I've been dreaming about building stuff like this (on a much simpler scale) for a long time. Perhaps a model that takes in past quarterly reports, commodity pr

I've been dreaming about building stuff like this (on a much simpler scale) for a long time. Perhaps a model that takes in past quarterly reports, commodity prices, executive performances and news regarding a company and calculates 'tones' surrounding a company. (for example appearance of 'good performance', 'meeting targets', 'vast improvements' would possibly yield a positive tone where as 'disappointing returns', 'failed endeavours' etc would yield negative). Then gauge past price sensitivities to various tones and see if there is any way to train the machine to speculate merely on these 'tones'.