Harnessing Business and Media Insights with Large Language Models

TL;DR

FALM addresses the challenge of applying large language models to the business and media domain by grounding a specialized model in Fortune Knowledge derived from articles, video transcripts, and ranking lists. It introduces time-aware reasoning, thematic trend modeling, and content referencing to improve accuracy and trust, while enabling direct data visualization via a Python-based code generation pipeline grounded to an external dataframe. The approach leverages a unified instruction-finetuning framework and a multi-layer safety and deployment guardrail system, supported by a scalable serverless architecture and vector retrieval components. Evaluations, including automated and human assessments, show substantial gains over baselines across open-ended QA, metric QA, ranking QA, content referencing, and visualization, along with strong safety performance, indicating practical readiness for business analysts seeking auditable analytics. Overall, FALM demonstrates a scalable, trustworthy tool for business analytics that couples precise knowledge grounding with interactive visualization and robust safety safeguards.

Abstract

This paper introduces Fortune Analytics Language Model (FALM). FALM empowers users with direct access to comprehensive business analysis, including market trends, company performance metrics, and expert insights. Unlike generic LLMs, FALM leverages a curated knowledge base built from professional journalism, enabling it to deliver precise and in-depth answers to intricate business questions. Users can further leverage natural language queries to directly visualize financial data, generating insightful charts and graphs to understand trends across diverse business sectors clearly. FALM fosters user trust and ensures output accuracy through three novel methods: 1) Time-aware reasoning guarantees accurate event registration and prioritizes recent updates. 2) Thematic trend analysis explicitly examines topic evolution over time, providing insights into emerging business landscapes. 3) Content referencing and task decomposition enhance answer fidelity and data visualization accuracy. We conduct both automated and human evaluations, demonstrating FALM's significant performance improvements over baseline methods while prioritizing responsible AI practices. These benchmarks establish FALM as a cutting-edge LLM in the business and media domains, with exceptional accuracy and trustworthiness.

Figures (2)

• Figure 1: This figure summarizes FALM's performance across various tasks compared to a state-of-the-art open-source LLM with prompt engineering as the baseline. FALM achieves significant improvements in: Open-Ended Question Answering (4.8x), Metric QA (3.9x), Ranking QA (8.7x), Content Referencing (1.6x), Data Visualization (2.5x). Notably, FALM's focus on the business and media domain also leads to a substantial increase in inference speed (3.8x). Furthermore, FALM demonstrates strong performance in handling harmful prompts, ensuring responsible AI practices.
• Figure 2: End to end System architecture design