Benchmarking"
In akc: Automatic Knowledge Classification

readRDS("benchmark_res.rds") -> res
knitr::opts_chunk$set(
  collapse = TRUE,
  eval = FALSE,
  comment = "#>"
)

Based on community detection to automatically classify the keywords, \CRANpkg{akc} can utilize different algorithms for clustering. In this vignette, a benchmark is provided to show the difference for various algorithms on multiple sizes of networks.

First, we'll load the needed packages.

library(akc)
library(dplyr)

Then, we prepare the needed data. The built-in data table biblio_data_table would be used here.

bibli_data_table %>% 
  keyword_clean() %>% 
  keyword_merge() -> clean_data

Next, a combination of network size and community detection algorithms are designed to be tested:

100:300 -> topn_sample
ls("package:akc") %>% 
  str_extract("^group.+") %>% 
  na.omit() %>% 
  setdiff(c("group_biconnected_component",
            "group_components",
            "group_optimal")) -> com_detect_fun_list

Finally, we'll implement the computation and record the results.

all = tibble()
for(i in com_detect_fun_list){
    for(j in topn_sample){
      system.time({
        clean_data %>% 
          keyword_group(top = j,com_detect_fun = get(i)) %>% 
          as_tibble -> grouped_network_table
      }) %>% na.omit-> time_info
      grouped_network_table %>% nrow -> node_no
      grouped_network_table %>% distinct(group) %>% nrow -> group_no
      grouped_network_table %>% 
        count(group) %>% 
        summarise(mean(n)) %>% 
        .[[1]] -> group_avg_node_no
      grouped_network_table %>% 
        count(group) %>% 
        summarise(sd(n)) %>% 
        .[[1]] -> group_sd_node_no
      c(com_detect_fun = i, 
        topn = j,
        node_no = node_no,group_no = group_no,
        avg = group_avg_node_no,
        sd = group_sd_node_no,time_info[1:3]) %>% 
        bind_rows(all,.) -> all
    }
}

res = all %>% 
  mutate_at(2:9,function(x) as.numeric(x) %>% round(2)) %>% 
  distinct(com_detect_fun,node_no,.keep_all = T) %>% 
  select(-topn,-contains("self")) %>% 
  setNames(c("com_detect_fun","No. of total nodes","No. of total groups",
             "Average node number in each group","Standard deviation of node number",
             "Computer running time for keyword_group function"))