Mining the emotional information in the audio of earnings conference calls : A deep learning approach for sentiment analysis of securities analysts' follow-up behavior