Research on AI-based ESG Rating Models: From Data Integration to Investment Decision Optimization

Environmental, social, and governance (ESG) factors have become a core consideration in global investment decisions. However, traditional ESG ratings suffer from limitations such as fragmented data sources, low processing efficiency, high subjectivity, and poor timeliness, hindering their in-depth application in investment practice. This study focuses on building an ESG rating model based on artificial intelligence (AI) technology, exploring its application value across the entire chain from data integration and rating generation to investment optimization. The study systematically examines the diversity and complexity of ESG data sources and proposes a framework for intelligent data collection and cleaning from multiple sources, integrating structured financial data with unstructured text and image data. At the model construction level, the application of natural language processing (NLP) technology in extracting key ESG issues and sentiment analysis is explored. This study employed machine learning (ML) and deep learning (DL) methods for feature construction and credit rating prediction, significantly improving the objectivity, dynamic responsiveness, and risk identification performance of ratings. The paper further explored the potential of AI-enabled ESG rating mechanisms in asset allocation optimization, risk control, and portfolio performance improvement. Strategy backtesting combined with historical data verified the model's practical effectiveness. Furthermore, the study explored key challenges and development paths, including model interpretability, algorithmic fairness, and data reliability. Experimental results demonstrate that the AI-based ESG assessment framework is valuable in improving information integration efficiency, enhancing rating accuracy, exploring deep risk correlations, and promoting proactive investment decisions, providing financial institutions with advanced digital analysis tools.