This paper presents an evolutionary optimization approach to generate rule-based trading strategies using financial sentiment data from StockTwits. The sentiment data includes bullish and bearish intensity scores (0-4) and message counts, normalized to create four variables: ibull, rbull, ibear, rbear. Each trading rule is encoded as an 8-gene chromosome consisting of four binary flags (indicating which conditions are active) and four real-valued thresholds (0-1). The rules determine when to enter or exit a long position based on sentiment thresholds. A genetic algorithm with elitism and three mutation operators evolves these rules to maximize a Sharpe-like ratio (expected return divided by standard deviation) over an in-sample period (2010-2013). The best strategies for each of the 30 DJIA stocks are then combined into an equally weighted portfolio, which is tested out-of-sample in 2014. Results show that the evolved sentiment portfolio outperforms both the Markowitz mean-variance portfolio and the 1/N portfolio on key risk metrics including semi-deviation, downside deviation, maximum drawdown, Value-at-Risk, and Expected Shortfall. The paper concludes that sentiment-based evolutionary trading strategies offer superior risk management and suggests future work incorporating transaction costs, rolling horizon recalibration, and more flexible rule generation via genetic programming.