Package 'metaextractoR'

Sample abstracts

Description

This dataset includes sample abstracts downloaded from one of my Covidence project. Pre-processing was done by running the clean_name function from janitor package.

Usage

abstracts

app_2

app_3

testing_stage_0_data

training_stage_0_data

Format

A data frame with 20 rows and 16 variables:

title

Title of the manuscript

authors

Authors

abstract

The column that contains all abstracts

published_year

Year of the publication

published_month

Month of the publication

journal

Journal's name

volume

Volume number

issue

Issue number

pages

Page number

accession_number

Accession number

doi

doi number

ref

Reference

covidence_number

Covidence number

study

Study

notes

Notes

tags

Tags

An object of class spec_tbl_df (inherits from tbl_df, tbl, data.frame) with 10 rows and 15 columns.

An object of class spec_tbl_df (inherits from tbl_df, tbl, data.frame) with 40 rows and 16 columns.

An object of class spec_tbl_df (inherits from tbl_df, tbl, data.frame) with 40 rows and 16 columns.

An object of class spec_tbl_df (inherits from tbl_df, tbl, data.frame) with 10 rows and 16 columns.

Source

Created for demonstration purposes

---

Pre-processing functions before shinyapps

Description

Pre-processing functions and saving intermediate data

Usage

add_predefined_vars(data, list_vars)

Arguments

data	a csv file contains abstract information. This could be the csv file downloaded from covidence
list_vars	a vector of data elements you want to extract. i.e. c("no_participants,"no_female","..")

Value

new dataset with additional empty columns

Examples

add_predefined_vars(abstracts,
                   c("no_participants", "no_aki", "age_mean", "age_sd"))

---

Launch the first shinyapp for to look into what variables are available in the abstract.

Description

This function launches a Shiny app that displays each abstract alongside the data elements you want to extract. The app provides an interactive interface for manually entering this information, helping to fill out the _manual column generated by the 'add_predefined_vars()' function in the training set.

Usage

glance_manual_app()

Value

Runs a shiny app.

Examples

if (interactive()) {
 glance_manual_app()
}

---

manual_validation_app

Description

This Shiny app allows you to manually review the results generated by the 'process_with_ollama()' function. By uploading a testing dataset with LLM-extracted values, you can systematically check and validate each field filled out by the model.

Usage

manual_validation_app()

Value

Runs a shiny app.

Examples

if (interactive()) {
  manual_validation_app()
}

---

Process the abstract with a large language model

Description

A function that batch process data extraction from text.

Usage

process_with_ollama(
  input,
  model = "llama3.1:8b",
  type_abstract,
  i,
  abstract_col
)

Arguments

input	A data frame contains abstracts and variables we want to extract. If you are not sure how the data should look like, please see the example dataset abstract. This data should be the testing data saved using the 'save_training_data()' function.
model	Large Language Model name you want to use i.e. "llama3.1:8b". If you are not sure what LLMs have been installed locally, please use the function: ellmer:::ollama_models() to find out what LLMs are available.
type_abstract	is created using the type_object function from the ellmer package. Objects represent a collection of named values and are created with type_object(). Objects can contain any number of scalars, arrays, and other objects. They are similar to named lists in R. If still not sure, please look up the documentation of ellmer::type_object().
i	number of abstracts you want to process at once.
abstract_col	column name for the column contains abstract.

Value

Returns a data.frame with results and time.

Examples

library(ellmer)
type_abstract <- type_object("Some key information from abstract.",
  no_patients_llm = type_integer("Find the total number of patients
                                  included in this study.",
                                  required = FALSE),
  no_AKI_llm = type_integer("Number of Acute Kidney Injury (AKI) patients
                             included in this study.",
                             required = FALSE),
  per_AKI_llm = type_number("Percentage of Acute Kidney Injury ",
                             required = FALSE),
  ICU_llm = type_boolean("Included only Intensive Care Unit (ICU) patients.",
                          required = FALSE),
  start_date = type_string("The starting date of the study written in
                            YYYY-MM-DD format",
                            required = FALSE),
  end_date = type_string("The end date of the study written in YYYY-MM-DD
                          format",
                          required = FALSE),
  age_mean_llm = type_number("Find the average age of the study cohort",
                              required = FALSE),
  age_median_llm = type_number("Find the median age of the study cohort",
                                required = FALSE)
)
## Not run: 
process_with_ollama(abstracts,
                    type_abstract = type_abstract,
                    i = 1,
                    abstract_col = "Abstract")

## End(Not run)

---

prompt_engineering_app

Description

This Shiny app is designed for model testing and prompt engineering. It uses the training set—with manually extracted values from 'glance_manual_app()' —as the reference for training and evaluation. When run, the function automatically creates a log_files folder in your current working directory. After you exit the prompt_engineering_app, a CSV file is saved in this folder, containing details of each prompt tested, results, runtime, model type, and accuracy. This log supports model selection, documentation, and reproducibility. The chosen model and prompt can be used for the 'process_with_ollama()' function.

Usage

prompt_engineering_app()

Value

Runs a shiny app.

Examples

if (interactive()) {
 prompt_engineering_app()
}

---

save_testing_data

Description

This function will save the testing abstract data with empty columns to the processed data folder. This csv file will be used in the shinyapp 2 The data will be stored in metaextroctor_process_data

Usage

save_testing_data(testing_abs)

Arguments

testing_abs

training abstracts including the variables you want to extract.

Value

a csv file saved in the metaextroctor_process_data file named training_stage_0_data.csv

---

save_training_data

Description

This function will save the training abstract data with empty columns to the processed data folder. This csv file will be used in the shinyapp 1 The data will be stored in metaextroctor_process_data

Usage

save_training_data(training_abs)

Arguments

training_abs

training abstracts including the variables you want to extract.

Value

a csv file saved in the metaextroctor_process_data file named training_stage_0_data.csv

---

Separates a data into training and testing datasets

Description

This function separates the abstracts into training and testing sets.

Usage

separate_training(data, percentage = 0.1)

Arguments

data	The csv file contains abstracts with
percentage	percentage of separation training sets. If percentage set to be 0.1, 10% of the data will be the training set and 90% of the data will be testing set.

Value

A list with two the train and test data.frames.

Examples

separate_training(abstracts, percentage = 0.4)

---