In recent years, with the verification of the carbon market's effectiveness in conserving energy and mitigating emissions, accurate carbon price prediction has attracted the interest of researchers and investors. However, carbon price forecasting is widely considered intractable due to its various non-stationary properties. A novel data-driven carbon prices forecasting approach using online news data and Google trends is applied in this paper. Word embedding algorithm is adopted to identify the text features of online carbon market news, which expressing the text information. Long Short Term Memory (LSTM) algorithm is applied to forecast carbon prices. Finally, a comparison analysis is employed, the results of which show that the proposed framework performs better than traditional statistical forecasting models with respect to predictive ability and robustness.