Research on Risk and Stock Price Forecasting of Listed Companies Based on Machine Learning and Sentiment Analysis

This study develops a risk-stock assessment framework by integrating financials, Weibo sentiment, and ML. For 5,100+ Chinese firms (2019–2024), we contextually imputed missing data via RF (vs. generic methods) and refined 21 indicators using time-lagged EDA (e.g., quarterly cash flow volatility). SNOWNLP analysis incorporated emoji-weighted lexicon (tech sector "up" +15% sentiment impact), correlating 960k posts to 3-day returns. The adaptive voting model (RF+Logistic) achieved 81.2% accuracy in identifying ±5% weekly anomalies. Unlike prior static models, our framework innovates by: (1) revealing sector-specific sentiment lags (consumer+48h vs. energy-12h); (2) dynamic risk scoring via Z-score trajectories + sentiment entropy. This work offers a process-aware approach, avoiding AIGC-prone generalities while preserving original methodology.