Replication-Ready Text Analysis Pipeline • krpoltext

This vignette demonstrates a reproducible text analysis pipeline using krpoltext with quanteda. For background on the data, see the Data Descriptor:

Lim, T.H. (2025). South Korean Election Campaign Booklet and Party Statements Corpora. Scientific Data, 12, 1030. https://doi.org/10.1038/s41597-025-05220-4

Step 1: Load and Filter

library(krpoltext)

ps <- load_party_statements()
ps_sub <- get_docs("party_statements", year = 2010:2022, .data = ps)

cat("Documents:", nrow(ps_sub), "\n")
cat("Years:", paste(range(ps_sub$year), collapse = "-"), "\n")

Step 2: Convert to quanteda Corpus

library(quanteda)

corp <- as_quanteda_corpus(ps_sub, docid_field = "id")
docvars(corp) |> head()

Step 3: Tokenize and Build DFM

The filtered column already contains morphologically parsed Korean text (via khaiii). You can use it directly or tokenize the raw text column:

toks <- tokens(corp, remove_punct = TRUE, remove_numbers = TRUE)
dfm_ps <- dfm(toks)
dfm_ps

Step 4: Exploratory Analysis

topfeatures(dfm_ps, 20)

# By partisan group
dfm_group <- dfm_group(dfm_ps, groups = partisan)
topfeatures(dfm_group, 10, groups = TRUE)

Step 5: Keyword Analysis (Keyness)

library(quanteda.textstats)

dfm_binary <- dfm_group(dfm_ps, groups = conservative)
keyness <- textstat_keyness(dfm_binary, target = "1")
head(keyness, 20)

Step 6: Campaign Booklet Analysis

cb <- load_campaign_booklet()
cb_assembly <- get_docs(
  "campaign_booklet",
  office = "national_assembly",
  .data = cb
)

corp_cb <- as_quanteda_corpus(cb_assembly, docid_field = "code")
dfm_cb <- corp_cb |>
  tokens(remove_punct = TRUE) |>
  dfm()

# Compare by election result
dfm_result <- dfm_group(dfm_cb, groups = result)
topfeatures(dfm_result, 10, groups = TRUE)

Saving Results

saveRDS(dfm_ps, "dfm_party_statements_2010_2022.rds")

Citation

citation("krpoltext")

Session Info

sessionInfo()