Quaglia_The Emotional Impact of Drone Warfare on the Ground.Rmd

---
title: 'The Emotional Impact of Drone Warfare on the Ground '
author: "Sofia Quaglia"
date: '2022-08-19'
output:
  pdf_document: default
  html_document: default
  word_document: default
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```



```{r, echo = FALSE}

# Loading libraries 

library(dplyr)
library(ggplot2)
library(knitr)
library(quanteda)
library(quanteda.textmodels)
library(quanteda.textstats)
library(quanteda.textplots)
library(rvest)
library(readr)
library(readxl)
library(readtext)
library(rio)
library(stopwords)
library(tidyverse)
```

_DATASET_

```{r, echo = FALSE}
# Load saved dataset and overview of dataset

dat_tweets <- read.csv("C:/Users/Sofias PC/OneDrive - University College Dublin/MSc Thesis Data Science and Politics/snscrape_tweets_june_kids_2009.csv")
names(dat_tweets)


#myfunc <- function(x,y){dat_tweets[dat_tweets$Datetime >= x & dat_tweets$Datetime <= y,]}

#DATE1 <- as.Date("2009-09-23")
#DATE2 <- as.Date("2010-09-30")

#Test_H2 <- myfunc(DATE1,DATE2)

# getting specific dates

#myfunc <- function(x,y){dat_tweets[dat_tweets$Datetime >= x & dat_tweets$Datetime <= y,]}

#DATE1 <- as.Date("2009-10-30")
#DATE2 <- as.Date("2009-10-30")

#dat_tweets <- myfunc(DATE1,DATE2)


# Transforming the dataset into a quanteda corpus
corp_tweets <- corpus(dat_tweets, text_field = "Text")

# Transforming the subset into a quanteda corpus 

#corp_subset <- corpus(Test_H2, text_field = "Text")

#corp_subset <- cbind(corp_subset, Date = Test_H2$Datetime)
```

_LEXICON_ 

```{r, echo = FALSE}
# importing the NRC lexicon

dat_dic <- rio::import("C:/Users/Sofias PC/OneDrive - University College Dublin/MSc Thesis Data Science and Politics/NRC-Emotion-Lexicon-v0.92-In105Languages-Nov2017Translations.xlsx")

# select only English and the relevant categories in a new data frame

dat_eng_dic <- dat_dic %>% 
  select(starts_with("English"), Positive:Trust)

names(dat_eng_dic)

head(dat_eng_dic)

nrow(dat_eng_dic)


# now transform dictionary to "long" format

dat_eng_dic_long <- dat_eng_dic %>% 
  rename(word = 'English (en)...1') %>% 
  gather(sentiment, score, -word)

head(dat_eng_dic_long)

# note: only words with a score of 1 belong to the respective category

# therefore, I only "filter" terms with scores of 1

dat_eng_dic_scored <- filter(dat_eng_dic_long, 
                            score == 1)

# you can check whether these scores make sense
head(dat_eng_dic_scored)

# get number of scored terms
nrow(dat_eng_dic_scored)

# remove "NO TRANSLATION" and get each term only once per category

dat_eng_dic_scored <- filter(dat_eng_dic_scored, word != "NO TRANSLATION") %>% 
  unique()

nrow(dat_eng_dic_scored)

dict_english <- as.dictionary(dat_eng_dic_scored)

```

_SENTIMENT ANALYSIS on the entire dataset _

```{r, echo = FALSE}

# tokenise and apply dictionary


dat_with_dict_tweets <- corp_tweets %>% 
  tokens() %>% 
  tokens_remove(stopwords(language = "en", source = 'marimo')) %>% 
  tokens(remove_punct = TRUE) %>% 
  tokens_lookup(dictionary = dict_english, nested_scope = "dictionary") %>% 
  dfm()

FINAL_dat_with_dic_tweets <- quanteda::convert(dat_with_dict_tweets, to = "data.frame")

View(FINAL_dat_with_dic_tweets)

summary(FINAL_dat_with_dic_tweets)

# Most frequent positive words
corp_tweets %>%
  tokens() %>% 
  tokens_keep(pattern = dict_english$Positive) %>%
  dfm() %>% 
  topfeatures(n = 30)

# Most frequent negative words

corp_tweets %>%
  tokens() %>% 
  tokens_keep(pattern = dict_english$Negative) %>%
  dfm() %>% 
  topfeatures(n = 30)

# Most frequent words related to discrete emotion anger
corp_tweets %>% 
  tokens() %>% 
  tokens_keep(pattern = dict_english$Anger) %>% 
  dfm() %>% 
  topfeatures(n = 30)

# Most frequent words related to discrete emotion fear
corp_tweets %>% 
  tokens() %>% 
  tokens_keep(pattern = dict_english$Fear) %>% 
  dfm() %>% 
  topfeatures(n = 30)

FINAL_dat_with_dic_tweets

# Getting rid of NA values 

FINAL_dat_with_dic_tweets_NO_NA <- na.omit(FINAL_dat_with_dic_tweets) 

# Getting the summary 

kable(summary(FINAL_dat_with_dic_tweets_NO_NA))


FINAL_dat_with_dic_tweets_NO_NA$average_positive <- mean(FINAL_dat_with_dic_tweets_NO_NA$positive)
FINAL_dat_with_dic_tweets_NO_NA$average_negative <- mean(FINAL_dat_with_dic_tweets_NO_NA$negative)
FINAL_dat_with_dic_tweets_NO_NA$average_anger <- mean(FINAL_dat_with_dic_tweets_NO_NA$anger)
FINAL_dat_with_dic_tweets_NO_NA$average_fear <- mean(FINAL_dat_with_dic_tweets_NO_NA$fear)


average_positive <- mean(FINAL_dat_with_dic_tweets_NO_NA$positive)
average_negative <- mean(FINAL_dat_with_dic_tweets_NO_NA$negative)
average_anger <- mean(FINAL_dat_with_dic_tweets_NO_NA$anger)
average_fear <- mean(FINAL_dat_with_dic_tweets_NO_NA$fear)


# Plot multivariable for sentiment on entire dataset 

ggplot(data=df1, aes(x=name, y=value)) + geom_bar(stat="identity") + labs(x = "Average Sentiment and Discrete Emotions", y = "Values")




```
_H2: Sentiment Analysis Using NRC lexicon on subset of dataset 2010-09-01/2010-12-31_
```{r, echo = FALSE}
# tokenise and apply dictionary

test_with_dic <- corp_subset %>% 
  tokens() %>% 
  tokens_remove(stopwords(language = "en", source = 'marimo')) %>% 
  tokens(remove_punct = TRUE) %>% 
  tokens_lookup(dictionary = dict_english, nested_scope = "dictionary") %>% 
  dfm()

test_with_dic <- quanteda::convert(test_with_dic, to = "data.frame")

View(test_with_dic)

summary(test_with_dic)

# Most frequent positive words
corp_subset %>%
  tokens() %>% 
  tokens_keep(pattern = dict_english$Positive) %>%
  dfm() %>% 
  topfeatures(n = 30)

# Most frequent negative words

corp_subset %>%
  tokens() %>% 
  tokens_keep(pattern = dict_english$Negative) %>%
  dfm() %>% 
  topfeatures(n = 30)

# Most frequent words related to discrete emotion anger
corp_subset %>% 
  tokens() %>% 
  tokens_keep(pattern = dict_english$Anger) %>% 
  dfm() %>% 
  topfeatures(n = 30)

# Most frequent words related to discrete emotion fear
corp_subset %>% 
  tokens() %>% 
  tokens_keep(pattern = dict_english$Fear) %>% 
  dfm() %>% 
  topfeatures(n = 30)

test_with_dic

# Getting the summary 

kable(summary(test_with_dic))

test_with_dic$average_positive <- mean(test_with_dic$positive)
test_with_dic$average_negative <- mean(test_with_dic$negative)
test_with_dic$average_anger <- mean(test_with_dic$anger)
test_with_dic$average_fear <- mean(test_with_dic$fear)

average_positive <- mean(test_with_dic$positive)
average_negative <- mean(test_with_dic$negative)
average_anger <- mean(test_with_dic$anger)
average_fear <- mean(test_with_dic$fear)


#create data frame with two columns
df2 <- data.frame(
  name = c("Average Positive", "Average Negative", "Average Anger", "Average Fear"),
  value = c(average_positive, average_negative, average_anger, average_fear)
)
# Plot multivariable for sentiment on subset 

ggplot(data=df1, aes(x=name, y=value)) + geom_bar(stat="identity") + labs(x = "Average Sentiment and Discrete Emotions", y = "Values") 







```


_ H1: Sentiment Analysis Using the Lexicoder Sentiment Dictionary_
```{r, echo = FALSE} 
# Tokenise and apply dictionary

dat_dict <- corp_tweets %>% 
  tokens() %>% 
  tokens_remove(stopwords(language = "en", source = 'marimo')) %>% 
  tokens(remove_punct = TRUE) %>% 
  tokens_lookup(dictionary = data_dictionary_LSD2015,
                nested_scope = "dictionary") %>% 
  dfm()

FINAL_dat_dict <- quanteda::convert(dat_dict, to = "data.frame")
View(FINAL_dat_dict)
summary(FINAL_dat_dict)

# SIMPLE FREQUENCY ANALYSIS  

## Most Frequent Words 

data(corp_tweets, package = "quanteda.textmodels")

tweets_dfm <- tokens(corp_tweets, remove_punct = TRUE, 
                     remove_symbols = TRUE) %>%
  tokens_remove(stopwords("en")) %>%
  dfm()
textstat_frequency(tweets_dfm, n = 10)

## Plotting the most frequent words 

tweets_freq_plot <- tweets_dfm %>%
  textstat_frequency(n = 50) %>%
  ggplot(aes(x = reorder (feature, frequency), 
             y = frequency)) + 
  geom_point() +
  coord_flip() +
  labs(x = NULL, y = "Frequency")

 tweets_freq_plot 

## Most Frequent Positive Words

corp_tweets %>% 
  tokens() %>% 
  tokens_keep(pattern = data_dictionary_LSD2015$positive) %>%
  dfm() %>% 
  topfeatures(n = 30)

# Most Frequent Negative Words

corp_tweets %>% 
  tokens() %>% 
  tokens_keep(pattern = data_dictionary_LSD2015$negative) %>%
  dfm() %>% 
  topfeatures(n = 30)

# Estimate Sentiment 

FINAL_dat_dict <- FINAL_dat_dict %>%
  mutate(sentiment = log((positive + neg_negative + 0.5) / (negative + neg_positive + 0.5)))

FINAL_dat_dict$doc_id <- corp_tweets$doc_id

# Plot Sentiment H1

p_sent <- ggplot(data = FINAL_dat_dict,
                 mapping = aes(x = sentiment, 
                     y = row.names(FINAL_dat_dict))) + geom_col() + labs(x = "Estimated Sentiment", y = "Pakistani Civilians Tweets") 
p_sent


```

_ H3: plot Sentiment Analysis on kids srike_
```{r, echo = FALSE} 

# getting the averages for fear anger and sentiment for individual dates in two datasets
#df10 <- data.frame(
  #name = c("Average Positive", "Average Negative", "Average Anger", "Average Fear"),
  #value = c(average_positive, average_negative, average_anger, average_fear),
  #date = c("2009-10-30")
#)

appendedDf_Kids <- rbind(df1, df2, df3)

appendedDf_Militants <- rbind(df5, df6, df7, df8, df9, df10)



# targeted sentiment and discrete emotion analysis june 2009

p_targeted <- ggplot(data = appendedDf_Kids,
aes(x = value,
y = reorder(date, value))) + geom_point() +
facet_grid(name~., space = "free", scales = "free_y") + #
labs(x = "Average Sentiment and Discrete Emotion", y = "Dates")

p_targeted

# targeted sentiment and discrete emotion analysis october 2009

p_targeted2 <- ggplot(data = appendedDf_Militants,
aes(x = value,
y = reorder(date, value))) + geom_point() +
facet_grid(name~., space = "free", scales = "free_y") + #
labs(x = "Average Sentiment and Discrete Emotion", y = "Dates")

p_targeted2

```

Note that the `echo = FALSE` parameter was added to the code chunk to prevent printing of the R code that generated the plot.