-
Notifications
You must be signed in to change notification settings - Fork 0
/
31_match_geopolitics_digital.R
293 lines (233 loc) · 9.72 KB
/
31_match_geopolitics_digital.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
## 31_match_geopolitics_digital
# Input: all_texts.rds, zs_subtopics.csv, CountryMentions_NM-dict_ParagraphLevel.rds, all_meta.rds
# load packages:
library(tidyverse)
library(magrittr)
# data merging: ####
# adaptable Classification Cutoff:
zs_max_cutoff = 0.7
# update data:
outfile <- "zs_subtopics.csv"
# googledrive::drive_download(paste0("TRIAS/", outfile), paste0("data/", outfile), overwrite = T)
file.copy(from = paste0("M:/user/schroeder/", outfile), to = paste0("./data/", outfile), overwrite = T)
# lookup EU membership status:
EU_at_time <-
tibble::tibble(
memberstate = c("AT", "BE", "BG", "HR", "CY", "CZ", "DK", "EE", "FI", "FR", "DE", "GR", "HU",
"IE", "IT", "LV", "LT", "LU", "MT", "NL", "PL", "PT", "RO", "SK", "SI", "ES", "SE", "GB"),
eu_accession = as.Date(c("1995-01-01", "1958-01-01", "2007-01-01", "2013-07-01",
"2004-05-01", "2004-05-01", "1973-01-01", "2004-05-01",
"1995-01-01", "1958-01-01", "1958-01-01", "1981-01-01",
"2004-05-01", "1973-01-01", "1958-01-01", "2004-05-01",
"2004-05-01", "1958-01-01", "2004-05-01", "1958-01-01",
"2004-05-01", "1986-01-01", "2007-01-01", "2004-05-01",
"2004-05-01", "1986-01-01", "1995-01-01", "1973-01-01"))
)
analysis_data <-
# Classification is the same for duplicate texts:
read_rds("data/all_texts.rds") %>%
left_join(.,
read_csv(paste0("data/", outfile)) %>%
select(text_para = text, target_max:digital_policy),
join_by(text_para)
) %>%
filter(!is.na(digital_policy)) %>%
# calculate combined classification:
mutate(zs_max = pmax(digital_communications,
internet_technologies,
digital_services,
digital_algorithms,
digitized_data,
digital_policy,
na.rm = T),
zs_digital = ifelse(zs_max > zs_max_cutoff, 1, 0)
) %>%
# add Country Mentions:
left_join(.,
read_rds("large_data/CountryMentions_NM-dict_ParagraphLevel.rds") %>%
select(-c(doc_pos, doc_key)) ,
join_by(id)
) %>%
# add Date:
left_join(.,
read_rds("data/all_meta.rds") %>%
select(doc_key, date),
join_by(doc_key)
)
# aggregate Country Mentions for EU (time-sensitive):
EU_mentions <- analysis_data %>%
pivot_longer(all_of(
countrycode::codelist %>% filter(eu28 == "EU") %>% pull(iso2c)),
names_to = "memberstate",
values_to = "cms"
) %>%
right_join(.,
EU_at_time,
join_by(memberstate)
) %>%
# was in EU at date?
mutate(eu = eu_accession <= date) %>%
mutate(eu = ifelse(memberstate == "GB" & date >= as.Date("2020-02-01"), F, eu)) %>%
select(-c(BI:WS)) %>%
group_by(eu, id) %>%
# count for EU members at date:
summarise(cm_EU = sum(cms)) %>%
ungroup() %>%
filter(eu) %>%
select(-eu)
# add EU mentions:
analysis_data %<>%
left_join(.,
EU_mentions,
join_by(id)
) %>%
# calculate country group mentions:
mutate(
cm_total = rowSums(analysis_data %>% select(BI:WS), na.rm = T),
cm_external = cm_total - cm_EU,
cm_CN_wider = rowSums(analysis_data %>% select(CN, TW, HK, MO), na.rm = T),
cm_BRICS = rowSums(analysis_data %>% select(BR, RU, IN, CN, ZA), na.rm = T),
cm_BRICSplus = rowSums(analysis_data %>% select(IR, AE, ET, EG), na.rm = T) + cm_BRICS
) %>%
# unselect other country mentions:
select(-c(BI:WS), US, CN, RU, ZA, IN, BR)
# aggregate on doclevel: ####
analysis_data_doclevel <-
analysis_data %>%
group_by(doc_key) %>%
summarise(
n_paras_classified = n(),
# aggregate classifications:
any_digital_para = max(zs_digital, na.rm = T),
n_digital_para = sum(zs_digital, na.rm = T),
share_digital_para = n_digital_para / n_paras_classified,
# aggregate classification scores:
max_digital_score = max(zs_max, na.rm = T),
mean_digital_score = mean(zs_max, na.rm = T),
# aggregate subscores:
across(digital_communications:digital_policy, max, .names = "{col}_max"),
across(digital_communications:digital_policy, mean, .names = "{col}_mean"),
# aggregate manual codings:
handcoding_max = max(target_max, na.rm = T) %>% ifelse(is.infinite(.), 0, .),
# aggregate Country mentions:
across(cm_EU:BR, sum), # Count CMs
across(cm_EU:BR, ~ ifelse(.x > 0, 1, 0), .names = "{col}_any"), # any CMs
across(cm_EU:BR, ~ ifelse(cm_total > 0, .x / cm_total, 0), .names = "{col}_share") # share of all CMs in doc
) %>%
# add metadata:
left_join(.,
read_rds("data/all_meta.rds"),
join_by(doc_key)) %>%
ungroup() %>%
select(-cm_total_share)
# save data: ####
write_rds(analysis_data, "./data/ParaLevelData_zs.rds")
write_rds(analysis_data_doclevel, "./data/DocLevelData_zs.rds")
### Validation (Doclevel) ####
cor(analysis_data_doclevel$share_digital_para, analysis_data_doclevel$cm_CN_wider_share)
# define cutoff function:
calc_cutoff_candidates <- function(cutoffvar, targetvar, seqstart = 0, seqend = 1, seqstep = .01, facetingvar = "cutoff") {
### targetvar: boolean!
### facetingvar needs to be present in long format data!
library(magrittr)
inputdata <- tibble(
targetvar = targetvar,
cutoffvar = cutoffvar,
facetingvar = facetingvar
)
df <- inputdata
cutoff_candidates <- tibble()
facets <- unique(facetingvar)
for (facet in facets) {
df <- filter(inputdata, facetingvar == facet)
for (cutoff in seq(seqstart, seqend, seqstep)) {
df %<>% mutate(pred = cutoffvar > cutoff)
TP <- sum(df$pred & df$targetvar)
TN <- sum(!df$pred & !df$targetvar)
FP <- sum(df$pred & !df$targetvar)
FN <- sum(!df$pred & df$targetvar)
precision <- TP / (TP + FP)
recall <- TP / (TP + FN)
f1 <- 2*(precision*recall)/(precision+recall)
accuracy <- (TP + TN) / (TP + TN + FP + FN)
baccuracy <- ((TP / (TP + FN)) + (TN / (TN + FP))) / 2 # Balanced accuracy (because imbalanced sample!) - https://neptune.ai/blog/balanced-accuracy
cutoff_candidates %<>%
bind_rows(., tibble_row(cutoff, precision, recall, f1, accuracy, baccuracy, TP, FP, TN, FN, facet))
}
}
return(cutoff_candidates)
}
# define plotting function:
plot_metrics <- function(data, scorename, plotname, facetingvar = NULL){
library(magrittr)
data %<>% mutate(facet = facetingvar)
# # Accuracy
# pl.accuracy <-
# ggplot(cutoffs, aes(x=cutoff, y = accuracy, group = facet), )+
# geom_point(color = "darkgreen")+
# geom_line(color = "darkgreen")+
# labs(title = "Accuracy",
# subtitle = "What overall share of texts is correctly classified?",
# y = "Accuracy\n",
# x = "Chosen semantic similarity cutoff")+
# theme_bw()
# Balanced Accuracy
pl.baccuracy <-
ggplot(data, aes(x=cutoff, y = baccuracy, group = facet), )+
geom_point(color = "darkgreen")+
geom_line(color = "darkgreen")+
labs(title = "<b>Balanced Accuracy</b>: What overall share of documents in each category is correctly classified?",
y = "Accuracy\n",
x = "")+
theme_bw()+
theme(plot.title = element_markdown())
# Precision and recall
pr <- data %>%
select(cutoff, precision, recall, facet) %>%
pivot_longer(cols = 2:3, names_to = "metric", values_to = "value")
pl.precrec <-
ggplot(pr, aes(x=cutoff, y = value, color = metric))+
geom_point()+
geom_line()+
scale_color_manual(values = c("#d95f02", "#7570b3"))+
labs(title = "<b>Precision & recall</b>",
subtitle = "<span style = 'color:#d95f02;'><b>Precision</b></span>: What share of retrieved documents is actually relevant?<br><span style = 'color:#7570b3;'><b>Recall<b></span>: What share of actually relevant documents is is retrieved?",
y = "Value\n",
x = "",
color = "")+
theme_bw()+
theme(legend.position = "none",
plot.subtitle = element_markdown(),
plot.title = element_markdown())
# F1 Score
pl.f1 <-
ggplot(data, aes(x=cutoff, y = f1, group = facet), )+
geom_point(color = "black")+
geom_line(color = "black")+
labs(title = "<b>F1 Score</b>: Harmonized mean of recall & precision",
y = "F1 score\n",
x = paste("\nClassification cutoff applied to", scorename)) +
theme_bw()+
theme(plot.title = element_markdown())
if (!is.null(facetingvar)) {
pl.f1 <- pl.f1 + facet_grid(. ~ facet)
pl.precrec <- pl.precrec + facet_grid(. ~ facet)
pl.baccuracy <- pl.baccuracy + facet_grid(. ~ facet)
# pl.accuracy <- pl.accuracy + facet_grid(. ~ facet)
}
# Combined plot
pl.comb <- # pl.accuracy /
pl.baccuracy / pl.precrec / pl.f1
ggsave(paste0("./output/plots/", plotname, ".png"),
pl.comb, width = 24, height = 27, units = "cm")
return(pl.comb)
}
doc_subtopics <- analysis_data_doclevel %>%
pivot_longer(c(digital_communications_max, internet_technologies_max, digital_services_max, digital_algorithms_max, digitized_data_max, max_digital_score),
names_to = "label",
values_to = "score")
cutoffs_doc <- calc_cutoff_candidates(cutoffvar = doc_subtopics$score, targetvar = doc_subtopics$handcoding_max, facetingvar = doc_subtopics$label)
plot_metrics(data = cutoffs_doc,
scorename = "Zeroshot Score with finegrained labels (document level)",
plotname = "Digital_classification_cutoff_doc",
facetingvar = cutoffs_doc$facet)