Código de análise de dados da matéria "De cobras a funk: a guinada na comunicação digital do Instituto Butantan", Núcleo Jornalismo

butantan_redes.R

###################################################################################################
###################################################################################################
#########
######### "De cobras a funk: a guinada na comunicação digital do Instituto Butantan"
#########
######### Lucas Gelape
######### Analise de dados da materia publicada no Nucleo Jornalismo
#########

## Pacotes
library(dplyr)
library(lubridate)
library(stringr)
library(tidytext)
library(tidyr)
library(funprog)
library(purrr)
library(tm)

## Funcoes
# Cria um "nao esta contido em"
'%nin%' <- Negate('%in%')
# Remove strings de dentro de um character
limpar_texto <- function(x){

  t <- unlist(strsplit(x, " "))
  vetor <- t[t %nin% stopwords]
  texto <- paste(vetor, collapse = " ")

}

###################################################################################################

##### TWITTER

# Abre o banco de tweets ate 17/03
twitter <- readRDS("butantan_tweets.rds")

# Acerta fuso horario e cria variaveis para analises e grafico
twitter <- twitter %>%
  mutate(created_at = created_at - hours(3),
         dia_mes_ano = as.Date(created_at),
         mes = month(created_at),
         dia = day(created_at)) %>%
  mutate(engajamento = retweet_count + favorite_count) %>%
  mutate(mes = factor(mes,
                      levels = c(6:12, 1:3),
                      labels = c("Jun", "Jul", "Ago",
                                 "Set", "Out", "Nov",
                                 "Dez", "Jan", "Fev",
                                 "Mar")))

### MEDIA TWEETS/DIA

# Calcula a media de tweets por dia (grafico)
tweets_dia <- twitter %>%
  group_by(mes) %>%
  summarise(dias = length(unique(dia)),
            total_mes = n(),
            tweets_dia = total_mes/dias) %>%
  ungroup()

### ENGAJAMENTO

# Calcula o engajamento mensal de posts do Butantan
engajamento_mes_twitter <- twitter %>%
  # Elimina os RT, mantendo somente os posts de autoria do Butantan
  filter(!is_retweet == "TRUE") %>%
  group_by(mes) %>%
  summarise(engajamento_mensal = sum(engajamento)) %>%
  ungroup()

# 5 postagens com maior engajamento
twitter %>%
  filter(!is_retweet == "TRUE") %>%
  slice_max(engajamento, n = 5) %>%
  mutate(link = paste0("https://twitter.com/butantanoficial/status/", status_id)) %>%
  select(link)

### RESPOSTAS A USUARIOS

# N. absoluto e porcentagem de tweets que sao respostas
replies <- twitter %>%
  mutate(resposta = ifelse(!is.na(reply_to_screen_name) & reply_to_screen_name != "butantanoficial",
                           1, 0)) %>%
  group_by(mes) %>%
  summarise(resposta = sum(resposta),
            percentual = (resposta/n())*100) %>%
  ungroup()

### CONTAGEM DE PALAVRAS

# Abrir stopwords gerais
source("https://gist.githubusercontent.com/lgelape/edcc0250f21bcc5710c0a9fd0488d1ea/raw/960e11e214d6a32df5dd38e4d0f251a992b46d57/stopwords_pt.R")

# Stopwords identificadas na analise
stopwords_butantan <- data.frame(word = c("https", "t.co", "butantan", "equipebutantan", "instituto",
                                          "10", "120", "12h45", "19", "2020", "oi", "ola", "ne"),
                                 source = "analise_propria")

# Banco e vetor finais de stopwords
stopwords_pt_final_noaccent <- bind_rows(stopwords_pt_final_noaccent, stopwords_butantan)
stopwords <- stopwords_pt_final_noaccent$word

# Limpa texto dos tweets
twitter <- twitter %>%
  mutate(text = tolower(stri_trans_general(text, "Latin-ASCII")),
         text = removePunctuation(text))

# Aplica a funcao de remover as stopwords e guarda como vetor
textos_limpos <- map(.x = twitter$text,
                     .f = limpar_texto)
textos_limpos <- do.call(c, textos_limpos)

# Salva o vetor como coluna do banco de dados
twitter$textos_limpos <- textos_limpos

# Identifica as 5 palavras mais repetidas em cada periodo de analise
palavras_mais_repetidas <- twitter %>%
  unnest_tokens(word, textos_limpos) %>%
  select(dia_mes_ano, word) %>%
  mutate(periodo = ifelse(dia_mes_ano < as.Date("2020-12-01"), 1, 2)) %>%
  group_by(periodo) %>%
  count(word) %>%
  slice_max(n, n = 5) %>%
  arrange(desc(n)) %>%
  mutate(posicao = rank(n)) %>%
  ungroup()

###################################################################################################

##### FACEBOOK

# Abre o banco de posts no Facebook
facebook <- readRDS("butantan_desde0106.rds")

# Cria variaveis uteis na analise
facebook <- facebook %>%
  mutate(dia_redondo = as.Date(date),
         mes_lubridate = month(date),
         dia = day(date)) %>%
  mutate(mes = factor(mes_lubridate,
                      levels = c(6:12, 1:3),
                      labels = c("Jun", "Jul", "Ago",
                                 "Set", "Out", "Nov",
                                 "Dez", "Jan", "Fev",
                                 "Mar"))) %>%
  mutate(engajamento =  actual_shareCount + actual_likeCount + actual_loveCount +
           actual_commentCount + actual_wowCount + actual_hahaCount +
           actual_sadCount + actual_angryCount + actual_thankfulCount +
           actual_careCount)

### POSTS/DIA

# Calcula o n. de posts por dia do mes
posts_dia <- facebook %>%
  group_by(mes) %>%
  summarise(dias = unique(days_in_month(mes_lubridate)),
            total_mes = n()) %>%
  ungroup() %>%
  mutate(dias = ifelse(mes == "Mar", 17, dias),
         tweets_dia = total_mes/dias)

###################################################################################################