​ I am not well versed in NLP, so hopefully someone can help me out here. I am looking at safety incidents for my organization. I want to compare the text of incident reports and observations to investigate if our observations are deterring incidents. I have a dataset of the incidents and a dataset of the observations. Both datasets have a free-text field that contains the description of the incident or observation. There is not really a good link between observations and incidents (as in, these observations were monitoring X activity on Y contract, and an incident also occurred during X activity on Y contract). My feeling is that the observations are just busy work; they don’t actually observe the activities that need safety improvement. The correlation between number of observations and number of incidents is minor, but I want to make a stronger case. I want to investigate this by using NLP to describe the incidents, then describe the observations, and see if there is a difference in content.…