wersimtext: Run Text models on corpora that were modified using the...

In jenswaeckerle/wersim: Simulating and Measuring Word Error

Run Text models on corpora that were modified using the WERSIM function

Description

This function calculates the word error rate between a hypothesis and a reference corpus.

Usage

wersimtext(x, measured_wer, new_wer, deletions_sim = 0.13,
  insertions_sim = 0.22, substitutions_sim = 0.65, num_sims,
  preprocessing = c("punctuation", "numbers"), mincount_wersim = 0,
  method, groupingvar_sim, direction = c(1, 2))

Arguments

x	A quanteda corpus to be modified
measured_wer	The word error rate (or an estimate thereof) of corpus x
new_wer	The word error rate for which the text model should be run. Generally, this is measured_wer plus some fixed incremental error such as 0.05
deletions_sim	The share of word error that should be introduced through deletions
insertions_sim	The share of word error that should be introduced through insertions
substitutions_sim	The share of word error that should be introduced through substitutions
num_sims	The number of simulations to be run
preprocessing	The preprocessing that should be done with the corpus. Defaults to c("punctuation","numbers") for excluding numbers and punctuation, can take on "min_term", stemming", "stopwords_en", "stopwords_de".
mincount_wersim	If "min_term" is part of "preprocessing", this parameters specifices the minimum number of times a word has to be in the corpus to be retained in the dfm
method	The text model that should be run on the simulated corpus. Can either be "sentiment" or "wordfish"
groupingvar_sim	The variable that groups the corpus
direction	For Wordfish: The parameters that are forwarded to the dir command in Wordfish (fixing the direction of the space). Defaults to c(1,2).

Value

A data frame with the grouping variable in column 1 and simulated quantities (sentiment or Wordfish estimates) in subsequent columns.

Examples

library(RecordLinkage)
library(quanteda)
library(wersim)
corp=corpus(data_corpus_dailnoconf1991)
dfm_corp=dfm(corp,groups="party")
wordfish_pos=austin::wordfish(quanteda::as.wfm(dfm_corp))
wordfish_pos_res=data.frame("theta"=wordfish_pos$theta,"party"=wordfish_pos$docs)
wersimulated_positions=wersimtext(corp,measured_wer=0,new_wer=0.05,deletions_sim=0.13,insertions_sim=0.22,substitutions_sim=0.65,num_sims=5,method="wordfish",groupingvar_sim="party")

jenswaeckerle/wersim documentation built on Dec. 7, 2022, 9:31 a.m.