LLM API Research with R

This repository collects R-focused workflows, utilities, and examples for integrating Large Language Model (LLM) APIs (e.g., OpenAI, Anthropic, etc.) into computational social science and media/communication research.

Goals

• Provide reusable R scripts for calling LLM APIs with robust error handling and logging.
• Demonstrate end-to-end pipelines for social media and text data (CSV-based) using tidyverse and httr/httr2.
• Facilitate reproducible research for topic modeling, sentiment, and annotation workflows that rely on LLMs.
• Connect to related repositories in the Sawood Anwar GitHub profile for thesis-related and NLP projects.

Repository Structure

• R/
  • llm_api_client.R: Core wrapper functions for making LLM API calls from R (POST requests, JSON handling).
  • llm_api_error_handling.R: Extra utilities for retries, rate-limit backoff, and structured error logging.
  • llm_batch_requests.R: Helpers for batched calls over data frames (e.g., news posts, comments).
• data/
  • README_data.md: Notes on expected data formats (CSV), privacy/anonymization, and example schemas.
  • example_posts.csv: Small synthetic dataset of COVID-19 news posts and reaction counts for quick testing.
• examples/
  • covid_news_reactions_llm_api.R: Example pipeline inspired by PhD thesis data (Facebook reactions & COVID-19 news).
  • simple_prompting_demo.R: Minimal reproducible example for first-time LLM API integration in R.
• logs/
  • Directory for storing API error logs (e.g., .csv or .json) and run metadata.

Data Notes

This repository assumes input data is in CSV format with columns such as post_id, page_name, post_text, created_at, and reaction counts. Data should be anonymized and compliant with platform terms and institutional ethics requirements.

See data/README_data.md for details.

Cross-links to Related Repositories

• LLM text analysis workflows in R: llm-text-analysis-r
• BERTopic topic modeling workflows: bertopic-media-topics
• Profile and other research repositories: GitHub profile

Together with llm-text-analysis-r and bertopic-media-topics, this repository is part of a small computational communication toolkit for LLMs and topic modeling.

Example Usage

A minimal example of running batched LLM calls over the synthetic COVID-19 news posts:

library(readr)
source("R/llm_api_client.R")
source("R/llm_api_error_handling.R")
source("R/llm_batch_requests.R")

posts <- read_csv("data/example_posts.csv")
results <- llm_batch_call(posts, text_col = "text", model = "gpt-4.1-mini")
head(results$llm_output)

Getting Started

1. Clone the repository:

   git clone https://github.com/sawoodanwar/llm-api-research-r.git

2. Install R packages:

   • tidyverse
   • httr or httr2
   • jsonlite
   • readr

3. Configure your API key(s):

   • Store them in .Renviron or environment variables (e.g., OPENAI_API_KEY).
   • Never commit secrets to the repository.

4. Run an example script, e.g.:

   source("R/simple_prompting_demo.R")

License

This repository is licensed under the MIT License. See LICENSE for details.