Joe Rappaport, Charles River Labs
Open text written in a context-heavy field can be tricky for generic sentiment models to accurately describe. This difficulty arises because the broader assumptions of the training dataset are not the same as your business’s data. The hurdles are jargon and lexicon, masked negativity, and framing. By crowd-sourcing an in-house model with your employees as the experts, you can leverage their business acumen to create broader external validity for future sentiment extraction within your organization. The objective of this talk is to inform you on each step of the process you will need to leverage this method for sentiment extraction of context-laden text in a fast, reliable, and objective manner. This method has the advantages of maximizing business insights and thereby returning value on your data. Pitfalls, breakthroughs, and surprises will be covered.