library(rtweet)
library(dplyr)

tmls <- get_timeline(c("BillGates", "elonmusk", "Wendys"), n=5000)
# Remove retweeted, quotes, and reply tweets
tmls <- tmls %>% filter(is.na(reply_to_user_id) & !is_quote & !is_retweet & is.na(media_type) & lang == "en")

# show sample
knitr::kable(tmls[sample(1:nrow(tmls),5),] %>% select(user_id, created_at, screen_name, text))

for (k in 1:nrow(tmls)) {
  if (!is.na(tmls$urls_url[[k]])) {
    urls <- tmls$urls_t.co[[k]]
    for (j in 1:length(urls)) {
      tmls$text[k] <- gsub(urls[j], "", tmls$text[k], fixed=T)
    }
  }
}

tmls$text <- trimws(tmls$text) # trim whitespace
tmls$text <- gsub("’", "\\'", tmls$text)
tmls$text <- gsub("[\r\n]", " " , tmls$text) # remove new lines
tmls$text <- gsub(":", "", tmls$text, fixed = T)
tmls$text <- gsub("&amp;", "&", tmls$text, fixed=T)
tmls$text <- gsub("&", "", tmls$text, fixed=T)
tmls$text <- gsub("\\s+", " ", tmls$text)

library(quanteda)

haikus <- NULL

for (k in 1:nrow(tmls)) {
  textSplit <- strsplit(tmls$text[k], "\\s")
  syl <- lapply(textSplit, nsyllable)[[1]]
  if (sum(syl, na.rm=T) == 17) {
    syl[is.na(syl)] <- 0
    cs <- cumsum(syl)
    # Check if this contains 5, 12, 17 (cumsum)
    if (all(c(5,12,17) %in% cs)) {
      # parse out info
      idx_v1 <- 1:which(cs == 5)[1]
      idx_v2 <- (which(cs == 5)[1] + 1):which(cs == 12)[1]
      idx_v3 <- (which(cs == 12)[1] + 1):which(cs == 17)[1]
      haikus <- rbind(haikus, cbind(data.frame(tmls[k,c("user_id","status_id","created_at","screen_name","text")]),  
                                    data.frame(verse1=paste0(textSplit[[1]][idx_v1], collapse=" "), 
                                               verse2=paste0(textSplit[[1]][idx_v2], collapse=" "),
                                               verse3=paste0(textSplit[[1]][idx_v3], collapse=" "))))
    }
  }
}

knitr::kable(haikus, align="c")