thank you. Thank you very much. I don't know whether we have some time of time to for another questions. We are just

158
00:33:13.290 --> 00:33:15.929
okay.

159
00:33:18.760 --> 00:33:35.549
and we've been on. So you know that all this people have enough time to to, for all the families have time, and next one is a subsidiary of Xavi. From the atos of law and a university and the faculty of Social Science

160
00:33:35.550 --> 00:33:51.780
Research Center for Computational Social Science in Hungary. And the name of this talk is investigating and dynamic social network of politicians constructed by the similarity similarity of their speeches.

161
00:33:52.140 --> 00:33:57.790
So I don't know whether it is ready. Yes, yes, I am so I can share my screen if it's okay.

162
00:33:58.000 --> 00:34:04.440
And yes, we can, all of all of us see your presentation so

163
00:34:05.210 --> 00:34:26.660
Zsófia Rakovics: so great. yes, thank you very much for having me here. I'm representing the Research Center for Computational Social science here, and I'm a Phd. Student at the professional and university in Budapest, Hungary, and I, I will ask. It was mentioned. Talk about the dynamic networks of politicians based on the similarity of their speeches. So let's continue, and let's say.

164
00:34:26.970 --> 00:34:47.659
Zsófia Rakovics: and start with the exciting part. I would like to start with with some motivation explaining why I think it's important to have this research or similar research like that. The main goal was to analyze the parliamentary speeches. we document embedding because I think it's a nice method to use when coding the

165
00:34:47.760 --> 00:35:09.729
Zsófia Rakovics: the the text of the the transcripts of speeches into vectors, and then trying to compare, or just to understand the relation of the speeches. So that's why it's a nice way of using, and that was a mango to try that method out on parliamentary speeches, and the goal was to the the further goal was to examine language and political polarization within the speeches.

166
00:35:09.730 --> 00:35:32.199
for the theoretical approach. I I both. It was nice to use the neo-popular bubble, the theory of neo populism. And I'm I'm referring him a book by Pete Chico, who is talking about the speculative, mediatized political public sphere. I think it's a nice and a theoretical framing for this research. That's why I'm having it here

167
00:35:32.279 --> 00:35:53.200
and about the level of the analysis, and there are 3 different layers we can. observe, and we can work on. The first one is the Parliament for speeches on the level of parliamentary speeches themselves. Then we can aggregate those and talk about the Parliamentary party representative. So the Mps. Of the poll, the the members of the Parliament.

168
00:35:53.210 --> 00:36:07.269
Zsófia Rakovics: and then we can also aggregate those results as well, and then have the level of political parties, which is also exciting to examine. So these are the 3 different levels. I will address in this presentation

169
00:36:07.580 --> 00:36:22.409
about the corpus and the and the the data I've used. It's related to the research of the research center of computational social science of, and it's related to the layers of the political public sphere. research!

170
00:36:22.420 --> 00:36:25.459
The corpus itself contains a

171
00:36:25.500 --> 00:36:38.239
Zsófia Rakovics: speeches between 1998 and 220 and the and it's kind of a huge core process for it in the bait big data session because these amount of speeches are not there and

172
00:36:38.240 --> 00:37:03.000
Zsófia Rakovics: easy to read through. That's why we are using these machine learning and national language processing techniques to you understand those. But for using those methods, we need to do some prep processing. So I was also working on the data cleaning that organization, the lemonization. These are steps which are necessarily has to be done to make the standard values of forms, and before the analysis to make it more advanced and more

173
00:37:03.000 --> 00:37:04.150
prepared for

174
00:37:04.150 --> 00:37:17.739
Zsófia Rakovics: For quantitative analysis. I have also applied named entity recognition. When we are when I have compounded words together which belong together, for example, names of institutions, etc., or names of people.

175
00:37:17.740 --> 00:37:38.410
I've also removed the Stopford and the compounded significant engrams words that are commonly used together, not as as part of a as institution or a name. But these are not named entities, but words that are really going together, and having a kind of a stronger relationship between them.

176
00:37:38.680 --> 00:38:02.840
Zsófia Rakovics: And I would like to talk about now about the about the documenting batting itself. for that I've chosen a verse movers in padding, and because of that I have started with with the word mover's distance. When we are using this method we have to compute the distance between the the lock documents, and for that there is a way. This is called vertical assistance. Kushner is now referred.

177
00:38:03.050 --> 00:38:31.710
and I'm it. It measures the similarity of 2 documents in the words in the word embedding space. It's good because we are not restricted to any. word, the embedding methods. But we can choose our preferred methods. I have used glove here, but it's also possible to use work to back a word embedding for this operation to have to word over distance if it's really flexible, and can be adjusted to the the preferred methods of the of the researcher.

178
00:38:31.830 --> 00:38:48.949
Zsófia Rakovics: What is A disadvantage, though, that it's really computational and storage intensive procedure. So if we have an a huge amount of documents, then it's a really huge matrix which will be constructed, and then it can be time and storage intensive.

179
00:38:48.950 --> 00:39:02.620
Zsófia Rakovics: and there is another procedure, a general procedure which needs to be used for the embedding. This is the distances to kennels and embeddings. here you can find the reference right? It's related to who and his colleagues

180
00:39:02.620 --> 00:39:09.860
who just worked on that. And I've used this to prepare or just get the kennel function of the distance metrics.

181
00:39:10.080 --> 00:39:33.739
Zsófia Rakovics: here. It's it's kind of the concept of the method the word movers embedding is using Co. The documents to try to compute the distances between documents. So in the left hand side of the plot you can see the verse, mover distance. It is the case when we have 2 different texts, document X and document Y, you can see some short sentences here, and we are just

182
00:39:33.900 --> 00:40:03.819
Zsófia Rakovics: computing the distance between the words or the d dictionaries of each document separately that you can see in the left hand side, and then the right hand side. You can see the black rectangles which are from at the other document generated for the computation of the distances, and with that we can and and spare some time and and and computation, because we will have, we will be able to generate a document and based on the differences of

183
00:40:03.820 --> 00:40:22.569
Zsófia Rakovics: the comp compared from the the of the document and the original documents we will be able to find and and compute the distances. And it's It's easier and helping us to go through quickly and then more efficiently on the problem. So that's what's on the right hand side, this word movers and padding the concept of it.

184
00:40:22.810 --> 00:40:42.990
Zsófia Rakovics: Let's continue. And now I would like to talk about the social network analysis of political parties, because that's I I have also targeted to try to understand the the social network of politicians and based on that, try to conclude some exciting social scientific right insights results.

185
00:40:42.990 --> 00:41:00.480
Zsófia Rakovics: So here we have 2 different levels of analysis. The First level is at the unit of the Parliamentary members and years. So I've aggregated the speeches for each Parliamentary members and each year, and then we can also have a go further and have a second level of analysis. It's a

186
00:41:00.490 --> 00:41:30.259
Zsófia Rakovics: unit of political party. And here, when we are constructing the aggregated volume for political parties based on the representatives of the political party, and by doing so we can construct the graph but which can be based on the similarity of the speeches. So in this analysis I have chosen to. It's chosen a threshold to see a and yeah chosen a threshold to measure the similarity of the speeches.

187
00:41:30.260 --> 00:41:44.940
Zsófia Rakovics: and if this threshold was exceeded by the the by 2 Mps, and then there was a link between the 2 people, and then it's it. It was kind of the the the and the method. I have constructed this graph.

188
00:41:44.940 --> 00:42:08.920
Zsófia Rakovics: It's just for visual understanding. It's just have food to construct graph based on the similarity of the speeches of the per the parliamentary members, and based on that, we can then conclude and go further and understand, for example, the homogeneity of the speeches, because if these graph is dance, then we can see that the the people are talking similarly to each other in the Hungarian Parliament.

189
00:42:09.160 --> 00:42:18.459
Zsófia Rakovics: And that's what I just aimed for. So I've just checked the transitivity of the graphs and that try to understand how homogeneous the speeches are.

190
00:42:18.880 --> 00:42:48.380
Zsófia Rakovics: and here is one output of it, and it's an internal. It shows the internal homogeneization of the language within the parties. So now we have, the party level as as the target on the horizontal access. We have the year, the different years of the targeted period. it's a 5 entire parliamentary terms, starting from 1,998 and ending at 19 2,218, and we have the transitivity

191
00:42:48.630 --> 00:43:16.660
Zsófia Rakovics: of so kind of the close. And it's kind of that. The homogeneousness of on the on the vertical access showing how homogeneous the speeches were together to to each other. And here we have 5 different political parties, to which for which we have a full time series. Msb. And Fidas. the F. This is the the currently ruling party. The governing party, led by Victor Orban and Msb. Used to be the the big.

192
00:43:16.850 --> 00:43:19.879
Zsófia Rakovics: Sorry. I like you to hear something. But I

193
00:43:20.900 --> 00:43:26.620
Zsófia Rakovics: okay, thank you. Yes. And Ms B. Is

194
00:43:26.750 --> 00:43:51.249
Zsófia Rakovics: is the the biggest opposition party, or used to be the biggest opposition party, and what we can observe is a slide that trending to the the higher transitivity values on this plot. That's what I just wanted to show you. So in general, we can see a a slight homogenization within the parties, and what is also possible as mentioned, these are graphs to construct the or and visualize the graphs. What we have the

195
00:43:51.270 --> 00:44:15.359
Zsófia Rakovics: based on the similarity of the speeches. I now would like to just show examples of it. I won't analyze those, but I would like to just show you this is a writing party called the Opique, and for 2 different time periods. How was the structure of the party? And you can see that there are 2 blocks of part party representatives in the left hand side of the picture and 3 different blocks of people in the right hand side.

196
00:44:15.360 --> 00:44:31.540
It's really exciting, because if you check another party, for example a Conservative Party, which is not yet in the Parliament. It's a MDF. It's a it's a Hungarian Democratic Forum, and and another party called the the Lamp, which is, which means that the

197
00:44:31.640 --> 00:44:54.309
Zsófia Rakovics: and the name of the party means that the politics can be different. We can see that it's a totally different structure of a political parties, and it's really exciting to see that they here we have individuals in different groups, and we don't have 2 or 3 blocks, but kind of each individual is a single, a single, a group of a party representative and party. group.

198
00:44:54.580 --> 00:45:23.269
Zsófia Rakovics: Yes, and here we have a and about another the last method I have applied about the classification of the parliamentary speeches, because I wanted to see whether, if I classify the speeches, the parliamentary speeches themselves. I will see more homogeneous tendencies, whether it's easier to to predict the party affiliation of the Speaker, and that's what I just the aim for I'm showing Cohen's Cup of values because it's

199
00:45:23.480 --> 00:45:42.960
Zsófia Rakovics: it's a try to adjust the imbalance of the number of speeches between the given parties and the other parties, and the in time it's not all at the same time, the same amount of speeches available. So for that, it's it's it's it's Co. Call Henska, which could be used. And again, the goal was to class it, to to see how.

200
00:45:42.960 --> 00:45:59.190
Zsófia Rakovics: how, to what extent can we classify the speakers correctly only by the transcript of the speeches themselves, and we would like to see how it's working and the here it is the results for this analysis. So here we have

201
00:45:59.240 --> 00:46:26.069
Zsófia Rakovics: 6 different political parties on that shown. And we can see, for example, the orange one is the the the ruling party feed that's led by the royal band, where we can see that the call has cup of values. So the correct classification of the model is really, almost monotonously increasing. So we can see here that there is a homogeneous tendency, a relative homogeneization across the parties. And it's exciting to

202
00:46:26.070 --> 00:46:33.310
Zsófia Rakovics: came to that conclusion. And here are the the last slide of my presentation about the the summary and the conclusions.

203
00:46:33.490 --> 00:46:58.139
Zsófia Rakovics: So it what this was a data-driven analysis based on document embedding, based on the document embedding vectors, vector spaces. I have constructed graphs of politicians and then, based on analyt analysis of these dynamic graphs. I try to understand the homogeneity and the whether I can see any polarizations within the speeches, and the conclusions are that there is a homogenization in the language used by the Parliamentary

204
00:46:58.140 --> 00:47:08.779
Zsófia Rakovics: party representatives, and we could see a language polarization observed based on the results and the references, and thank you for your attention.

205
00:47:14.290 --> 00:47:33.359
Thank you very much for your presentation. I I can see that all of you are using the word mover as embedding and and his colleague. So we have time for one question. is anybody interested in asking something? Yes, of course, Professor Robert.

206
00:47:37.700 --> 00:48:02.159
Sorry this was It's not really a it's more just a clarification. I I really lot enjoyed the presentation and the the figure where you're showing the dynamic networks of the political the the Mps based on their similarity of like speeches. Can you just clarify that? The first visualization you showed was that the same party over 2 years? And then it was like a fracturing. It went from 2 main groups to

207
00:48:02.160 --> 00:48:26.669
Zsófia Rakovics: actually 3. Is that is that the interpretation of that? Yes, on the first plot. It was the same party for 2 different parliamentary periods, and then there was a fractioning. Yes, and this part is really exciting, because it's really moving totally different directions as it's aim for in the kind of the beginning of this sort of the Parliamentary party life, it's it's really exciting to analyze that. And yes, so yes, you you understood correctly. That was the first time

208
00:48:26.670 --> 00:48:28.680
party and the first visual.

209
00:48:29.050 --> 00:48:41.180
And so when you saw that that made sense to you that that That result confirmed what you knew about that party. Yes, so yes, it was the literature which could be kind of proven

210
00:48:41.180 --> 00:49:04.440
Zsófia Rakovics: that, based on the similarity of the pitches. The constructed clusters show the the same hierarchies and structure what we know from the literature, and that's something we we will for sure, with the research group, analyze further and understand what kind of the the, the, the arch of this change, and how we can understand that the for for this writing party of the Hungarian Parliament.

211
00:49:04.660 --> 00:49:06.740
Zsófia Rakovics: Yes, thank you. Thanks.

212
00:49:06.930 --> 00:49:12.019
Thank you again. So the Anarchic. Maybe we'll see you later. I think so. Thank you.

213
00:49:12.460 --> 00:49:13.290
Zsófia Rakovics: Thank you.

214
00:49:16.070 --> 00:49:37.700
So moving on our next panelists are Rover Hembreak, and Jen's Dor being goes from the Federal Institute for vocational education and training in Germany, and the name of the their. Their their presentation is surveys and big data close in the gut, and they just post a question. How is it they got?

215
00:49:37.880 --> 00:49:39.960
Are they breathing or

216
00:49:41.200 --> 00:49:44.450
no means that if you make these music something?

217
00:49:46.490 --> 00:50:05.430
No, here, because if I'm not wrong, it was in person. It was about the version. Right? So it's a bit. But that will give us more time for questions for the next panelists, maybe. Sophia Rakovic, are you ready, then, to to do the presentation of your nerves.

218
00:50:05.470 --> 00:50:06.510
Hello!

219
00:50:06.560 --> 00:50:07.990
Zsófia Rakovics: yes, I am.

220
00:50:08.220 --> 00:50:09.429
Zsófia Rakovics: Then I can.

221
00:50:09.670 --> 00:50:34.970
Here we have again.

222
00:50:35.000 --> 00:50:43.639
and then come on. We have that, then time to ask you more questions, because I was asking my mind some question for you. Now. Thank you.

223
00:50:43.850 --> 00:50:45.630
Zsófia Rakovics: I will then share my screen.

224
00:50:47.530 --> 00:51:12.879
Zsófia Rakovics: I hope you can see this now, and if this is the correct one, I hope so. This only one thing you said to know. I I don't want to be a rude oh, no, no, for sure. I I'm not always seeing the chat. Yes, so please interrupt me. If it's time to. It's out, we can oppose it, because then the panelists cannot see. So I I'm so sure what I have to tell you through the microphone again. Yeah, yeah, sure, of course.

225
00:51:13.190 --> 00:51:24.490
Zsófia Rakovics: so, yeah, this is the second research. I would like to share you. I share it with you. And this is this language and political organization through 2 decades of parliamentary speeches.

226
00:51:24.490 --> 00:51:43.250
Zsófia Rakovics: And I would like to start again about the about the data and the corpus used. It's the same purpose. So here there will be the same information available for you. This is related to the the research of a research center for computational social science.

227
00:51:43.250 --> 00:52:07.979
Zsófia Rakovics: And then this corpus is again the parliamentary pitches between 1,998 and 220, and I have made some data cleaning and preprocessing methods and stuff, and try to compound significant words together and recognize named entities and remove stop-thirds before the analysis. The motivation of this research was to

228
00:52:08.000 --> 00:52:35.759
Zsófia Rakovics: detect the changes in the political strategy communication over the 2 decades. And then the goal was to examine the political polarization through the the the language, the transcript of the speeches, and I have used natural language processing to analyze, and these speeches and the transcript of the speeches. the methods applied was a very keyness analysis to identify the dominant

229
00:52:35.760 --> 00:53:00.270
Zsófia Rakovics: a words of for the political parties, and I have also used records similarities to try to understand and measure this at the different The similarities between the sets of keywords and to map patterns of the change from one set to another to to other parties. It's just really exciting to have this. It's a simple measure.

230
00:53:00.270 --> 00:53:13.789
It's really exciting and and easy to use, but still the results could be exciting the insights from this. And I've used classification model to predict the political party of the The membership of the the Parliament.

231
00:53:13.790 --> 00:53:37.680
Zsófia Rakovics: and for this presentation only I will focus on Fides Caddy Andpe, which is the governing party in Hungary since the 2,010. It's a party coalition. Fidas is a led by Victor Orban, and the Cadian pay is a Christian Democratic party of the People's Party, as they call themselves, and these are the articulation and governing Hungary since 2,010.

232
00:53:38.070 --> 00:53:47.419
Zsófia Rakovics: So I will focus for this presentation only on those because of time, and I try to be a cautious of time. Yes. So the

233
00:53:47.590 --> 00:54:08.230
Zsófia Rakovics: the first of all, about the keenness and the Jacquard similarity, and the and the in the methods use. So for with the skin as analysis we can examine which were the distinguished groups in the in the carcass, and it's a Chi-square-based statistical based method. It's used independence that's

234
00:54:08.230 --> 00:54:28.479
Zsófia Rakovics: tests on a to to see whether they are appearing or the frequency of those are higher than what? What we could expect from the Corpus, and in other groups of speeches and the parties in in this case, because we have political parties at the end. If we aggregate the speeches to

235
00:54:28.480 --> 00:54:54.729
Zsófia Rakovics: those, and we just see this by using this independent test. So we can have to perform this test separately for each and every word, and try to understand those and try to get the results. And based on that we will be able to define keywords. These are dominant words of each party, political party compared to others. So these are the words which are most characteristic for the given party.

236
00:54:54.730 --> 00:55:18.850
Zsófia Rakovics: and this checkout similarity, it measures the the similarity of 2 different sets. It's a calculate The just divides the number of values with the within the intersect of the 2 sets and device that with the with the union of the 2 sets, so you could see it here. I just form it in another way, there in the slides, so you can see that.

237
00:55:18.850 --> 00:55:39.640
Zsófia Rakovics: But we are dividing the number of observation in both sets by the number of observation in the union of the 2 sets. And that's the form of definition. And here we can see for fides the the rocket similarities over time. So we will be able to see how similar the set of the keywords are if you're

238
00:55:39.640 --> 00:55:50.900
Zsófia Rakovics: including the that time as a dimension into the analysis. So on the X, the horizontal access. We have the at the time we different here for the targeted period.

239
00:55:50.900 --> 00:56:15.459
Zsófia Rakovics: and on the vertical axis we have, there's record similarities. So the similarity measure, showing how close the keyvert sets are together and what we can observe, that is and we we we have the previous here as a reference. So we are comparing the set of the previous year to the actual year, and seeing what are the results, how different these sets are!

240
00:56:15.460 --> 00:56:36.380
Zsófia Rakovics: And we can see an increasing and monotonously increasing trend if you're checking the line. And of course, if you're checking the the the actual data points, we can see some trends and tendencies. And this is because we have here, red lines showing the the parliamentary elections. These lines are really

241
00:56:36.380 --> 00:57:00.599
Zsófia Rakovics: a crucial when we are analyzing parliamentary speeches, because, of course, the campaign has starts and the finishes before and after. So it's it's really having a period or just a having an effect on the speeches and the nature of the speeches. And that's why the red lines, the vertical lines are included to help us understand we, in which term we are within this 4 year period, what we have in Hungary

242
00:57:00.920 --> 00:57:21.759
Zsófia Rakovics: for 4 years of one parliamentary election. So it's just really exciting to have this tendencies and try to understand the overall results as well. And as we could see, it's a it's becoming more a similar to each other, more. How many homogeneous compared to the previous year than the actual set keyboard set of this of the parties of feed us here.

243
00:57:21.790 --> 00:57:47.940
Zsófia Rakovics: and yes, we have the dominant keywords. There are some examples I have a collected for you, and the first is for the first and and Parliamentary term, 1998, and 200 to for fides at the Cov. When this part is a governing party in Hungary, the dominant keywords for school education, youth, ministry of finance, investment, tourism, and transportation.

244
00:57:48.000 --> 00:58:11.720
Zsófia Rakovics: In the next 2 parliamentary terms, when they were in opposition The dominant words were such as a State Secretary, Minister for, and Su. Chan, who is a a a. A politician of Msb. And now it's a politician of Theika. So he had just formed a new party, and he used to be at that time the Prime Minister. Between this timeframe, between

245
00:58:11.740 --> 00:58:40.349
Zsófia Rakovics: 204 and 209, and another word was C, which was really dominant for feed us in this period. And then, I have just provided you with some examples for the next, next to 2 parliamentary terms, from 2,010 to 2,014, and from 2,014 and 2,018 we had for Fides Scadium pay, because then at that time they were already a party coalition governing the Hungary at the Hungarian Parliament.

246
00:58:40.350 --> 00:59:05.099
Zsófia Rakovics: and then that's why we have now the joined the keywords for these parties, and this was fundamental law of Hungary, which is the constitution of Hungary, which has been re-written by them in the the early years of 2,008 of the first period, 2,010 and 2,014. They have just a Re written this, and we have Ruler Development go down by now who used to be the Prime Minister

247
00:59:05.100 --> 00:59:19.699
between 2,009 and 2,010, we have Bill nationality fides and ruling parties. So these are the the keywords which, helping helping us understand the substantive content about the the political party communication.

248
00:59:20.650 --> 00:59:38.789
Zsófia Rakovics: And then I've just used classification models. I would like to show you. Now, first of all, I would like to talk about the the methods themselves. How I have prepared for this analysis. I've used the data, validation and test set and the the proportion of each and every set has been defined as A

249
00:59:38.790 --> 00:59:56.660
Zsófia Rakovics: as as written here. So 60% for training set 20 for validation and 24 test set, and I've used for the set for defining these sets. a randomly, a stratified sample for each year for each parliamentary term.

250
00:59:56.660 --> 01:00:16.360
Zsófia Rakovics: and that's what you could see on the slide as well, and the classification based on keyword, was also on another step the pro done and applied, and for that I have for each parliamentary term chosen the 300 most relevant keyness keywords, and then,

251
01:00:16.360 --> 01:00:43.559
Zsófia Rakovics: then use those as the the explanatory variables of the of them more classification model. So here the frequencies associated with these keywords were considered as as the explanatory variables in the in the actually boost model, it's a Gen. in in extreme, Gradient boosting method I have chosen. It's an ensemble method which is really flexibly. It could be flexibly used. And it's really

252
01:00:43.560 --> 01:00:50.800
Zsófia Rakovics: providing us with nice insights and results. So that's what I just decided to go with.

253
01:00:50.800 --> 01:01:19.859
Zsófia Rakovics: And the the binary output was what I have just decided to go for so for the first category for the of the classification. So the the output. I wanted to see whether the parties belonging to feed as Caddy, Mp. Or whether this speech belongs to feed as cadmium or or non-fi desktop and So all other groups, if the speech belong to to all other groups. Then, that was the 0 category for this output classification.

254
01:01:19.860 --> 01:01:48.160
Zsófia Rakovics: Yeah, thank you. And then we have the results here. So when we are considering time and trying to understand the performance of the classification models, we can see an increasing tendency in the Kappa values. Again, I've chosen to use Kappa values because these are correct for the imbalance of the frequencies of the the sets to groups. and that's what you could see here. So we have. if we are checking the clients, copper values

255
01:01:48.160 --> 01:02:13.160
Zsófia Rakovics: and monotonously increasing tendency. And that's kind of the main insight. I would like to share you now, and the summary and the conclusions. So this was again an Nlp base. That data driven analysis on the parliamentary speeches. The goal was to understand the characteristics of the language polarization, and try to find the association with political polarization studied by the the literature. There is a huge literature on that.

256
01:02:13.160 --> 01:02:22.230
So it's really exciting to try to understand whether the Imperial, whether the data could prove what we have seen in the literature already, and in other researches.

257
01:02:22.230 --> 01:02:32.280
Zsófia Rakovics: And then, yeah, I've chosen the evaluation, the performance metrics to understand the the, the the home organization or polarization of the speeches.

258
01:02:32.280 --> 01:02:56.540
Zsófia Rakovics: And that's a conclusion. We can say that there are patterns of of change in the political parties and the we could map that for both governing and opposition parties. In this presentation I have shown only the governing party, the currently governing party, but of course I have done this, and on this further and for the opposition parties. There was a observations as well for for these tendencies and for these patterns of change.

259
01:02:56.660 --> 01:03:07.240
Zsófia Rakovics: and we can see that there is a language polarization observed for this, fides Scadium Bay party and party coalition of the Hungarian Parliament.

260
01:03:07.430 --> 01:03:09.810
Zsófia Rakovics: and thank you for your attention. Then.

261
01:03:10.420 --> 01:03:11.480
in 2 min

262
01:03:13.400 --> 01:03:40.790
presentation, maybe there is any question here for the audience and for Sophia, or in case test, nobody wants to ask anything. I have just a question, or is more than a question is like our proposal. I don't know whether you know the the measurement call Jpj for measuring polarization, which is based on facilitation.

263
01:03:41.860 --> 01:03:43.000
Zsófia Rakovics: yeah.

264
01:03:43.440 --> 01:03:46.700
a few of her, because we we have.

265
01:03:47.770 --> 01:04:01.929
Zsófia Rakovics: sorry. I think it was just a cut this. I. I couldn't hear exactly what you have heard. But yeah, the method. I haven't tried that yet, but it's I have heard about it, and I will experiment with that. Thank you for for advising and recommending it.

266
01:04:02.090 --> 01:04:19.599
Yeah, it is a very good measurement. And we have here the expert maybe you can. I'm gonna just gonna publish in the chat one of the the piece references, and then we can share it with the whole audience. Thank you very much of the congratulations.

267
01:04:24.390 --> 01:04:52.370
So then moving on with our last presentation today.

268
01:04:52.370 --> 01:05:03.650
examining the different political sites, memorialization of using tools of natural language processes and narrative psychology. Are we ready?

269
01:05:04.820 --> 01:05:08.710
Ildikó Barna: Yes, I think I'm I'm ready. Can you see my slides?

270
01:05:09.290 --> 01:05:17.470
Yes, we can see it, but pretty well welcome in digital Garner. You are very welcome when I, when you are ready you can just start.

271
01:05:18.420 --> 01:05:31.090
Ildikó Barna: First of all, thank you very much for for having us here. as you said, the the paper is co-authored. And fortunately, one of my co-authors ran out and name at this here with us.

272
01:05:31.240 --> 01:05:38.030
Ildikó Barna: so she's also participating here for

273
01:05:39.190 --> 01:05:46.549
Ildikó Barna: thinking about keeping the time, so I I down my presentation to to be able to keep it more strictly.

274
01:05:47.300 --> 01:05:58.759
Ildikó Barna:  the Treaty of Chernon, signed in the 1,920 formally ended World War I. Between Hungary and the Allied powers. The treaty resulted in

275
01:06:00.840 --> 01:06:08.529
Ildikó Barna: the treaty resulted in Hungary losing 72% of his territory and almost two-thirds of its population.

276
01:06:08.890 --> 01:06:32.309
Ildikó Barna: The 303 is among the most important events in the twentieth centuries Hungarian history. Its significance goes beyond history. It's impact is outstanding from a political, social, and societal psychological point of view. And it's influence continues to this day, as Eva Covach argues, Sri LAN has a hegemonic position in Hungary, and social memory

277
01:06:32.850 --> 01:06:49.289
Ildikó Barna: in the collective representation of Hungarian history, that 363 plays a central role as an insult suffered by external groups. Earlier studies have shown that this representation led to a specific collective victim role in the construction of Hungarian national identity.

278
01:06:49.810 --> 01:07:04.479
the constant instrumentalization of history, memory, and identity. Politics have been important building blocks of the Orban regimes. and not only from 2,010, but also between 1998 and 2,002, when they also from the Government

279
01:07:05.010 --> 01:07:21.250
Ildikó Barna: after 2,010, and has been at the center of the Orban regime's memory. Politics resulting in a so-called 3 and uncult one of the Orban governments, first measure in 2,010 was that to grant citizenship to Hungary and Minorities abroad.

280
01:07:21.490 --> 01:07:30.919
Ildikó Barna: The measure was presented as the nation's unification. However, the votes of those New Hungarian citizens were also crucial for fides.

281
01:07:31.090 --> 01:07:42.449
Ildikó Barna: Soon after that the Government declared the fourth of June, when the treaty was signed, the day of national belonging further instrumentalizing history, and putting historical debates at the service of politics

282
01:07:42.700 --> 01:07:51.939
Ildikó Barna: in 2,020 was the one hundredth anniversary of the treaty. The Parliament has declared this year, year the year of national cohesion.

283
01:07:53.130 --> 01:08:03.570
Ildikó Barna: Several programs have been organized memorials have been erected in various parts of the country, and speeches have been given by the Prime Minister and other politicians.

284
01:08:03.680 --> 01:08:12.329
Ildikó Barna: The official narratives was completed by the Declaration on the Defense on national identity adopted by the Parliament on the day of the treaty was signed.

285
01:08:12.810 --> 01:08:26.450
Ildikó Barna: In addition to the weekly narrative of the past, the urban regime lived as efforts to create a positive historical identity became apparent. That is, a nation that overcomes difficulties and survives.

286
01:08:27.500 --> 01:08:45.299
Ildikó Barna: Our research aimed to explore the thematic structure of the 3. And on this course and the narrative psychological indicators associated with these identified topics in the call with quarant political public sphere representative represented in online media.

287
01:08:45.500 --> 01:09:10.680
Ildikó Barna: Our main research questions are the following, which are the main topic points of media representation of what are the narrative features of these 3, and on teams, and what are the similarities and differences of the discourse on 3, and on this guide, by different topics and narrative features in the media of different political positions, namely, proper garment pro government, far right and left liberal

288
01:09:10.690 --> 01:09:31.069
Ildikó Barna: our May hypotheses is that the discourse of has distinct topics which also differ in terms of narrative indicators. Further, more, we hypothize that the either ideological position of the medium we at the given article has been published has an impact on both in the Kematic indicators and also its narrative characteristics.

289
01:09:31.660 --> 01:09:40.080
Ildikó Barna: Since our research. We do conclusions about the narrative of the different political sites based on the political orientations of the new size.

290
01:09:40.160 --> 01:09:48.879
Ildikó Barna: We have to ask the question, Does the media out of less of different political position, represent the memory of different political sites.

291
01:09:49.460 --> 01:09:55.009
Ildikó Barna: To answer this question we need to briefly talk about the media landscape of Hungary.

292
01:09:55.280 --> 01:10:07.360
Ildikó Barna: after 2,010, the over Orban regime fundamentally reshaped the countries media landscape, the government has been systematically building a centralized media network.

293
01:10:07.400 --> 01:10:14.930
Ildikó Barna: The main components of these are the the following, the total control over the public service media and the State press agency.

294
01:10:14.960 --> 01:10:43.070
Ildikó Barna: unprecedented level of over and over ownership concentration in the private media sector and also a near monopoly in the daily printed radio sectors. At the same time they also position uses the uses, the remaining independent and critical sources, and the far right has a few of its own outlets. Therefore we are you that we can grasp the different size of of memory, size, memory, politics, by analyzing these sources.

295
01:10:43.930 --> 01:11:09.109
Ildikó Barna: And now about the data set in the corpus which is collected by Santi one and the base, the social listening company, and then text analytics platform, and contained all Hungarian articles. But it'll be published between September 2,017, and the September 2,020 containing the word tri-non and all it's conjugate conducted form.

296
01:11:09.190 --> 01:11:28.410
Ildikó Barna: We selected those new sizes. Political orientation should be could be clearly identified, and we dealing the duplicates and the Semit applicants in quasi duplicates. So the full quote for side was what sizable was around 11,000 articles from 36 different sides.

297
01:11:28.500 --> 01:11:35.400
Ildikó Barna: So it's a simple keyverse search was used. Several of the articles do not focus on, but just mentions it.

298
01:11:35.410 --> 01:11:54.909
Ildikó Barna: So we use topic modeling to solve this problem, which also separates the the articles according to to this. And in this way the subcurpose. That is, act. The actual object of the research could be identified as a set of a few topics

299
01:11:55.420 --> 01:12:01.990
Ildikó Barna: So the final court was contained around 303,500 articles.

300
01:12:03.620 --> 01:12:25.470
Ildikó Barna: In our research we utilize 2 methods Lda and the narrative category called content. And this is to our best knowledge. This is the first time it was combined in a research to to identify the heater late and thematic structure of the given corpus. We use the L, they a topic modeling. And

301
01:12:25.700 --> 01:12:46.179
Ildikó Barna: as you probably all know, all the processes. Texas, a bag of words, namely, 3 texts as a set of isolated words, ignoring word order, and systematic relation after the necessary repress of processing steps. it was very similar. What zoom is. It was talked about

302
01:12:46.180 --> 01:12:58.390
Ildikó Barna: 24 topics were extracted, based on coherence, measure, and interpretability, and 7 out of the 24 ever connected directly to memory politics.

303
01:12:58.800 --> 01:13:25.790
Ildikó Barna:  Our other method, the narrative, categorical content analysis, or, in short, narcot, is a tool to perform automated content analysis of narratives, narrators unwittingly place codes in the written and or spoken text, which can be decoded to do on conclusions. And narcot is based on narrative psychology, and quantifies the correct frequency of certain forms in text.

304
01:13:25.860 --> 01:13:31.469
Ildikó Barna: It is a specific feature of the narcot that it makes the analysis through basic dimensions of

305
01:13:31.620 --> 01:13:55.410
Ildikó Barna: The basic dimensions are agency, psychological perspective, evaluation of characters. He was of spatial temporal perspective, negation, and social reference. It is more complex than a back of Morse model order of the verse is important. It also examines syntax, syntactic, relationships and endings play an important role in the in in identifying them.

306
01:13:56.670 --> 01:14:12.540
Ildikó Barna: Unfortunately, I don't have the time to cover all several 7 topics. So I talk about 3 of them. These 3 are closely connected to the previously mentioned memory, politic features connected to 3, and on responsibility together, Ness and Heroism.

307
01:14:13.870 --> 01:14:26.170
Ildikó Barna: To pick 2 shows the various quality, because size, different interpretation of the historical context of 3 unknown. The discourse on the political on all political side, is about responsibility, but with different emphasis.

308
01:14:26.180 --> 01:14:41.580
Ildikó Barna: The articles in the pro Government press are mainly about the Hungarian Soviet Republic and the rep responsibility of the Communists and the left. The narrative. They are because in the far right press is similar, but often means mixed with anti-semitism.

309
01:14:41.780 --> 01:14:57.020
Ildikó Barna: In addition, these articles, those are core close, close power with contemporary Hungarian politics, and described, left being Liberal politicians as quote, disgusted by Theian, or as not caring about the country state greatest tragedy

310
01:14:57.190 --> 01:15:21.529
Ildikó Barna: art, because in the left liberal class are also about responsibility, but dispute that only the left was responsible for the articles clearly show the different perceptions of historical figures on different sides of the political spectrum. The most tracking in Rt. Feature is a negative evaluation. It is the left liberal side which uses negative evolution and negation more frequently, and activity less.

311
01:15:21.550 --> 01:15:26.129
Ildikó Barna: This indicates that he tries to minimize this. All the responsibility of that.

312
01:15:29.820 --> 01:15:43.090
Ildikó Barna: Yes. Most of the articles in Topic 10 published in the pro government and far right media are about the emotionally charged speeches of politicians, especially if to Orban about 300.

313
01:15:43.090 --> 01:16:06.350
Ildikó Barna: In these speeches the Hungarian people are presented as heroes. A nation that a quote has a vet or storms of history, we'll be stuck but by bullets or so we will not be stuck by bullets and or swords, even if not knocked off its feet it will rise again and again, and the 100 years after the defeat in the Ha in the World War and the 3 on dictate

314
01:16:06.350 --> 01:16:11.750
Ildikó Barna: today on the stage of European history as champion of survivor. End. Quote

315
01:16:11.870 --> 01:16:23.570
Ildikó Barna: the articles in the left Liberal press in this topic also deal with this species. But, unlike the other 2 political sites, they do not just publish them, but provide critical commentary.

316
01:16:23.650 --> 01:16:40.650
Ildikó Barna: The relevance of the topics. Interpretation is highlighted by the frequent use of us. Reference, however, it is also apparent that the nation heroization is controversial as positive balance, negative emotion, negation, and meta narrative perspective are also frequently used.

317
01:16:40.850 --> 01:16:53.430
Ildikó Barna: there is also an interaction with political ideology. It is the right ideology that sees the heroic component as most relevantly highlighted by the most frequent use of us. Reference

318
01:16:53.570 --> 01:17:16.820
Ildikó Barna: the articles in topics 14. And this is my last topic to show published in the pro-government and far right press focus much on today's response to the the central is belonging, and togetherness with Hungarians beyond the borders quote. It was the belief in belonging that kept the Hungarians who were torn from their homeland alive, and quote

319
01:17:16.880 --> 01:17:37.029
Ildikó Barna: in this context many articles deal with the decision of the Hungarian Parliament in 2,010 that declared the force of June, the day of national belonging the law of On dual, that citizenship is depicted as a realistic response to 3 on the term motherland is also typically used in this topic.

320
01:17:37.150 --> 01:17:47.929
Ildikó Barna: At the same time, the left being press deals with these themes with much less emotional charge, and many times criticize the right thing. Policy officially revisionize them.

321
01:17:47.960 --> 01:18:03.410
Ildikó Barna: The narrative analysis of the topic provides a direct confirmation of the topic, modeling in as much as This topic is predicted by the frequent use of Us. Reference. It seems to be a universal features in since there are no interaction with political ideologies.

322
01:18:03.530 --> 01:18:12.879
Ildikó Barna: And to conclude, we found that the dig scores on 300 centers around this thing topics that also differ in terms of narrative indicators.

323
01:18:12.880 --> 01:18:37.499
Ildikó Barna: So it is really, exciting and very fruitful to combine these 2 methods. Most of the narrative indicators have discriminated power across the topics. This discourse it framing the first by either ideological position, however, that of the pro government and the farm media very much overlaps most of the narrative characteristics are significant to different by ideological

324
01:18:37.500 --> 01:18:46.789
Ildikó Barna: The position. however, the similarity of the pro government and the far, right near media is also also visible here. Thank you very much.

325
01:18:51.370 --> 01:18:57.760
Thank you. The kobama. Is there any question for her in among the audience?

326
01:18:58.720 --> 01:19:12.169
No, because I just want to ask you something. I was thinking up. We know that it's very difficult, a very difficult topic, because you must indicate a number of top, because before hand

327
01:19:12.410 --> 01:19:26.169
to the algorithm. So have you calculated any measure of coherence. And have you had an expert who studies the solution with more coherence to see which one is optimal?

328
01:19:27.610 --> 01:19:44.990
Ildikó Barna: actually, what we do in our pipeline most of the time is that we choose a a topic number from like 10 to 30. And we run the model 10 times for each number of topics.

329
01:19:44.990 --> 01:20:09.919
Ildikó Barna: and then we have a a mattress with all the coherence measures, and also the the keyboard of the different topics, and we choose the the actual, the final topic number by examining this matters of coherence measures, and also we we also. look at the dispersion of

330
01:20:09.920 --> 01:20:17.110
Ildikó Barna: of the coherence measures and also taking into account the interpret credibility of the topics.

331
01:20:18.330 --> 01:20:24.660
Thank you very much for your answer. It was very clear. So now we have another question for you.

332
01:20:24.890 --> 01:20:35.529
Please go on. Yes, yes, I have a question concerning the coherence numbers. I think the process that you describe makes a lot of sense to you to calculate them, and then look at the individual

333
01:20:35.530 --> 01:21:00.530
models that really out from them. Which coherence measure do you find most helpful? Because when I experimented with atomic models, I found that they were sometimes, if you look at, for example, Cv or you get to really different results for the same topic model and then choosing, you know. Well, you have to look into the models and make sense of what you see in my, that's at least my experience so far.

334
01:21:00.530 --> 01:21:07.350
Do you have any suggestions or tips on how to choose which which models to look at.

335
01:21:08.310 --> 01:21:24.380
Ildikó Barna:  a. As far as I know, I I actually, I'm not the expert on the on the on the computational side. So maybe I'm not. I I'm not able to answer your question clearly, and I'm sorry for that.

336
01:21:24.480 --> 01:21:38.009
Ildikó Barna: the Aldi the the the the give the the coherence, measure what we what we use, and and but I have to look at it

337
01:21:38.060 --> 01:21:42.369
Ildikó Barna: more closely and and sorry for not being able to answer your question.

338
01:21:44.810 --> 01:22:00.570
Ildikó Barna: Thank you very much. Anyway. I just just to let you know that. The question was asked by a step. And now from the the University of Germany.

339
01:22:00.730 --> 01:22:02.250
Ildikó Barna: and

340
01:22:02.260 --> 01:22:06.209
Ildikó Barna: get back to you if if I can, if

341
01:22:06.390 --> 01:22:31.650
yeah, I I hope I can. But I I'll just let you know that the Ss. And node was one of our our percenters in our session yesterday, a bit at the one. So you can access. Okay, okay, I I I will. I will contact him. That is the amazing things of these events, and they so the next question is going to be asked. Good by, Professor Gavara from the

342
01:22:31.650 --> 01:22:42.020
the the University of Madrid. Go on. Hi, thank you very much for your presentation regarding the last question of the other professor.

343
01:22:42.140 --> 01:22:49.679
If you know which language program we are using in order to compute the Latin-directed algorithm.

344
01:22:49.830 --> 01:22:52.030
Do you know that because it

345
01:22:52.050 --> 01:23:01.410
according to the problem and the function that you are using? Probably I could respond to the to the coordinates measure.

346
01:23:01.490 --> 01:23:04.349
So I don't know the language program

347
01:23:04.780 --> 01:23:08.659
that you were using in order to compute the Lda algorithm.

348
01:23:09.770 --> 01:23:14.860
Ildikó Barna:  I think I'm not sure of that.

349
01:23:15.440 --> 01:23:23.259
Ildikó Barna: I'm sorry for that. Okay, for example, probably it could be you. You mean that. we are using python.

350
01:23:23.810 --> 01:23:51.800
Okay with python. Okay, I don't know. Probably one of the most use experience measures are those who who calculate the currents according to the similarity between words. And I, I think we talk about that yesterday, in fact, because that's a tricky question, indeed, because which is the similarity between virus and hospital, for example, it may be launched to to the same topic

351
01:23:51.800 --> 01:24:12.690
mit ctl, and but they are not similar in in, in letters in it. So that's because it's highly recommended to to take a look at the difference resort given by by the topics, and not always a trust in the highest correlation, but taking a deeper look into the the different results and 250.

352
01:24:12.800 --> 01:24:18.760
Ildikó Barna: And to decide who makes more sense. Yeah, what I really find.

353
01:24:18.760 --> 01:24:44.379
Ildikó Barna: just just to add, what I really find very useful is the interpretation of the of the topic. So it's it really helps to. In most of the time it has greatly to decide how many topics is import is is is a good, I mean good solution. You know it depends on you know you all. You never know exactly what is the good solution

354
01:24:44.390 --> 01:25:06.740
Ildikó Barna: you just get a have a kind of have it a a a good guess of it. And I think it is also very important to interpret, because many times it's very technically used and just looking at the numbers. And and I think it's not enough to to be able to use as well.

355
01:25:07.370 --> 01:25:29.110
Yeah, absolutely in this, because sometimes we we find that there are lots and lots of different topics that it probably made at the best solution. But they are not, and even we can find different words in different topics. The same word in different topics and everything. So that that was a good practice. Thank you very much.

356
01:25:29.940 --> 01:25:35.089
Thank you. Nikki Obama. Is there any other question?

357
01:25:37.200 --> 01:25:53.110
And just to conclude this session. Thank you very much for all the attendance for coming, and today I know that it was one of the last of today, and you have been yes, and maybe going to another presentation.

358
01:25:53.270 --> 01:26:12.489
Some thoughts that came to my mind, as you may all know, Ellen must have left us like blind, completely blind. I mean researchers. in terms of analyzing theater. So maybe it's time just to rethink and join versus between researchers and universities and institutions

359
01:26:12.490 --> 01:26:39.140
and then to to join us. To to get access to that platform again and to unload data. Maybe there are some 2 research, many questions. Of course, one is a economics, and the price we we would have to pay for to access that data. But another of those reasons could be some, but, praxis, that those platforms are doing nowadays.

360
01:26:39.170 --> 01:26:54.720
We sometimes may involve a kind of threaten for democracy. So just let think about that, all of us and be in new ways to get new ways to access that data again. Thank you for coming.