incivility_networks_scripts.R

# Required Libraries #

library(tidyverse)
require(lubridate)
library(igraph)
library(ineq)
require(tidytext)
require(viridis)
library(network)
library(ggraph)
library(ggrepel)


### MP HANDLES ## - from other R script - so we can match them to the tweet mentions
mp_df <- read.csv("data/combined_MP_twitter_handles.csv", stringsAsFactors = F)
# summary(mp_df)



# Function to calculate the structural properties on the giant component of the RT network #
network_property_calc <- function(network){
  
  
  largest_component <- components(network)$membership
  largest_component_id <- which.max(table(largest_component))
  network <- induced_subgraph(network, which(largest_component == largest_component_id))
  
  
  degree_centrality <- degree(network, mode = "all")
  closeness_centrality <- closeness(network, normalized = T)
  betweenness_centrality <- betweenness(network, normalized = T) # by the n of vertex pairs
  eigenvector_centrality <- eigen_centrality(network)$vector
  Density <- edge_density(network)
  Assortativity <- assortativity_degree(network)
  Clustering_coeff <- transitivity(network, type = "localaverage")
  Clustering_coeff_gini <- Gini(transitivity(network, type = "localaverage"))
  Transitivity_global <- transitivity(network, type = "global")
  Geodesic_mean <- mean_distance(network)
  Geodesic_gini <- Gini(distances(network))
  Geodesic_max <- diameter(network)
  
  # Network-level metrics
  Density <- edge_density(network, loops = FALSE)
  Assortativity <- suppressWarnings(assortativity_degree(network))
  Clustering_coeff <- suppressWarnings(transitivity(network, type = "localaverage"))
  Clustering_coeff_gini <- suppressWarnings(Gini(transitivity(network, type = "local", isolates = "zero")))
  Transitivity_global <- transitivity(network, type = "global")
  Geodesic_mean <- suppressWarnings(mean_distance(network, directed = TRUE))
  Geodesic_gini <- suppressWarnings(Gini(distances(network)))
  Geodesic_max <- suppressWarnings(diameter(network, directed = TRUE))
  
  # Output dataframe
  result <- data.frame(
    n_nodes = gorder(network),
    n_edges = gsize(network),
    Density,
    avg_degree = mean(degree(network, mode = "all")),
    avg_in_degree = mean(degree(network, mode = "in")),
    avg_out_degree = mean(degree(network, mode = "out")),
    Gini_Degree = ineq::Gini(degree_centrality),
    Closeness = mean(closeness_centrality, na.rm = T),
    Betweenness = mean(betweenness_centrality),
    Eigenvector = mean(eigenvector_centrality),
    Gini_Closeness = ineq::Gini(closeness_centrality),
    Gini_Betweenness = ineq::Gini(betweenness_centrality),
    Gini_Eigenvector = ineq::Gini(eigenvector_centrality),
    Assortativity,
    Clustering_coeff,
    Clustering_coeff_gini,
    Transitivity_global,
    Geodesic_mean,
    Geodesic_gini,
    Geodesic_max
  )
  
  return(result)
  
}


## Creates RT networks from tweets mentions ##

retweet_networks <- function(data, month_name){
  
  # Extracting the twitter mentions ##
  Rtweets_data <- data %>%
    mutate(mentions = str_extract_all(text, "@\\w+"),
           text = str_replace_all(text, "[[:punct:]]", ""))  %>% 
    unnest(mentions) %>%
    select(-text) %>% # remove text as we don't use that variable
    mutate(mentions = str_replace_all(mentions, "@", ""),
           mentions_lc = str_to_lower(mentions),
           politician = if_else(str_to_lower(mentions) %in% str_to_lower(mp_df$handles), 1, 0),
           across(where(is.character), as.factor))
  
  
  # We extract the mentions per tweet and the edge is the tweet itself but an edge attribute is
  # the n of rtweets (size of edge) and label/civility (color)

  # create user attribs
  user_attribs <- data %>%
    mutate(screen_name = str_replace_all(screen_name, "[^[:ascii:]]", ""), 
           screen_name = as.character(screen_name)) %>% 
    group_by(screen_name) %>%
    summarise(
      friendships = mean(friendships,na.rm = TRUE), #avg as they may have tweeted multiple times
      uncivility_prop = sum(classification =="uncivil")/ n(), # calculates the proportion of uncivil out of all
      rtweets = mean(rtweets,na.rm = TRUE), 
      politician =  if_else(str_to_lower(screen_name) %in% str_to_lower(mp_df$handles), 1, 0),
      .groups = 'drop'
    ) %>% 
    select(screen_name,uncivility_prop, friendships, rtweets,politician) %>% 
    ungroup() %>% 
    unique
  
  # create the edge attribs for our tweeting users
  edge_attributes <- Rtweets_data %>%
    group_by(screen_name,mentions_lc) %>%
    summarise(
      mean_RTs = mean(rtweets, na.rm = TRUE),
      classification = first(classification), # whether civil/uncivil
      .groups = 'drop'
    ) %>% 
    mutate(screen_name = as.character(screen_name))%>% 
    rename(from = screen_name, to = mentions_lc)
  
  
  # subset data for ONLY uncivil tweets
  Rtweets_data_uncivil <- Rtweets_data %>%
    filter(classification =="uncivil")
  
  # initial edgelist
  graph_edgelist <- Rtweets_data_uncivil %>%
    select(from = screen_name, to = mentions_lc) %>% 
    mutate(across(where(is.factor), as.character))
  
  
  # create the network from the data
  month_network <- graph_from_data_frame(graph_edgelist, directed =TRUE)
  
  #removes the loop and/or multiple edges from a graph
  # month_network <- simplify(month_network)
  
  
  # check which sources users are in the network
  source_users <- intersect(user_attribs$screen_name, graph_edgelist[[1]])
  
  # check which target users are in the network
  target_users <- intersect(str_to_lower(user_attribs$screen_name), graph_edgelist[[2]])
  
  # subsetting user attribs to actual nodes in network
  user_attribs_filter <- user_attribs %>%
    filter(screen_name %in% source_users | str_to_lower(screen_name) %in% target_users) 
  
  # assigning user attribs 
  labels1 <- setNames(user_attribs_filter$friendships, user_attribs_filter$screen_name)
  labels2 <- setNames(user_attribs_filter$uncivility_prop, user_attribs_filter$screen_name)
  labels3 <- setNames(user_attribs_filter$politician, user_attribs_filter$screen_name)
  
  
  # Match and assign the attribute to node
  V(month_network)$popularity <- labels1[match(V(month_network)$name, names(labels1))]
  V(month_network)$uncivility_prop <- labels2[match(V(month_network)$name, names(labels2))]
  V(month_network)$politician <- labels3[match(V(month_network)$name, names(labels3))]
  V(month_network)$label <- ifelse(V(month_network)$politician == 1, V(month_network)$name, NA)
  
  graph_edges_df <- as_data_frame(month_network, what = "edges") 
  
  #remove the dupli edges
  edge_attributes_filter <- graph_edges_df %>% 
    left_join(edge_attributes, by = c("from", "to"))
  
  # assign edge attribs
  E(month_network)$RTs <- edge_attributes_filter$mean_RTs
  E(month_network)$type <- edge_attributes_filter$classification
  E(month_network)$type <- as.character(E(month_network)$type)
  
  
  ## network property calculations
  property_summary <- network_property_calc(month_network)
  
  property_summary <- property_summary %>%
            mutate(month = month_name,
                   across(where(is.numeric), \(x) round(x, 3)))
  
  # export network properties
  write.csv(property_summary, paste0("uncivil_RT_structural_properties_",  month_name, ".csv"), row.names = FALSE)
  
  
  # data saving - convert into a save-able object
  graph_edgelist <- as_edgelist(month_network)
  
  # change col names
  graph_edgelist %>% 
    as.data.frame() %>% 
    select("source" = "V1", "target"="V2") %>% 
    write.csv(paste0("uncivil_RT_edgelist_",  month_name, ".csv"), row.names = FALSE)

  #save csv
  write_graph(month_network, file = paste0("uncivil_RT_graph_",  month_name, ".graphml"), format = "graphml")
  
  
  ## creating a df of the node attribs (as it doesn't get saved otherwise)
  vertex_attributes <- data.frame(
    id = V(month_network)$name,
    popularity = V(month_network)$popularity,
    uncivility_prop = V(month_network)$uncivility_prop,
    politician = V(month_network)$politician
  )
  
  # Exporting to CSV 
  write.csv(vertex_attributes,  paste0("uncivil_RT_node_attributes", month_name, ".csv"), row.names = FALSE)
  
}

## Let's make some graphs 
retweet_networks(tweets_jan,"January")
retweet_networks(tweets_feb,"February")
retweet_networks(tweets_march,"March")
retweet_networks(tweets_april,"April")
retweet_networks(tweets_may,"May")
retweet_networks(tweets_june,"June")


## Importing monthly network properties
tweets_jan_properties <- read.csv("outputs/RT networks/uncivil_RT_structural_properties_January.csv", stringsAsFactors = F)
tweets_feb_properties <- read.csv("outputs/RT networks/uncivil_RT_structural_properties_February.csv",header = T, stringsAsFactors = F)
tweets_march_properties <- read.csv("outputs/RT networks/uncivil_RT_structural_properties_March.csv",header = T, stringsAsFactors = F)
tweets_april_properties<- read.csv("outputs/RT networks/uncivil_RT_structural_properties_April.csv",header = T, stringsAsFactors = F)
tweets_may_properties <- read.csv("outputs/RT networks/uncivil_RT_structural_properties_May.csv",header = T, stringsAsFactors = F)
tweets_june_properties <- read.csv("outputs/RT networks/uncivil_RT_structural_properties_June.csv",header = T, stringsAsFactors = F)

# # combining them again 
all_tweets_properties <- rbind(tweets_jan_properties,tweets_feb_properties,tweets_march_properties,
                               tweets_april_properties,tweets_may_properties,tweets_june_properties)

#Latex table of combined properties
properties_latex <- kable(all_tweets_properties, format = "latex", booktabs = TRUE)
properties_latex

# export data into csv
write.csv(all_tweets_properties, "Combined_monthly_RT_network_properties.csv", row.names = FALSE)