LLMs Transform Scientific Literature Discovery

A recent chapter examines 34 peer-reviewed studies that utilize generative large language models (LLMs) for retrieving and screening scientific literature. This research was identified through a Boolean search within the OpenAIRE Graph, narrowing down from 1,589 records to 34 selected papers. It discusses the LLMs implemented, their accessibility and customization, prompting strategies, architectural methods, sources of ground truth, and evaluation standards. With the rapid expansion of academic publications, pinpointing relevant works has become increasingly challenging, as traditional search methods still rely on manually crafted queries and labor-intensive reviews. In contrast, generative LLMs provide a more adaptable solution, facilitating literature retrieval and the assessment of studies based on eligibility criteria.