> “It’s a big job,” Lyon said. “Our human labelers will have to assess and label at least 10,000 tweets.” This is where my concern comes in. For example the st

> “It’s a big job,” Lyon said. “Our human labelers will have to assess and label at least 10,000 tweets.” This is where my concern comes in. For example the statement "Dang immigrants are taking our God given jobs." vs the statement "An influx of immigrants into a region increases the amount of labor available and decreases the cost of said labor reducing wages." does the second one constitute hate speech? I could see many people arguing it does and just as many people arguing contrary to that. Does this mean the model is really just trained to identify speech the labelers disagree with? Further the other big problem is that it only tracks twitter, and the disconnect between reality and twitter I've found is fairly broad. Most people I know do not use twitter and those that do almost none of them tweet in any meaningful capacity. So why the heck is that a good barometer. I've found twitter to not be a reflection of a population, but rather a funhouse mirror of the population. EDIT: Updated first paragraph and fixed some punctuation.