You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
mamlr/R/dupe_detect.R

71 lines
4.2 KiB

#' Get ids of duplicate documents that have a cosine similarity score higher than [threshold]
#'
#' Get ids of duplicate documents that have a cosine similarity score higher than [threshold]
#' @param row Row of grid to parse
#' @param grid A cross-table of all possible combinations of doctypes and dates
#' @param cutoff_lower Cutoff value for minimum cosine similarity above which documents are considered duplicates (inclusive)
#' @param cutoff_upper Cutoff value for maximum cosine similarity, above which documents are not considered duplicates (for debugging and manual parameter tuning, inclusive)
#' @param es_pwd Password for Elasticsearch read access
#' @param es_super Password for write access to ElasticSearch
#' @param words Document cutoff point in number of words. Documents are cut off at the last [.?!] before the cutoff (so document will be a little shorter than [words])
#' @param localhost Defaults to true. When true, connect to a local Elasticsearch instance on the default port (9200)
#' @return dupe_objects.json and data frame containing each id and all its duplicates. remove_ids.txt and character vector with list of ids to be removed. Files are in current working directory
#' @export
#' @examples
#' dupe_detect(1,grid,cutoff_lower, cutoff_upper = 1, es_pwd, es_super, words, localhost = T)
#################################################################################################
#################################### Duplicate detector ################################
#################################################################################################
dupe_detect <- function(row, grid, cutoff_lower, cutoff_upper = 1, es_pwd, es_super, words, localhost = T) {
params <- grid[row,]
print(paste0('Parsing ',params$doctypes,' on ',params$dates ))
query <- paste0('{"query":
{"bool": {"filter":[{"term":{"doctype": "',params$doctypes,'"}},
{"range" : {
"publication_date" : {
"gte" : "',params$dates,'T00:00:00Z",
"lt" : "',params$dates+1,'T00:00:00Z"
}
}}]
} } }')
out <- elasticizer(query, es_pwd = es_pwd, localhost= localhost)
if (class(out$hits$hits) != 'list') {
dfm <- dfm_gen(out, text = "full", words = words)
if (sum(dfm[1,]) > 0) {
simil <- as.matrix(textstat_simil(dfm, margin="documents", method="cosine"))
diag(simil) <- NA
df <- as.data.frame(which(simil >= cutoff_lower & simil <= cutoff_upper, arr.ind = TRUE)) %>%
rownames_to_column("rowid") %>%
mutate(colid = colnames(simil)[col]) %>%
.[,c(1,4)] %>%
group_by(colid) %>% summarise(rowid=list(rowid))
text <- capture.output(stream_out(df))
# write(text[-length(text)], file = paste0(getwd(),'/dupe_objects.json'), append=T)
simil[upper.tri(simil)] <- NA
# write(unique(rownames(which(simil >= cutoff_lower & simil <= cutoff_upper, arr.ind = TRUE))),
# file = paste0(getwd(),'/remove_ids.txt'),
# append=T)
dupe_delete <- data.frame(id=unique(rownames(which(simil >= cutoff_lower & simil <= cutoff_upper, arr.ind = TRUE))),
dupe_delete = rep(1,length(unique(rownames(which(simil >= cutoff_lower & simil <= cutoff_upper, arr.ind = TRUE))))))
bulk <- c(apply(df, 1, bulk_writer, varname='duplicates', type = 'set'),
apply(dupe_delete, 1, bulk_writer, varname='_delete', type = 'set'))
if (length(bulk) > 0) {
res <- elastic_update(bulk, es_super = es_super, localhost = localhost)
}
return(paste0('Checked ',params$doctypes,' on ',params$dates ))
} else {
return(paste0('No results for ',params$doctypes,' on ',params$dates ))
}
} else {
return(paste0('No results for ',params$doctypes,' on ',params$dates ))
}
### Dummy code to verify that filtering out unique ids using the bottom half of the matrix actually works
# id_filter <- unique(rownames(which(simil >= cutoff, arr.ind = TRUE)))
# dfm_nodupes <- dfm_subset(dfm, !(docnames(dfm) %in% id_filter))
# simil_nodupes <- as.matrix(textstat_simil(dfm_nodupes, margin="documents", method="cosine"))
# diag(simil_nodupes) <- NA
# which(simil_nodupes >= cutoff, arr.ind = TRUE)
}