Prompt Refinement or Fine-tuning? Best Practices for using LLMs in Computational Social Science Tasks

TL;DR

This paper evaluates various strategies for applying large language models to social science tasks, recommending best practices like model selection, prompting techniques, and fine-tuning based on task and data availability.

Contribution

It provides a comprehensive benchmark and practical guidelines for using LLMs effectively in computational social science research.

Findings

Larger vocabulary and pre-training data improve performance

AI-enhanced prompting outperforms simple zero-shot methods

Fine-tuning on task-specific data yields better results

Abstract

Large Language Models are expressive tools that enable complex tasks of text understanding within Computational Social Science. Their versatility, while beneficial, poses a barrier for establishing standardized best practices within the field. To bring clarity on the values of different strategies, we present an overview of the performance of modern LLM-based classification methods on a benchmark of 23 social knowledge tasks. Our results point to three best practices: select models with larger vocabulary and pre-training corpora; avoid simple zero-shot in favor of AI-enhanced prompting; fine-tune on task-specific data, and consider more complex forms instruction-tuning on multiple datasets only when only training data is more abundant.