A comprehensive review of open source large language models for the analysis of earnings call reports and financial documents including applications, datasets, and long term challenges

Large Language Models (LLMs) have become part of daily tasks for everyone. Large scale research work is undertaken by researchers and the industry sector to explore the new applications of LLMs in various domains, including finance. Earning conference calls and financial documents offer insights into a company’s performance, market outlook, and investor sentiments. This review explores the open source LLMs and their usage in analysis of earning conference call reports and financial reports. The initial part of the article focuses on bibliometric analysis to understand the requirements of review gaps and need analysis of growing importance of LLMs in earnings conference call analysis. The PRISMA approach based bibliometric analysis presented uncovering publication trends, Keyword cooccurrence, thematic research clusters, and the growing trends. The article further presents extensive review articles under three important thematic structures (i) LLM adaptation and infrastructure for financial language (ii) retrieval, reasoning and decision support from earnings calls (iii) socioeconomic signals, sentiment and managerial discourse. Additionally, detailed coverage and discussion on publicly available datasets, commonly used for earnings call transcript analysis, along with associated challenges related to preprocessing, and evaluation metrics relevant to financial text analytics are articulated. This review further elaborates open research challenges in earnings call analysis relevant to regulatory compliance, data provenance, and auditability. We have also covered data privacy concerns, hallucination risks, and computational constraints encountered during application level LLMs development. The article also highlights the growing influence of multimodal AI approaches. Finally, potential future research directions and long term challenges, encouraging researchers to further work to address the research gaps are discussed.