beto_cruz_ads.R

library(dplyr)
library(tidyverse)
library(tidytext)
library(plotly)
library(stringr)

# Pull in spreadsheet
df <- read.csv("fbads.csv", header=TRUE, stringsAsFactors = FALSE) 
df$date <- as.Date(df$created_at) # Add date column
df %>% glimpse()

summary(df)

# Prompt users to look for a specific term or politician
betomentions <- df %>%
  filter(str_detect(message, "Beto")) %>%
  mutate("candidate" = "O'Rourke")
betomentions %>% glimpse()

cruzmentions <- df %>%
  filter(str_detect(message, "Cruz")) %>%
  mutate("candidate" = "Cruz")  
cruzmentions %>% glimpse()

betocruz_mentions <- bind_rows(betomentions, cruzmentions) %>%
  mutate("Beto" = ifelse(advertiser == "Beto O'Rourke", T, F))

betocruz_mentions %>% glimpse()

ggplot(betocruz_mentions, aes(date, advertiser, text = message, color = Beto)) + 
  geom_point(aes(size=impressions)) +
  scale_color_manual(values=c("black", "dodgerblue")) +
  ggtitle("Beto O'Rourke and Ted Cruz Facebook ads by advertiser over time") +
  theme(legend.position = 'none')

# Plotly

allads <- ggplot(betocruz_mentions, aes(date, advertiser, text = message, color = Beto)) + 
  geom_point(aes(size=impressions)) +
  scale_color_manual(values=c("black", "dodgerblue")) +
  ggtitle("Beto O'Rourke and Ted Cruz Facebook ads by advertiser over time") +
  theme(legend.position = 'none')

ggplotly(allads, tooltip=c("text","impressions"))



# Top 20 advertisers

top_advs <- betocruz_mentions %>%
  count(advertiser, Beto) %>% 
  arrange(desc(n)) %>%
  top_n(20)

ggplot(top_advs, aes(fill = Beto, reorder(advertiser, n), n)) + 
  geom_bar(stat="identity") + 
  coord_flip() +
  scale_fill_manual(values=c("black", "dodgerblue")) +
  theme(legend.position = 'none') +
  xlab("advertiser") +
  ggtitle("Top 20 advertisers whose Facebook ads mention Beto or Cruz")



# Look at official ads

betoads <- df %>%
  filter(advertiser == "Beto O'Rourke") %>%
  mutate("candidate" = "O'Rourke")%>% glimpse()

cruzads <- df %>%
  filter(advertiser == "Ted Cruz") %>%
  mutate("candidate" = "Cruz") %>% glimpse()

betocruz <- bind_rows(betoads, cruzads) %>% glimpse()

ggplot(betocruz, aes(date, fill = candidate)) + geom_histogram() +
  scale_fill_manual(values=c("firebrick", "dodgerblue"))  +
  facet_wrap(~candidate)

ggplot(betocruz, aes(date, advertiser, text = message, color = candidate)) + 
  geom_point(aes(size=impressions)) +
  scale_color_manual(values=c("firebrick", "dodgerblue")) +
  theme(legend.position = 'none')



##### Sentiment analysis 

betocruz$post <- betocruz$message

tokenized_comments <- betocruz %>%
  select(paid_for_by, advertiser, post, candidate, date, impressions, title, message) %>%
  unnest_tokens(word, message) %>%
  anti_join(stop_words) %>%
  group_by(word, paid_for_by, advertiser, date, post, impressions, candidate, title) %>%
  tally() %>%
  arrange(desc(n))

tokenized_comments %>% glimpse()

sentiments <- read.csv("labMT2english.csv", sep ="\t" )
labMT <- sentiments %>%
  select(word, happs)

all_sentiment <- tokenized_comments %>%  
  inner_join(labMT, by = "word") %>%
  group_by(post, candidate, paid_for_by, advertiser, date, impressions, title) %>%
  summarize(sentiment = mean(happs)) %>%
  arrange(desc(sentiment)) %>%
  mutate("score" = sentiment-5.372) 

all_sentiment %>% glimpse()

cruz_sent <- all_sentiment %>%
  filter(advertiser == "Ted Cruz") %>% glimpse()
  
mean(cruz_sent$score)

beto_sent <- all_sentiment %>%
  filter(advertiser == "Beto O'Rourke") %>% glimpse()

mean(beto_sent$score)


# Plot sentiment over time
#betocruzsent1 <- ggplot(all_sentiment, aes(date, score)) + geom_smooth() + geom_point()+ facet_grid(~advertiser)

# Plot sentiment over time and split by advertiser

ggplot(all_sentiment, aes(date, score, color=score, text = post, score = score)) + geom_point() + geom_smooth() +
  facet_wrap(~candidate) +
  scale_color_gradient(low= "red", high="blue")+
  ggtitle("Sentiment of Beto O'Rourke and Ted Cruz Facebook ads over time")


betocruzsent <- ggplot(all_sentiment, aes(date, score, color=score, text = post, score = score)) + geom_point() + geom_smooth() +
  facet_wrap(~candidate) +
  scale_color_gradient(low= "red", high="blue")+
  ggtitle("Sentiment of Beto O'Rourke and Ted Cruz Facebook ads over time")

ggplotly(betocruzsent, tooltip=c("text","word"))

# Sentiment of overall words and by advertiser


betocruz_mentions$post <- betocruz_mentions$message

tokenized_all_ads <- betocruz_mentions %>%
  select(paid_for_by, advertiser, post, candidate, date, impressions, title, message) %>%
  unnest_tokens(word, message) %>%
  anti_join(stop_words) %>%
  group_by(word, paid_for_by, advertiser, date, post, impressions, candidate, title) %>%
  tally() %>%
  arrange(desc(n))

tokenized_all_ads %>% glimpse()

sentiments <- read.csv("labMT2english.csv", sep ="\t" )
labMT <- sentiments %>%
  select(word, happs)

all_ads_sentiment <- tokenized_all_ads %>%  
  inner_join(labMT, by = "word") %>%
  group_by(post, candidate, paid_for_by, advertiser, date, impressions, title) %>%
  summarize(sentiment = mean(happs)) %>%
  arrange(desc(sentiment)) %>%
  mutate("score" = sentiment-5.372) 

all_ads_sentiment %>% glimpse()

sent_advertiser <- all_ads_sentiment %>%
  group_by(advertiser) %>%
  summarize("avgsent" = mean(score)) %>%
  arrange(desc(avgsent))
sent_advertiser %>% glimpse()

# Plot of mean sentiment by advertiser
ggplot(sent_advertiser, aes(reorder(advertiser, avgsent), avgsent, fill=avgsent >0)) + 
  geom_bar(stat = "identity") + 
  scale_fill_manual(values=c("red", "blue")) + 
  theme(legend.position = 'none') +
  ylab("Average sentiment") +
  xlab("") +
  coord_flip()

ggplot(all_ads_sentiment, aes(date, score, color=score, text = post, score = score)) + geom_point() + geom_smooth() +
  facet_wrap(~candidate) +
  geom_smooth() +
  scale_color_gradient(low= "red", high="blue")+
  ggtitle("Sentiment of Beto O'Rourke and Ted Cruz Facebook ads over time")


View(all_ads_sentiment %>% filter(score < -0.5) %>% glimpse())