Social Network Analysis of Tweet Bigrams from Sentiment of four popular Learning Management Systems

Early work using ggraph and igraph

1. PURPOSE

1a. Motivation and Focus

We completed a sentiment analysis on four popular *Learning Management Systems (LMS); Google Classroom, Canvas, Moodle, and Blackboard, in a previous case study. We evaluated emotions and sentiment towards each LMS by assessing the most common uniwords. We evaluated strengths and weaknesses by pulling public opinion from the Twitter Resting API.

We will evaluate connections between the most frequent combination of words (Bigrams) through Social Network Analysis. Here we’ll explore this option for the tweets made in English for the 4 LMS previously observed.

Guiding Questions:

1. Are there networks from the co-occurrence of Bigrams?
2. What does the mathematical analysis of the Bigram network tell us?

1b. Load Libraries

Let’s first load our libraries to read in packages that we will use to answer our questions. We will also create a function called replace_reg() by looking for and deleting strings (nonsense words), numbers, and other manual stop characters and numbers.

Show code

library(tidytext)
library(tidyverse)
library(network)
library(sna)
library(visNetwork)
library(threejs)
library(ndtv)
library(qgraph)
library(splitstackshape)
library(tidyr)
library(stringr)
library(readxl)
library(readr)

# For visualizations
library(vtree)
library(igraph)
library(ggraph)
library(tidygraph)
library(networkD3)
library(ggplot2)

# regex for parsing tweets
replace_reg <- "https?://[^\\s]+|&amp;|&lt;|&gt;|&d2l;|&aristotlemrs;|&aleks;|\bRT\\b"

# Custom Color Palette
my_colors <- c("#05A4C0", "#85CEDA", "#D2A7D8", "#A67BC5", "#BB1C8B", "#8D266E")

2. METHOD

Our initial read-in data frame includes 4521 tweets objects in the text to evaluate. After tokenizing the Bigrams and restructuring the data objects, we will include Bigrams mentioned more than five times. Once tidyed, the data consists of 1407 bigrams left to evaluate with a social network approach.

2a. Read and Restructure Data

• Read in the previously evaluated LMS data for Google Classroom tweets, Blackboard tweets, Canvas tweets, and Moodle tweets.
• Subset columns to pull only index, lms and text columns.
• Visualize the initial number of tweets for each LMS.

Show code

tweets <- read_excel("data/tweets.xlsx")

#select lms, text and index, grouping by lms
tweets_data2 <- tweets %>% 
  select(c('index', 'lms', 'text')) %>% 
  group_by(lms) %>%
  na.omit()


vtree(tweets_data2, "lms", horiz=FALSE, palette = 4, sortfill = TRUE, title="Initial LMS Tweets Data")

2b. Tidytext and Initiate Bigram

• Using the tidytext and dplyr packages we split the text into tokens creating a table with two-tokens-per-row. The token is under a column called “bigram.”
• Subset the bigram columns by separating and adding two columns “first” and “second.””
• Remove stop words and manual stop words using our str_detect() dictionary and remove words that are not letter strings.
• Group our lms_bigram() to count up the bigrams, summarize keeping only those that appear more than 5 times.
• Visualizing the new numbers for each LMS.

Show code

# split into word pairs
lms_bigrams <- tweets_data2 %>% 
  mutate(text = str_replace_all(text, replace_reg, "")) %>%
  unnest_tokens(bigram, text, token = "ngrams", n = 2)

# remove stop words
lms_bigrams <- lms_bigrams %>%
  separate(bigram, into = c("first","second"), sep = " ", remove = FALSE) %>%
  anti_join(stop_words, by = c("first" = "word")) %>%
  anti_join(stop_words, by = c("second" = "word")) %>%
  filter(str_detect(first, "[a-z]") &
         str_detect(second, "[a-z]"))

#count up new birgams and create a new column called n only keep more than 5 counts
lms_bigrams_count <- lms_bigrams %>%
  group_by(lms, bigram, first, second)%>%
  summarise(n=n())%>%
  filter(n >= 5)%>%
  arrange(-n)%>%
  ungroup()

#visualize new number of rows (previously counting tweets)
vtree(lms_bigrams_count, "lms", horiz=FALSE, palette = 4, sortfill = TRUE, title="Bigram LMS Tweets Data")

3. EXPLORE

3a. Subset Coloumns and visualize Bigrams

• Select first, second and n columns create lms_bigram_tble data frame to use later in the Social Network Analysis.
• Visually inspect Bigrams mentioned more than 35 times through an igraph to observe any communities in the network.

Show code

# Rename and reorder columns (so we can make the graphs more easily)
lms_bigram_tbl <- lms_bigrams_count %>%
  dplyr::select(c('first','second', 'n'))


bigram_graph <- lms_bigram_tbl %>%
  filter(n > 35) %>%
  graph_from_data_frame()


set.seed(123)

a <- grid::arrow(type = "closed", length = unit(.15, "inches"))

ggraph(bigram_graph, layout = "fr") +
    geom_edge_link(aes(edge_alpha = n), show.legend = FALSE, arrow = a) +
    geom_node_point(color = "lightblue", size = 5) +
    geom_node_text(aes(label = name), vjust = 1, hjust = 1) +
    theme_void()

4. MODEL

To evaluate Social Network for Bigrams we will create a igraph and table graph class network. This will easily allow for visualizing the connections and explaining the network mathmatically.

4a. Edges and Nodes

Nodes are the unique words - each word has an identification ID edges are the bigrams, meaning that they show how frequently we find a combination of 2 words (represented by their unique ID).

• Create a source for the first word of the bigram
• Create a destination for the second word of the bigram
• Create Nodes
• Create Edges
• Select only to, from and weight
• Inspect i. head of Node and ii. head of Edges and save for later use.

Show code

#filter to Bigrams that are mentioned more than 35 times.
lms_df <- lms_bigram_tbl %>%
  filter(n > 35) 
# Distinct first (part of bigram)
sources <- lms_df%>% 
                distinct(first) %>% 
                rename(label = first)

# Distinct second (part of bigram)
destinations <- lms_df %>% 
                    distinct(second) %>% 
                    rename(label = second)

#NODES AND EDGES BELOW:

# ----- NODES -----
# Unique Items + create unique ID
nodes <- full_join(sources, destinations, by="label") %>% rowid_to_column("id")

# ----- EDGES -----
# Adds unique ID of Item 1 to data
edges <- lms_df %>% 
            left_join(nodes, by = c("first" = "label")) %>% 
            rename(from = id)

# Adds unique ID of Item 2 to data
edges <- edges %>% 
            left_join(nodes, by = c("second" = "label")) %>% 
            rename(to = id) %>% 
            rename(weight = n)

# Select only From | To | Weight (frequency)
edges <- edges %>% select(from, to, weight)

1. Inspect head of node
   Show code

   # inspect head of nodes and edges
   nodes %>% 
     head(5)
   

# A tibble: 5 x 2
     id label  
  <int> <chr>  
1     1 google 
2     2 canvas 
3     3 essay  
4     4 pearson
5     5 aleks