Comparing AWS and GCP NLP API for Sentiment Analysis and a Case for Custom Model

Transfer learning is the premise of learning from data in one domain, and then being able to transfer that knowledge to another domain. In this case, we leverage large existing text datasets to build an initial model, and then tailor that model to customer feedback and apply it to our specific problem (for this blog post, classifying feedback sentiment).

We chose to use this approach for customer feedback since it combines the benefits of a general model (large amounts of data) with the benefits of a domain-specific model (targeted to customer feedback). It looks like we were validated, as it outperformed Google NLP API and AWS Comprehend - both general models - on our set of customer feedback data.

I would imagine that AWS Comprehend utilizes a lot of customer feedback when training its internal models. However, they do advertise Comprehend as a general purpose NLP service, so they may have taken steps to make the models generalize to the non-feedback domain. In any case, on our set of sentiment data for customer feedback, Comprehend is outperformed by Google’s NLP API and our own internal model.

As for Azure, we actually did not factor their service into our comparison. This is definitely something to consider in future blog posts.