Take-home_Ex2

Author

Han Shumin

Published

May 19, 2023

Modified

June 11, 2023

Overview

The task

With reference to Mini-Challenge 2 of VAST Challenge 2023 and by using appropriate static and interactive statistical graphics methods, I will be revealing the:

Use visual analytics to identify temporal patterns for individual entities and between entities in the knowledge graph FishEye created from trade records. Categorize the types of business relationship patterns if any.

Methodology

To analyse the temporal patterns, timeline visualization and network visualization will be used to identify the patterns and explore the possible the business relationship between entities.

Visualizing Temporal Patterns:

Timeline Visualization: Create a timeline visualization that shows the activities of companies over time. Each company can be represented as a separate entity on the timeline, and their fishing activities (both legal and illegal) can be displayed as events or bars. This allows analysts to compare the temporal patterns of different companies and identify any suspicious behavior or recurring patterns.

Temporal Network Visualization:

Represent the relationships between companies and their fishing activities as a network, where nodes represent companies and edges indicate their fishing activities. By incorporating temporal information into the visualization, such as color-coding or varying the thickness of edges based on the time of occurrence or numbers of interactions between nodes, analysts can identify patterns of illegal fishing and observe if companies reappear under different names.

Data Preparation

Install and load the packages

The following code chunks will install and load the required packages.

Show the code

pacman::p_load(jsonlite, tidygraph, ggraph, visNetwork, tidyverse, igraph, ggiraph, ggplot2, ggthemes, patchwork, plotly, ggstatsplot, hrbrthemes)

Load the dataset in JSON format

In the code chunk below, from JSON() of jsonlite package is used to import mc2_challenge_graph.json into R environment.

Show the code

MC2 <- fromJSON("data/mc2_challenge_graph.json")

Data Wrangling

Extracting the nodes and links

The code chunk is used to extract nodes/edges data tables from MC2 list object and save the output in a tibble data frame object called MC2_nodes and MC2_edges.

Show the code

MC2_nodes <- as_tibble(MC2$nodes) %>%
  select(id, shpcountry, rcvcountry)



MC2_edges <- as_tibble(MC2$links) %>%
  mutate(Arrivaldate = ymd(arrivaldate)) %>%
  mutate(year = year(Arrivaldate)) %>%
  mutate(month = month(Arrivaldate)) %>%
  select(source, target, Arrivaldate, year, month, hscode, valueofgoods_omu, volumeteu, weightkg, valueofgoodsusd) %>%
  distinct()

mutate() is used two times to create two derive fields.
ymd() of lubridate package is used to covert arrivaldate field from character data type into date data type.
year() of lubridate package is used to convert the values in ArrivalDate field into year values.
month() of lubridate package is used to convert the values in ArrivalDate field into month values.
select() is used not only to select the field needed but also to re-organise the sequent of the fields.

Check for missing values

Show the code

#Check for missing values
any(is.na(MC2_nodes))

[1] TRUE

Show the code

any(is.na(MC2_edges))

[1] TRUE

Show the code

# Calculate the percentage of NA values in each column
nodes_na_pct <- colMeans(is.na(MC2_nodes)) * 100

# Print the NA percentages
print(nodes_na_pct)

        id shpcountry rcvcountry 
   0.00000   64.66624    8.41335

Show the code

# Calculate the percentage of NA values in each column
edges_na_pct <- colMeans(is.na(MC2_edges)) * 100

# Print the NA percentages
print(edges_na_pct)

          source           target      Arrivaldate             year 
         0.00000          0.00000          0.00000          0.00000 
           month           hscode valueofgoods_omu        volumeteu 
         0.00000          0.00000         99.99471          9.38167 
        weightkg  valueofgoodsusd 
         0.00000         54.36443

The valueofgoods_omu has 99% of na values, therefore we can remove this variable from the edges table. In the nodes table, we have 64.6% and 8.4$ na values in shipping country and receiving country columns, we can replace the missing country names with “others”.

Show the code

#drop the valueofgoods_omu column, and remove rows with missing value in volumnteu column
MC2_edges_clean <- MC2_edges[, -which(names(MC2_edges) == "valueofgoods_omu")]

glimpse(MC2_edges_clean)

Rows: 5,309,087
Columns: 9
$ source          <chr> "AquaDelight Inc and Son's", "AquaDelight Inc and Son'…
$ target          <chr> "BaringoAmerica Marine Ges.m.b.H.", "BaringoAmerica Ma…
$ Arrivaldate     <date> 2034-02-12, 2034-03-13, 2028-02-07, 2028-02-23, 2028-…
$ year            <dbl> 2034, 2034, 2028, 2028, 2028, 2028, 2028, 2028, 2028, …
$ month           <dbl> 2, 3, 2, 2, 9, 10, 4, 6, 9, 9, 2, 2, 4, 4, 3, 9, 3, 3,…
$ hscode          <chr> "630630", "630630", "470710", "470710", "470710", "470…
$ volumeteu       <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
$ weightkg        <int> 4780, 6125, 10855, 11250, 11165, 11290, 9000, 19490, 6…
$ valueofgoodsusd <dbl> NA, NA, NA, NA, NA, NA, 87110, 188140, NA, 221110, 586…

Show the code

# replace na with others in shipping country and receive country
MC2_nodes_clean <- dplyr::mutate(MC2_nodes, 
                                shpcountry = ifelse(is.na(shpcountry), "others", shpcountry),
                                rcvcountry = ifelse(is.na(rcvcountry), "others", rcvcountry))

glimpse(MC2_nodes_clean)

Rows: 34,576
Columns: 3
$ id         <chr> "AquaDelight Inc and Son's", "BaringoAmerica Marine Ges.m.b…
$ shpcountry <chr> "Polarinda", "others", "Oceanus", "others", "Oceanus", "Kon…
$ rcvcountry <chr> "Oceanus", "others", "Oceanus", "others", "Oceanus", "Utopo…

HSCODE Mapping

From the VAST MC2 datanotes we know that Harmonized System code for the shipment can be joined with the hscodes table to get additional details.

The hscode table used in this exercise was extracted from the World Customs Organization website.

The hscode system is used by more than 200 countries, it comprises more than 5000 commodity groups, each identified by a first six digit code. Generally, the first six digits of the HS code are the same in all countries. Different countries, however, may add further digits to detail commodities in more detail without amending the first six digits. In this exercise, as we will be only interested in the fish/seafood products. In hscode system, Fish and crustaceans, molluscs and other aquatic invertebrates etc. start with 301 to 309. Therefore, we will map the hscode to each type of fish products by identify the first 3 digits.

Show the code

MC2_edges_clean_mapped <- MC2_edges_clean %>%
  mutate(fishtype = case_when(
    startsWith(hscode, "301") ~ "live fish",
    startsWith(hscode, "302") ~ "fresh fish",
    startsWith(hscode, "303") ~ "frozen fish",
    startsWith(hscode, "304") ~ "fish meat",
    startsWith(hscode, "305") ~ "processed fish",
    startsWith(hscode, "306") ~ "crustaceans",  #like lobster or shrimps
    startsWith(hscode, "307") ~ "molluscs",  #like Oysters or Abalone
    startsWith(hscode, "308") ~ "aquatic invertebrates", #like Sea cucumbers?
    startsWith(hscode, "309") ~ "seafood flours",  #fish powder, shrimp powder?
    TRUE ~ "not fish"
  ))

Visualization

Visualising Nodes Distribution

Show the code

source_nodes_g <- MC2_nodes_clean %>%
  left_join(MC2_edges_clean_mapped %>% select(source, fishtype), by = c("id" = "source")) %>%
  select(id, fishtype)

source_nodes_g <- source_nodes_g %>%
  count(fishtype) %>%
  arrange(n) %>%
  mutate(fishtype = factor(fishtype, levels = fishtype))

source_nodes_g <- source_nodes_g %>%
  plot_ly(x = ~reorder(fishtype, n), # Reorder the x-axis categories based on count
          y = ~n, 
          type = "bar", 
          color = I('#808de8'), 
          text = ~n, 
          textposition = "auto", 
          hoverinfo = "text", # Set hover information to display text
          texttemplate = "%{y}"
          ) %>%
  layout(
    title = list(text = "Distribution of source nodes by fishtype", x = 0.5),
    yaxis = list(title = "No. of Companies"),
    xaxis = list(title = "fishtype", tickangle = -90),   # Set x-axis title and tick angle
    showlegend = FALSE
  )

source_nodes_g

Upon evaluating the node distribution plot delineated by product type, it becomes clear that entities not related to the fish or seafood industries account for the majority of the data. Followed by ‘fish meat’, ‘crustaceans’, and ‘frozen fish’ in decreasing order.

Given our task is to analyze patterns among fishing companies, it would be prudent to narrow down our data set and filter out irrelevant noise. In this context, companies identified as ‘not fish’ that do not engage in fish or seafood related operations may skew the analysis and should be removed.

Timeline Visualization

Show the code

grp1 <- MC2_edges_clean_mapped %>%
  group_by(year, month, fishtype) %>%
  summarise(no_shnpment = n()) %>%
  filter(fishtype != "not fish") %>%
  ungroup()

tt <- c(paste("Year:", grp1$year, "<br>Month:", grp1$month, "<br>fishtype:", grp1$fishtype, "<br>NoShipment:", grp1$no_shnpment))

fig1 <- grp1 %>%
  mutate(month = factor(month, levels = 1:12, labels = 1:12)) %>%
  ggplot(aes(x = month, y = no_shnpment, fill = fishtype, text = tt)) +
  geom_bar(position = "stack", stat = "identity") +
  scale_fill_viridis(discrete = TRUE) +
  labs(title = "No of shipment per month, 2028 - 2034", x = 'Month', y = 'No of shipment') +
  theme(legend.position = "none") +
  xlab("") +
  scale_x_discrete(labels = 1:12)

fig1 <- ggplotly(fig1, tooltip = "text")

# Create subplot layout
subplot <- subplot(fig1, nrows = 1, shareX = TRUE)

subplot

We examined above plot of the monthly shipment count, organized by fish type (excluding non-fish products), revealed distinct temporal patterns. Notably a dip in shipments was observed in April, followed by a gradually raising trend to a peak in October. This is then succeeded stable trend in November and December, suggesting a stabilization of activity during these months.

Such patterns could be attributed to the majority of fisheries operating in accordance with seasonal cycles, which affect the timing and intensity of their fishing activities. Consequently, our subsequent analyses will refine the focus to those months demonstrating higher fishing activities.

Time series Analysis

Shipment By Product Type

Show the code

grp2 <- MC2_edges_clean_mapped %>%
  group_by(year, month, fishtype) %>%
  summarise(no_shnpment = n()) %>%
  filter(fishtype!="not fish") %>%
  ungroup()

# using ggplot2 for creating the plot with facet_wrap
p <- ggplot(grp2, aes(x = month, y = no_shnpment, color = fishtype)) +
  geom_line(aes(group = fishtype), line = list(shape = "spline", smoothing = 0.2)) +
  geom_point() +
  labs(title = "Total shipment per month by fish type, 2028 - 2034", 
       x = "Month", 
       y = "Num of Shnp") +
  facet_wrap(~year, nrow = 1) + 
  theme(legend.position = "bottom") +
  scale_x_continuous(breaks = 1:12, labels = 1:12)  # Set x-axis breaks and labels

# converting ggplot2 object to plotly object
fig2 <- ggplotly(p)

# print the plot
fig2

Utilizing data specific to fish and seafood-related shipments, we created a time series plot to study the temporal trends. We can observe the number of shipments reached a significant peak in the year 2033. This spike was particularly seen in the months of July, August, and September for the top four fish products: ‘fish meat’, ‘crustaceans’, ‘frozen fish’, and ‘molluscs’.

This observation could be leveraged in our subsequent network graph analysis. By reducing the node count to concentrate on shipments pertaining to these top four fish products during these specific months in 2033, we could streamline our analysis.

Prepare for Edges

Show the code

MC2_edges_aggregated <- MC2_edges_clean_mapped %>%
  filter(year %in% c(2033)) %>%
  filter(month %in% c(7, 8, 9)) %>%
  filter(fishtype != "not fish") %>%
  group_by(source, target, fishtype) %>%
  summarise(Weight = n()) %>%
  filter(source != target) %>%
  filter(Weight > 20) %>%
  ungroup()

Prepare for Nodes

Show the code

# Filter rows in nodes based on matching ids in edges target and source

id1 <- MC2_edges_aggregated %>%
  select(source) %>%
  rename(id = source)
id2 <- MC2_edges_aggregated %>%
  select(target) %>%
  rename(id = target)

MC2_nodes_extracted <- rbind(id1, id2)  %>%
  distinct()


MC2_nodes_extracted <- MC2_nodes_extracted %>%
  left_join(MC2_edges_aggregated %>% select(target, fishtype), by = c("id" = "target")) %>%
  select(id, fishtype)

# Let's add a column with the group of each name. It will be useful later to color points
MC2_nodes_extracted$group <- MC2_nodes_extracted$fishtype

Creating the Graph Dataframe

Show the code

MC2_graph <- tbl_graph(nodes = MC2_nodes_extracted,
                        edges = MC2_edges_aggregated, 
                        directed = TRUE) 


MC2_graph

# A tbl_graph: 274 nodes and 164 edges
#
# A directed acyclic simple graph with 134 components
#
# A tibble: 274 × 3
  id                                              fishtype group
  <chr>                                           <chr>    <chr>
1 2 Wharf S.A. de C.V. Delivery                   <NA>     <NA> 
2 Adriatic Catch Ltd. Liability Co Transportation <NA>     <NA> 
3 Adriatic Tuna AS Solutions                      <NA>     <NA> 
4 Agua Limited Liability Company Services         <NA>     <NA> 
5 Andhra Pradesh   Sextant Oyj Forwading          <NA>     <NA> 
6 Angeline Sea NV Worldwide                       <NA>     <NA> 
# ℹ 268 more rows
#
# A tibble: 164 × 4
   from    to fishtype    Weight
  <int> <int> <chr>        <int>
1     1   112 fish meat       27
2     2   115 crustaceans     24
3     3   122 crustaceans     36
# ℹ 161 more rows

Show the code

set_graph_style()

g <- ggraph(MC2_graph, 
            layout = "fr") + 
  geom_edge_link(aes(width=Weight), 
                 alpha=0.2) +
  scale_edge_width(range = c(0.1, 5)) +
  geom_node_point(aes(colour = group), 
                  size = 2)
  
g + facet_nodes(~group)+
  th_foreground(foreground = "grey80",  
                border = TRUE) +
  theme(legend.position = 'bottom')

Plot Network Graph

Plotting of Network using Eigenvector Centrality

Degree centrality only takes into account the number of edges for each node, but it leaves out information about ego’s alters.

However, we might think that power comes from being tied to powerful people. If A and B have the same degree centrality, but A is tied to all high degree people and B is tied to all low degree people, then intuitively we want to see A with a higher score than B.

Eigenvector centrality takes into account alters’ power. In our Network graph, Eigenvector score for each node will be calculated using igraph package.

Due to the large number of edges within the interaction, we will focus only the top 20% nodes based on their eigenvector centrality score.

Note

Note: This part of codes cites from Senior Mr Jordan Ong’s Take-Home_Exercise 6 in 2021-2022 April Term).

Show the code

quantile_graph <- quantile(eigen_centrality(MC2_graph)$vector,
         probs = seq(0, 1, 1/5)
         )
V(MC2_graph)$size = eigen_centrality(MC2_graph)$vector

MC2_graph_aggregated <- delete_vertices(MC2_graph, V(MC2_graph)[size < quantile_graph[5]])


set.seed (1234)
layout1 <- layout_with_fr(MC2_graph_aggregated)

quantile_graph_aggregated <- quantile(V(MC2_graph_aggregated)$size, #identify top 20% of the new vertices
         probs = seq(0, 1, 1/5)
         )


V(MC2_graph_aggregated)$color <- ifelse (V(MC2_graph_aggregated)$size > quantile_graph_aggregated[5], "darkgoldenrod3", "azure3") #color yellow if vertices is top 20%
E(MC2_graph_aggregated)$color <- "grey"
V(MC2_graph_aggregated)$size <- V(MC2_graph_aggregated)$size/0.065 
#Increase the size of nodes based on their centrality score, only those with high score will be visible

V(MC2_graph_aggregated)$id <- ifelse (V(MC2_graph_aggregated)$size*0.065 > quantile_graph_aggregated[5],V(MC2_graph_aggregated)$id,NA)
#label the vertices if vertices belongs to the top 20%


plot(MC2_graph_aggregated, edge.arrow.size = 0.25, edge.arrow.mode = "-", 
     vertex.label = V(MC2_graph_aggregated)$id, vertex.label.cex = 0.65, 
     vertex.label.font = 1, main = "Which companies are having  to other nodes?")

A higher number of links could potentially indicate a higher level of activity, collaborations, or partnerships, which might increase the likelihood of a company undergoing name changes.

Show the code

vertex_attr(MC2_graph_aggregated, index = V(MC2_graph_aggregated)$size*0.065 > quantile_graph_aggregated[5])

$id
 [1] "Balkan Cat ОАО Transport"            
 [2] "Blue Horizon Family &"               
 [3] "Sea Breezes S.A. de C.V. Freight "   
 [4] "Yenisei  Eel GmbH & Co. KG Services" 
 [5] "Caracola del Sol Services"           
 [6] "Costa de la Felicidad Shipping"      
 [7] "Orange River   Incorporated Shipping"
 [8] "Selous Game Reserve  S.A. de C.V."   
 [9] "hǎi dǎn Corporation Wharf"           
[10] "Mar del Este CJSC"                   
[11] "Pao gan SE Seal"                     

$fishtype
 [1] NA            NA            NA            NA            "crustaceans"
 [6] "crustaceans" "crustaceans" "crustaceans" "crustaceans" "crustaceans"
[11] "crustaceans"

$group
 [1] NA            NA            NA            NA            "crustaceans"
 [6] "crustaceans" "crustaceans" "crustaceans" "crustaceans" "crustaceans"
[11] "crustaceans"

$size
 [1]  5.481011  8.338480 13.154593  5.523848 10.253177  7.721414  5.481894
 [8]  6.076499  5.462136 15.384615  7.830756

$color
 [1] "darkgoldenrod3" "darkgoldenrod3" "darkgoldenrod3" "darkgoldenrod3"
 [5] "darkgoldenrod3" "darkgoldenrod3" "darkgoldenrod3" "darkgoldenrod3"
 [9] "darkgoldenrod3" "darkgoldenrod3" "darkgoldenrod3"

Through the nodes attributes of the top 20% of Eigenvector Centrality Score, we see that majority of nodes are having the same product type which is crustaceans. We will now analyse to see if the product type affect the Eigen betweenness score. A non-parametric test is conducted to provided statistical evidence whether to reject or accept the null hypothesis.

Hypothesis testing: Does types of fishing affect eigen betweenness score?

Show the code

MC2_edges_aggregated_r <- MC2_edges_clean_mapped %>%
  #filter(year %in% c(2028)) %>%
  filter(fishtype !="not fish") %>%
  group_by(source, target, fishtype) %>%
  summarise(Weight = n()) %>%
  filter(source != target) %>%
  filter(Weight > 1) %>%
  ungroup()

# Filter rows in nodes based on matching ids in edges target and source

id3 <- MC2_edges_aggregated_r %>%
  select(source) %>%
  rename(id = source)
id4 <- MC2_edges_aggregated_r %>%
  select(target) %>%
  rename(id = target)

MC2_nodes_extracted_r <- rbind(id3, id4)  %>%
  distinct()


MC2_nodes_extracted_r <- MC2_nodes_extracted_r %>%
  left_join(MC2_edges_aggregated_r %>% select(target, fishtype), by = c("id" = "target")) %>%
  select(id, fishtype)


MC2_graph_r <- tbl_graph(nodes = MC2_nodes_extracted_r,
                        edges = MC2_edges_aggregated_r, 
                        directed = TRUE) 

MC2_graph_r

# A tbl_graph: 33753 nodes and 30273 edges
#
# A directed multigraph with 26526 components
#
# A tibble: 33,753 × 2
  id                                   fishtype
  <chr>                                <chr>   
1 " Direct Herring Company Transit"    <NA>    
2 " Direct LLC Marine biology"         <NA>    
3 " Direct Shark Oyj Marine sanctuary" <NA>    
4 "-1515"                              molluscs
5 "-1515"                              molluscs
6 "-28"                                <NA>    
# ℹ 33,747 more rows
#
# A tibble: 30,273 × 4
   from    to fishtype    Weight
  <int> <int> <chr>        <int>
1     1 16469 molluscs         5
2     1 16470 molluscs         6
3     1  2703 crustaceans      6
# ℹ 30,270 more rows

Show the code

V(MC2_graph_r)$size = eigen_centrality(MC2_graph_r)$vector


MC2_graph_analysis_r <- as.data.frame(MC2_graph_r)

p1 <- ggbetweenstats(
  data = MC2_graph_analysis_r,
  x = fishtype,
  y = size,
  xlab = "fishtype",
  ylab = "EV Centrality \nScore",
  title = "Will betweenness score be affected by fishing type?",
  type = "np", #conduct non-parametric test
  conf.level = 0.95,
  mean.ci = TRUE,
  package = "ggsci",
  palette = "default_jco"
) +
  ggplot2::theme(
    axis.title.y = element_text(angle = 0, size = 9),
    axis.title.x = element_text(size = 9),
    plot.title = element_text(color = "dimgrey", size = 12, hjust = 0.5)
)

p1

From the statistical plot we have P-values is less than 0.05, it indicates that there is a statistically significant difference in the eigen betweenness scores across different fishing types.

Therefore, we can conclude that the fishing type has a significant effect on the eigen betweenness scores in the network. This finding implies that the type of fishing has an influence on the structural importance or centrality of the nodes in the network, as measured by eigen betweenness.

VisNetwork Graph

The community detection is performed using the cluster_edge_betweenness function from the igraph package. This function calculates the edge betweenness centrality for each edge in the graph and then uses it to detect communities.

Show the code

edges_df1 <- MC2_graph %>%
  activate(edges) %>%
  as.tibble()

nodes_df1 <- MC2_graph %>%
  activate(nodes) %>%
  as.tibble() %>%
  rename(label = id) %>%
  mutate(id=row_number()) %>%
  select(id, label)

# Perform community detection using the cluster edge betweenness
communities <- cluster_edge_betweenness(MC2_graph)

# Get the cluster membership of each node
membership <- membership(communities)

# Add the cluster membership information to the nodes data frame
nodes_df1$group <- membership

# Plot the graph with clustered nodes using visNetwork
visNetwork(nodes_df1, edges_df1) %>%
  visIgraphLayout(layout = "layout_with_fr") %>%
  visEdges(arrows = "to",
           smooth = list(enabled = TRUE,
                         type = "curvedCW"), 
           color = list(highlight = "lightgray")) %>%
  visOptions(highlightNearest = list(enabled = TRUE,
                                     degree = 1,
                                     hover = TRUE,
                                     labelOnly = TRUE),
             nodesIdSelection = TRUE,
             selectedBy = "group") %>%
  visLayout(randomSeed = 1234)

cluster_edge_betweenness

the large number of nodes forming a single largest community suggests that the graph has strong connections and relatively low edge betweenness centrality across its edges.
This indicates a cohesive and densely connected network where most nodes are closely linked, resulting in a single dominant community.

cluster_louvain is a community detection algorithm implemented in the igraph package in R. It is based on the Louvain method, which is a popular and efficient algorithm for detecting communities in networks. The Louvain method aims to optimize a quality function known as modularity, which measures the strength of division of a network into communities.

Show the code

edges_df2 <- MC2_graph %>%
  activate(edges) %>%
  as.tibble()

nodes_df2 <- MC2_graph %>%
  activate(nodes) %>%
  as.tibble() %>%
  rename(label = id) %>%
  mutate(id=row_number()) %>%
  select(id, label)

# Convert the graph to undirected
MC2_graph_undirected <- as.undirected(MC2_graph)

# Perform community detection using the Louvain algorithm on the undirected graph
communities <- cluster_louvain(MC2_graph_undirected)

# Get the cluster membership of each node
membership <- membership(communities)

# Add the cluster membership information to the nodes data frame
nodes_df2$group <- membership

# Plot the graph with clustered nodes using visNetwork
visNetwork(nodes_df2, edges_df2) %>%
  visIgraphLayout(layout = "layout_with_fr") %>%
  visEdges(arrows = "to",
           smooth = list(enabled = TRUE, type = "curvedCW"), 
           color = list(highlight = "lightgray")) %>%
  visOptions(highlightNearest = list(enabled = TRUE, degree = 1, hover = TRUE, labelOnly = TRUE),
             nodesIdSelection = TRUE,
             selectedBy = "group") %>%
  visLayout(randomSeed = 1234)

cluster_louvain

In the cluster_louvain plot, it presences 8 small clusters suggests that the algorithm has identified distinct communities within the graph based on the optimization of modularity. These smaller clusters indicate subsets of nodes that are more densely connected internally and have fewer connections between them compared to the cluster_edge_betweenness result.

In our case, we would like to identify a clearer communities that potentially represent separate business cluster, cluster_louvain can be good choice. As it tends to produce well-separated communities with higher modularity score so I can capture smaller but tightly-knit communities within a larger network.