Abstract
This process performs all the preprocessing and wrangling of the necessary data-sets to run the occupation identification algorithm. The frequency table and the multilingual mappings as they are encoded in the ISCO-ESCO classification are used. The data are cleaned, transformed and weighted. The more common a word is the less weight it gets reflecting the fact that commonly used words encode less information. The weight is also analogous to the frequency of an ESCO code measured in the given CVs.# http://rmarkdown.rstudio.com/html_document_format.htm
sourceTimeNeeded <- c(0);
source.starting.time <- proc.time()[3]Information about the libraries, environment, sources used and their execution is reported. Aditional information is provided within section tabs. Navigating through the report is also possible through the table of contents. Tables reported, can be dynamically filtered, searched ordered and exported into various formats.
librariesVersion <- c()
for(i in 1:length(libraries))
librariesVersion <- c(librariesVersion, paste(packageVersion(libraries[i] )))
librariesLoaded <- lapply(libraries, require, character.only = TRUE) ## Loading required package: dplyr##
## Attaching package: 'dplyr'## The following objects are masked from 'package:data.table':
##
## between, first, last## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union## Loading required package: DT## Loading required package: jsonlite## Loading required package: text2vec## Loading required package: stopwords## Loading required package: parallel../000.core/00.01.libraries.R completed in 1.98 seconds
sourceTimeNeeded <- c( sourceTimeNeeded, timeNeeded)
source.starting.time <- proc.time()[3]
## Base functions
# EPAS-DS
# @authors ds@eworx.gr
repository <- "/data/generic/"
getSourcePath <- function(filename, baseFolder = repository){
return(paste(baseFolder, filename, sep = ""))
}
readData <- function(filename, colClasses = c(), baseFolder = repository, header = TRUE, sep = "\t", encoding = "UTF-8", stringsAsFactors = TRUE, na.strings = c("", "NULL"), verbose = FALSE){
if(length(colClasses) == 0)
return (data.table::fread(input = getSourcePath(filename, baseFolder), header = header, sep = sep, encoding = encoding, stringsAsFactors = stringsAsFactors, verbose = verbose, showProgress = TRUE, na.strings = na.strings ) )
return (data.table::fread(input = getSourcePath(filename, baseFolder), colClasses = colClasses, header = header, sep = sep, encoding = encoding, verbose = verbose, showProgress = TRUE, na.strings = na.strings ) )
}
#rds for small disk space & fst for fast load
saveBinary <- function(data, filename = filename, baseFolder = repository, format = "rds"){
fileName <- getSourcePath(filename, baseFolder)
dir.create(dirname(fileName), recursive = TRUE, showWarnings = FALSE)
if(format == "rds") saveRDS(data, fileName)
if(format == "fst") fst::write_fst(data, fileName)
}
#alternative for rough read write operations
saveRDS_ <- function(object, file){
dir.create(dirname(file), recursive = TRUE, showWarnings = FALSE)
saveRDS(object, file)
}
#rds for small disk space & fst for fast load
loadBinary <- function(filename, baseFolder = repository, format = "rds", as.data.table = TRUE){
if(format == "rds"){return(readRDS(getSourcePath(filename, baseFolder)))}
if(format == "fst"){return(fst::read_fst(getSourcePath(filename, baseFolder), as.data.table = as.data.table))}
}
rowColumns <- function(data){
return(paste( format(nrow(data), big.mark=","), "Rows X ", ncol(data), "Columns"))
}
publishIncludeCss <- function(){
sourceFile <- "/data/jobs/wp41.analysis/000.core/include.css"
destinatinoFile <- "/data/tmpfs/results/include.css"
if (!file.exists(destinatinoFile)) {
return (file.copy(sourceFile, destinatinoFile))
}else{
return(TRUE);
}
}
#as the mountstorage is on memory make sure the asset include.css is there.
summariseTable <- function(data){
return(data.frame(unclass(summary(data)), check.names = FALSE, stringsAsFactors = FALSE))
#return(do.call(cbind, lapply(data, summary)))
}
factoriseCharacterColumns <- function(data){
for(name in names(data)){
if( class(data[[name]]) =="character"){
data[[name]] <- as.factor(data[[name]])
}
}
return(data)
}
############################
# https://rstudio.github.io/DT/010-style.html
#https://rpubs.com/marschmi/RMarkdown
capitalise <- function(x) paste0(toupper(substring(x, 1, 1)), substring(x, 2, nchar(x)))
styliseDTNumericalColumn <- function(data, result, columnName, color, columnsName_original ){
if(columnName%in% columnsName_original){
result <- result %>% formatStyle(
columnName,
background = styleColorBar(data[[columnName]], color),
backgroundSize = '100% 90%',
backgroundRepeat = 'no-repeat',
backgroundPosition = 'center'
)
}
return(result)
}
reportTabularData <- function(data, anonymize=TRUE){
if(anonymize)return()
columnsName <- names(data)
columnsName <- lapply(columnsName, capitalise)
columnsName_original <- names(data)
result <-
DT::datatable(
data,
class = 'cell-border stripe',
filter = 'top',
rownames = FALSE,
colnames = columnsName,
extensions = 'Buttons',
options = list(
pageLength = 20,
columnDefs = list(list(className = 'dt-left', targets = "_all")),
dom = 'Bfrtip',
buttons = c('copy', 'csv', 'excel', 'pdf'),
searchHighlight = TRUE,
initComplete = JS(
"function(settings, json) {",
"$(this.api().table().header()).css({'border': '1px solid'});",
"}"
)
)
)
result <- styliseDTNumericalColumn(data,result, "Count", 'steelblue', columnsName_original)
result <- styliseDTNumericalColumn(data,result, "sourceTimeNeeded", '#808080', columnsName_original)
result <- styliseDTNumericalColumn(data,result, "timeNeeded", '#808080', columnsName_original)
#result <- styliseDTNumericalColumn(data,result, "percentMatch", '#5fba7d', columnsName_original)
result <- styliseDTNumericalColumn(data,result, "percentMatch", '#4682b4', columnsName_original)
return(result)
}
fonts <- list(
sans = "DejaVu Serif",
mono = "DejaVu Serif",
`Times New Roman` = "DejaVu Serif"
)
#read_xml_to_list <- function(filepath, is.gz = FALSE){
# if(is.gz){
# temp_data <- paste0(repository, "data/delete.me")
# result <- xmlToList(xmlParse(gunzip(filepath, destname = temp_data, remove =FALSE)))
# Sys.chmod(file.path(temp_data), "777", use_umask = FALSE)
# unlink(temp_data)
# result
# }else{
# xmlToList(xmlParse(filepath))
# }
#}
#transpose_list_to_dt <- function(data_list){
# dt <- t(as.data.table(data_list))
# dt <- as.data.table(dt)
# dt[, (names(dt)) := lapply(.SD, unlist), .SDcols = 1:ncol(dt)]
# dt[, (names(dt)) := lapply(.SD, unlist), .SDcols = 1:ncol(dt)]
# names(dt) <- names(data_list[[1]])
# dt
#}
cleansingCorpus <- function(
htmlString, rem.html =TRUE, rem.http = TRUE, rem.newline = TRUE,
rem.nonalphanum = TRUE, rem.longwords = TRUE, rem.space = TRUE,
tolower = TRUE, add.space.to.numbers = TRUE, rem.country.begin = FALSE,
rem.nonalphanum.begin = FALSE, rem.space.begin = FALSE
){
if(rem.html){text <- gsub("<.*?>", " ", htmlString)} # removing html commands
if(rem.http){text <- gsub(" ?(f|ht)tp(s?)://(.*)[.][a-z]+", " ", text)} #removing http destinations
if(rem.newline){text <- gsub("[\r\n\t]", " ", text)}
if(rem.nonalphanum){text <- gsub("[^[:alpha:]]", " ", text)} #removing non-alphanumeric
if(rem.longwords){text <- gsub("\\w{35,}", " ", text)} ##Removing words with more than 30 letters
if(rem.space){text <- gsub("\\s+", " ", text)} #removing excess space
if(tolower){text <- tolower(text)}
if(add.space.to.numbers){ #add space between number and letters
text <- gsub("([0-9])([[:alpha:]])", "\\1 \\2", text)
text <- gsub("([[:alpha:]]|[.])([0-9])", "\\1 \\2", text)
}
if(rem.space.begin){text <- gsub("^[[:space:]]*", "", text)}
if(rem.country.begin){text <- gsub("^EU", "", text)} #remove country codes from the beginning of the text
if(rem.nonalphanum.begin){text <- gsub("^[?–-]*", "", text)} #remove special characters identified in the beginning of text
if(rem.space.begin){text <- gsub("^[[:space:]]*", "", text)}
trimws(text)
}
#This function removes dates that are "relics" from the xml parsing
removeDates <- function(text){
days <- "(Sunday,|Monday,|Tuesday,|Wednesday,|Thursday,|Friday,|Saturday,)"
months <- "(January|February|March|April|May|June|July|August|September|October|November|December|Months)"
date_form1 <- paste(days, months, "([0-9]|[0-9][0-9]), [0-9][0-9][0-9][0-9]")
date_form2 <- "\\?[0-9][0-9][0-9][0-9]"
text <- gsub(date_form1, " ", text)
gsub(date_form2, " ", text)
}
xmlToDataTable <- function(xmlData, itemNames){
itemList <- lapply(itemNames,
function(x){
xml_text(xml_find_all(xmlData, paste0(".//item/", x)))
}
)
names(itemList) <- xmlItems
as.data.table(itemList)
}
cleanCorpusHtml <- function(text){
unlist(lapply(text, function(x){
if(nchar(x) > 0){
# because nodes were starting with tag keywords in li, we relocate at the end so the information remains and the description
# starts with the main content
html <- gsub(">","> ", x) # add spaces after html tags so these aren't concatenated
xml <- read_xml(html, as_html = TRUE)
lis <- xml_find_all(xml, ".//li")
xml_remove(lis)
text <- paste( paste(xml_text(xml), collapse ="") , paste(xml_text(lis) , collapse =""), collapse ="")
text <- gsub("\\s+"," ", text)
}else {""}
}))
}
#Split equally a vector into chunks of number n_chunks
equal_split <- function(vct, n_chunks) {
lim <- length(vct)
fstep <- lim%/%n_chunks
idx_list <- list()
for(i in seq(n_chunks - 1)){
idx_list[[i]] <- vct[((i-1)*fstep + 1):(i*(fstep))]
}
idx_list[[n_chunks]] <- vct[((n_chunks - 1)*fstep + 1):(lim)]
return(idx_list)
}
#Function that takes a vector, and returns thresholded first 10 sorted indexes
getThresholdOrderRwmd <- function(vct, idVec, threshold = 1e-6, numHead = 10){
vct <- ifelse(vct > threshold, vct, Inf)
indexVec <- head(order(vct), numHead)
idVec[indexVec]
}
#Function to read xml nodes in description
maintainElements <- function(nodes, elementType = "a", attribute = "href"){
xml_attr(xml_find_all(nodes, paste0(".//", elementType)), attribute)
}
#Function to add results to datatable
elementsToDataTable <- function(result, elementType){
if(length(result) > 0)
data.table(elementType = elementType, attributeValue = result)
else
data.table()
}
#Function to retrieve urls from text
keepHtmlElements <- function(feedItem){
nodes <- read_xml(paste0("<div>", feedItem, "</div>"), as_html = TRUE)
rbind(
elementsToDataTable(maintainElements(nodes, "a", "href"), "link"),
elementsToDataTable(maintainElements(nodes, "img", "src"), "image"),
elementsToDataTable(maintainElements(nodes, "img-src", "src"), "image")
#All "img-src" are NA
)
}
#retrieve list of parameters in a http request query
getQueryParams <- function(url){
query <- httr::parse_url(url)$query
queryValues <- unlist(query)
queryNames <- names(query)
dat <- data.table(varName = queryNames, value = queryValues)
dat[queryValues != ""]
}
keepCountryName <- function(string){
string <- gsub(".*_", "", string)
gsub("\\..*", "", string)
}
keepNTokens <- function(string, num){
tokenList <- strsplit(string, split = " ")
sapply(tokenList, function(tokens){
tokens <- sort(tokens)
tokensShift <- shift(tokens, -num, fill = FALSE)
paste(tokens[tokens != tokensShift], collapse = " ")
})
}
findTFIDF <- function(corpus, stopwords, normalize = "double", min_char = 1) {
tokensList <- strsplit(corpus[, text], " ")
names(tokensList) <- corpus[, code]
tokensDT <- lapply(tokensList, as.data.table) %>%
rbindlist(idcol = TRUE) %>%
setnames(c("class", "term"))
tokensDT <- tokensDT[!term %in% stopwords][nchar(term) > min_char]
#inverse document frequency smooth
idfDT <- tokensDT[!duplicated(tokensDT)][, .(docFreq = .N), by = "term"]
idfDT[, idf := log(length(unique(tokensDT$class)) / (docFreq + 1)) + 1]
tfDT <- tokensDT[, .(term_count = .N), by = c("class", "term")]
if(normalize == "double")tfDT[, tf := 0.5 + 0.5 * term_count / max(term_count), by = "class"]
if(normalize == "log")tfDT[, tf := log(1 +term_count)]
merge(tfDT, idfDT, on = "term")[, tfIdf := tf*idf ][, .(term, class, tfIdf)]
}
tidyJsonData <- function(jsonList){
if(length(jsonList) == 0)return(NULL)
unlistOccupations <- jsonList %>% unlist
codesMaleFemale <- names(unlistOccupations)
epasMapping <- data.table(unlistOccupations)
epasMapping[ , code := gsub("\\.[[:alpha:]]$", "", codesMaleFemale)]
uniqueMappingBoolean <- epasMapping[ , unlistOccupations != c(unlistOccupations[-1], F), by = code]$V1
codesEpasDB <- epasMapping[uniqueMappingBoolean]
names(codesEpasDB) <- c("title", "code")
codesEpasDB[ , title := cleansingCorpus(title)]
}
############################################################################################################# R version 4.0.5 (2021-03-31)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.2 LTS
##
## Matrix products: default
## BLAS/LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.8.so
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=C
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] parallel stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] stopwords_2.2 text2vec_0.6 jsonlite_1.7.2 DT_0.18
## [5] dplyr_1.0.6 rmarkdown_2.8 data.table_1.14.0
##
## loaded via a namespace (and not attached):
## [1] Rcpp_1.0.6 pillar_1.6.0 bslib_0.2.5
## [4] compiler_4.0.5 jquerylib_0.1.4 highr_0.9
## [7] tools_4.0.5 digest_0.6.27 evaluate_0.14
## [10] lifecycle_1.0.0 tibble_3.1.1 lattice_0.20-41
## [13] pkgconfig_2.0.3 rlang_0.4.11 Matrix_1.3-2
## [16] mlapi_0.1.0 RhpcBLASctl_0.20-137 yaml_2.2.1
## [19] xfun_0.23 stringr_1.4.0 knitr_1.33
## [22] generics_0.1.0 vctrs_0.3.8 sass_0.4.0
## [25] htmlwidgets_1.5.3 grid_4.0.5 tidyselect_1.1.1
## [28] glue_1.4.2 R6_2.5.0 fansi_0.4.2
## [31] lgr_0.4.2 purrr_0.3.4 magrittr_2.0.1
## [34] ellipsis_0.3.2 htmltools_0.5.1.1 float_0.2-4
## [37] rsparse_0.4.0 utf8_1.2.1 stringi_1.6.2
## [40] crayon_1.4.1../000.core/00.02.base.functions.R completed in 0.1 seconds
jsonRepoIsco <- getSourcePath("input/iscoLikeOccupations")
fileNames <- list.files(jsonRepoIsco, full.names = TRUE)
jsonCountriesIsco <- keepCountryName(fileNames)allJsonListIsco <- lapply(fileNames, fromJSON, flatten = FALSE)
jsonIscoCodes <- allJsonListIsco[jsonCountriesIsco == "en"]Type of data: list.
Number of elements: 13
escoRepo <- getSourcePath("input/esco_isco_bundle/esco")
fileNames <- list.files(escoRepo, full.names = TRUE)
escoCountries <- keepCountryName(fileNames)Type of data: list.
Number of elements: 28
fileName <- getSourcePath("input/esco_isco_bundle/isco/ISCOGroups_en.csv" )
iscoDT <- fread(fileName)Type of data: data.table, data.frame.
Number of elements: 619, 7
Type of data: data.table, data.frame.
Number of elements: 1844332, 7
01.00.load.data.R completed in 31.04 seconds
The process uses the ESCO classification and the alternative labels to calculate multilingual lookup tables as well as weighted mappings between tokens and ESCO codes.
no_cores <- detectCores() - 1
cl <- makeCluster(no_cores, type = "FORK")
allEscoTidyList <- parLapply(cl, allEscoList, function(data){
data[, code := gsub(".*/", "", conceptUri )]
data[, preferredLabel := trimws(gsub("\\n", " ", preferredLabel))]
data[, altLabels := trimws(gsub("\\n", " ", altLabels)) %>% keepNTokens(2)]
data[, description := trimws(gsub("\\n", " ", description)) %>% keepNTokens(2)]
data[is.na(altLabels), altLabels := ""]
data[is.na(description), description := ""]
data[, text := paste(preferredLabel, altLabels, description) %>% cleansingCorpus]
data[, .(text, code)]
})
stopCluster(cl)Tokenizing descriptions, removing stopwords and mapping each word to a token.
stopwordsLang <- c(stopwords_getlanguages("snowball"), stopwords_getlanguages("misc"))
escoCountries <- names(allEscoTidyList)
validStopwords <- escoCountries[escoCountries %in% stopwordsLang]
no_cores <- detectCores() - 1
cl <- makeCluster(no_cores, type = "FORK")
weightedTokensList <- parLapply(cl, seq_along(allEscoTidyList), function(i){
escoOccupations <- allEscoTidyList[[i]]
lang <- names(allEscoTidyList)[i]
stopWords <- NULL
if(lang %in% validStopwords)stopWords <- stopwords(lang)
findTFIDF(escoOccupations, stopWords) %>%
setNames( c("word", "code", "word_weight"))
})
stopCluster(cl)
names(weightedTokensList) <- names(allEscoTidyList)sortedVocabularyList <- lapply(weightedTokensList, function(x){
x[!duplicated(x$word)][order(-word_weight)][ , word]
})
names(sortedVocabularyList) <- names(allEscoTidyList)02.00.process.data.R completed in 39.74 seconds
regEsco <- "^[A-z0-9]{8}-[A-z0-9]{4}-"
escoCodesCounts <- workExpDT[grepl(regEsco, code)][, .(count = .N), by = "code"][order(-count)]iscoCodesCounts <- workExpDT[nchar(code) == 5][,.(count = .N), by = "code"][order(-count)]
iscoLookup <- jsonIscoCodes %>% tidyJsonData
iscoTextCount <- iscoLookup[iscoCodesCounts, on = "code"][!is.na(title)]
setnames(iscoTextCount, "title", "text")freeText <- workExpDT[is.na(code)][!is.na(label)][, .(id, locale, label, index)] %>% unique
freeText[, label := cleansingCorpus(label)]03.00.process.data.R completed in 43.37 seconds
outputRepo <- getSourcePath("jobsOutput/workExperience/")
filename <- "weightedTokensListEscoEpas.rds"
saveBinary(weightedTokensList, filename, outputRepo)Datasource : /data/generic/jobsOutput/workExperience/weightedTokensListEscoEpas.rds of 48,528,142 bytes.
filename <- "sortedVocabularyListEscoEpas.rds"
saveBinary(sortedVocabularyList, filename, outputRepo)Datasource : /data/generic/jobsOutput/workExperience/sortedVocabularyListEscoEpas.rds of 2,754,990 bytes.
Datasource : /data/generic/jobsOutput/workExperience/freeText.rds of 45,378,195 bytes.
Datasource : /data/generic/jobsOutput/workExperience/escoCodesCounts.rds of 62,790 bytes.
Datasource : /data/generic/jobsOutput/workExperience/processedWorkExp.fst of 233,815,823 bytes.
04.00.save.data.R completed in 22.32 seconds
Completed in 138.56 seconds.