Crude oil is one of the most powerful types of energy and the fluctuation of its price influences the global economy. Therefore, building a scientific model to accurately predict the price of crude oil is significant for investors, governments and researchers. However, the nonlinearity and nonstationarity of crude oil prices make it a challenging task for forecasting time series accurately. To handle the issue, this paper proposed a novel forecasting approach for crude oil prices that combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Long Short-Term Memory (LSTM) with attention mechanism and addition, following the well-known “decomposition and ensemble” framework. In addition, a news sentiment index based on Chinese crude oil news texts was constructed and added to the prediction of crude oil prices. And we made full use of attention mechanism to better integrate price series and sentiment series according to the characteristics of each component. To validate the performance of the proposed CEEMDAN-LSTM_att-ADD, we selected the Mean Absolute Percent Error (MAPE), the Root Mean Squared Error (RMSE) and the Diebold-Mariano (DM) statistic as evaluation criterias. Abundant experiments were conducted on West Texas Intermediate (WTI) spot crude oil prices. The proposed approach outperformed several state-of-the-art methods for forecasting crude oil prices, which proved the effectiveness of the CEEMDAN-LSTM_att-ADD with the news sentiment index.