This function applies AI-assisted analysis to each document in a character vector or a docvar in a corpus, using a structured type_object() to define the expected response format. It supports few-shot learning examples for better performance.
Usage
ai_text(
.data,
chat_fn,
type_object,
few_shot_examples = NULL,
verbose = TRUE,
result_env = NULL,
...
)Arguments
- .data
A character vector of texts
- chat_fn
A function like chat_openai(). See https://ellmer.tidyverse.org/articles/structured-data.html for details.
- type_object
an ellmer type_object()
- few_shot_examples
Optional few-shot learning examples (data frame with
text,score)- verbose
logical; whether to print progress
- result_env
An environment to store results and allow resuming
- ...
additional arguments passed to
chat_fn
Value
character; the response from the LLM with a length equal to the number of input documents; each single element defined by type_object() is added as a character vector
Examples
if (FALSE) { # \dontrun{
library(quanteda)
results <- quanteda::data_corpus_inaugural[1:3] %>%
ai_text(chat_fn = chat_openai, model = "gpt-4o",
api_args = list(temperature = 0, seed = 42),
type_object =
type_object("Summary of the document",
summary = type_string("Summarize the document in a few sentences.")),
)
} # }