Analysis of #IMGC14 tweets

IMGC14_twitter_analysis.R

library(twitteR)
library(tm)
library(wordcloud)
library(RColorBrewer)
library(dplyr)
library(ggplot2)

# load tweets
download.file("ftp://ftp.jax.org/petrs/other/IMGC14.rds", destfile = "IMGC14_tweets.rds", mode="wb")
tweet <- readRDS("IMGC14_tweets.rds")

# extract time, user, fav.count and retweet count
dt <- data.frame(time = sapply(tweet, function(x) as.character(x$created)), 
                 user = sapply(tweet, function(x) x$screenName),
                 nRt  = sapply(tweet, function(x) x$retweetCount),
                 nFav = sapply(tweet, function(x) x$favoriteCount))
# time zone change
dt$time <-  as.POSIXct(format(as.POSIXct(dt$time, tz="UTC"), tz="America/Thunder_Bay"))

### Number of tweets per user
dt %>% group_by(user) %>% 
  summarise(n=n()) %>%
  arrange(-n, user) %>%
  filter(n>1) %>%
  ggplot(aes(x=reorder(user,n),y=n)) +
    geom_bar(stat = "identity") + 
    ylab("Number of tweets") + 
    xlab("User") +
    ggtitle("Users with at least two #IMGC14 tweets") +
    coord_flip() +
    theme(panel.border = element_rect(colour = 'darkgrey', fill = NA)) 
ggsave("users.jpeg")

# number of users (79) 
n_distinct(dt$user)

# number of tweets (1546)
nrow(dt)

# number of Steve Munger's tweets (679)
dt %>% group_by(user) %>% 
  summarise(n=n()) %>%
  arrange(-n, user) %>%
  head(n=1)


### Time distribution of tweets

ggplot(aes(x=time), data=dt) + 
  geom_density(fill="#1B9E77", col="#1B9E77", adjust=0.2) +
  xlim(as.POSIXct("2014-10-26 4:00:00 EDT"), as.POSIXct("2014-10-30 04:00:00 EDT")) +
  xlab("Time") + ylab("Intensity of tweeting") + ggtitle("Twitter timestamps") +
  theme(panel.border = element_rect(colour = 'darkgrey', fill = NA)) 
  

### Most retweeted and favorited tweets
tweet[which.max(dt$nRt)]
tweet[which.max(dt$nFav)]

### WORD CLOUD

# make corpus
tweet_text <- sapply(tweet, function(x) x$getText())
tweet_text_corpus <- Corpus(VectorSource(tweet_text))

# clean corpus
tweet_text_corpus <- tm_map(tweet_text_corpus, content_transformer(tolower))
tweet_text_corpus <- tm_map(tweet_text_corpus, removePunctuation)
tweet_text_corpus <- tm_map(tweet_text_corpus, function(x) removeWords(x,stopwords("english")))
tweet_text_corpus <- tm_map(tweet_text_corpus, function(x) removeWords(x,c("imgc14", "stevemunger", "redsoxgal0407", "mice")))

# plot wordcloud
wordcloud(tweet_text_corpus, scale=c(3.5,0.5), max.words=100, min.freq=2,
          random.order=FALSE, rot.per=0.35, use.r.layout=FALSE, colors=brewer.pal(8, "Dark2"))