In this work, the application of the Triple-Barrier Method and Meta-Labeling techniques are explored using XGBoost to develop a sentiment-based trading signal for the S&P 500 stock market index. The results indicate that sentiment data possess predictive power; however, substantial work remains before a fully implementable strategy can be realized.

Systematic signal refinement: PIT data → triple-barrier labels → LGBM/MLP → calibrated classwise blend → probability-weighted portfolio with realistic costs.