Leveraging LLMs for Systematic Reviews: Evaluating Venture Capital Investment Strategies Across Large Datasets

Abstract

The exponential growth of venture capital (VC) research creates challenges in systematically identifying relevant evidence on investment strategies. Traditional keyword searches ensure precision but miss conceptually related studies due to vocabulary mismatch. Semantic search broadens retrieval, yet risks including loosely related records, while large language models (LLMs) promise context-aware classification but introduce issues of accuracy, transparency, and reproducibility. This study compares keyword-based, semantic, LLM-assisted screening of a large VC corpus using Web of Science, EconLit, and Scopus. Experiments with ChatGPT and Claude reveal major differences in recall and precision depending on prompt design and semantic depth. Findings underscore the importance of hybrid approaches, combining manual rigor with LLM support, to balance recall, reduce bias, and improve efficiency in systematic reviews.

Ähnliche Arbeiten