R/get_sentiment_tweets.R

In TwitterAutomatedTrading: Automated Trading Using Tweets

#' get_sentiment_tweets
#'
#'
#' This function computes the sentiment from tweets. Remind to connect with twitter using your API Key.
#'
#' @param ntweets Number of tweets to be searched
#' @param time_tweet Time in hours where the tweets will be filtered
#' @param terms_list Terms to be searched
#' @param time_zone The time zone
#' @param positive_dictionary The list of positive terms of the dictionary
#' @param negative_dictionary The list of negative terms of the dictionarya tibble with the words counting
#' @param sentiment_index_type The sentiment type to be used according to the dictionary, positive, negative or both. Default is both, positive and negative
#'
#' @return A list with: (1) - the sentiment index, (2) a tibble with the words counting, (3) a tibble with the negative words counting and (4
#' @export
#' @importFrom dplyr bind_rows
#' @importFrom dplyr count
#' @importFrom dplyr filter
#' @importFrom dplyr tbl_df
#' @import twitteR
#' @importFrom purrr map_df
#' @importFrom tidytext unnest_tokens
#' @import magrittr
#'
#'
#' @examples
#' \dontrun{
#' #Not run:
#' ntweets <- 500
#' time_tweet <- 6
#' terms_list <- c("IBOVESPA OR bovespa OR ibov OR petroleo OR $SPX OR $SPY OR $EWZ")
#' time_zone <- "Brazil/East"
#' positive_dictionary <- my_dictionary[['positive_terms']]
#' negative_dictionary <- my_dictionary[['negative_terms']]
#' sentiment_index <- get_sentiment_tweets(ntweets = ntweets,
#' terms_list = terms_list,
#' time_tweet = time_tweet,
#' time_zone = time_zone,
#' positive_dictionary = positive_dictionary,
#' negative_dictionary = negative_dictionary
#' )
#'
#' sent_idx <- sentiment_index[[1]]
#' sent_wrd <- sentiment_index[[2]]
#' sent_pos <- sentiment_index[[3]]
#' sent_neg <- sentiment_index[[4]]
#' }
#'
get_sentiment_tweets <- function(ntweets,
                                 time_tweet,
                                 terms_list,time_zone,
                                 positive_dictionary,
                                 negative_dictionary,
                                 sentiment_index_type){
  n <- ntweets
  tm <- time_tweet*3600


  aa <- twitteR::searchTwitter(terms_list,n=n)
  aaa <- dplyr::tbl_df(purrr::map_df(aa, as.data.frame))


  if(missing(sentiment_index_type)){
    sentiment_index_type <- 'both'
    }





  #tweets_df=aaa
  tweets_df <- dplyr::bind_rows(aaa)
  tweets_df$created <- format(tweets_df$created,tz = time_zone)
  st_date <- as.POSIXct(Sys.time()-tm)
  start_date <- format(st_date,tz= time_zone)
  e_date <- as.POSIXct(Sys.time())
  end_date <- format(e_date,tz= time_zone)
  tweets_df1 <- unique(tweets_df %>%
                         dplyr::filter(created >= start_date & created <= end_date)
  )

  tweets_df2 <- tweets_df1 %>%
    dplyr::filter(isRetweet == FALSE)




  words <- tweets_df2 %>%
    tidytext::unnest_tokens(word,text)
  words$word <- tolower(words$word)

  word_count <- words %>%
    dplyr::count(word,sort=TRUE)


  pos <-  positive_dictionary

  pos_fil<- dplyr::filter(words, word %in% pos)
  positive_count <- pos_fil %>%
    dplyr::count(word,sort=TRUE)
  positive=sum(positive_count$n)

  neg <- negative_dictionary
  neg_fil<- dplyr::filter(words, word %in% neg)
  negative_count <- neg_fil %>%
    dplyr::count(word,sort=TRUE)
  negative=sum(negative_count$n)

  if(sentiment_index_type == 'positive') {
  LG_sent_score=(positive)/(1 + (positive+negative) )
  LG_sent_score[is.na(LG_sent_score)] = 0
  } else if(sentiment_index_type == 'negative') {
    LG_sent_score=(negative)/(1 + (positive+negative) )
    LG_sent_score[is.na(LG_sent_score)] = 0
  } else {
    LG_sent_score=(positive-negative)/(1 + (positive+negative) )
    LG_sent_score[is.na(LG_sent_score)] = 0
      }

  outputs <- list(LG_sent_score , word_count, positive_count , negative_count )
  return(outputs)
}