Research on Financial News Sentiment Analysis and Stock Market Prediction Based on Natural Language Processing
Abstract
Financial news provides important signals for explaining short-term movements in the Chinese securities market. Existing methods, however, insufficiently integrate target-specific Chinese financial semantics, sentiment uncertainty, dynamic stock relations, delayed news effects, and heterogeneous market information. This study proposes a Dynamic Knowledge-Enhanced Sentiment Graph Transformer (DKE-SGFormer). It combines Chinese financial news encoding with knowledge-enhanced sentiment estimation to generate polarity distributions and uncertainty measures. A dynamic multi-relational graph and lag-aware temporal graph transformer captures cross-stock information propagation and delayed market responses while gated multimodal fusion integrates textual, relational, and market representations. Transaction-cost-aware backtesting further suggests stronger risk-adjusted portfolio performance. The findings provide an integrated framework for financial news analysis and stock prediction in the Chinese market.