Researchers have developed Pharos-ESG, a novel framework designed to parse and analyze Environmental, Social, and Governance (ESG) reports. This system addresses challenges posed by the irregular layouts and implicit hierarchies within these documents by employing multimodal parsing, contextual narration, and hierarchical labeling. Pharos-ESG integrates modules for reading-order modeling, hierarchy-aware segmentation, and multimodal aggregation to transform visual and textual data into coherent natural language, enriched with ESG, GRI, and sentiment labels. The framework's effectiveness has been demonstrated through extensive experiments, outperforming existing document parsing and general multimodal models. Additionally, the researchers have released Aurora-ESG, a large-scale dataset of ESG reports from China, Hong Kong, and the U.S., to support financial research and decision-making. AI
IMPACT This framework could streamline the analysis of ESG reports, potentially improving investment decisions and corporate accountability.
RANK_REASON The cluster describes a new framework and dataset presented in an academic paper. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
