Sentiment API Market Comparison: Negative Sentences

There is a significant number of Sentiment Analysis services in the market. At least seven, according to our count, offer a free evaluation API demo. Maybe there are more. As a result, identifying the pros and cons of each one of them is a complex and time-consuming task.

To simplify this task, at Bitext we have started comparing the Sentiment Analysis services offered via public API demos whose terms of use do not prevent its use in comparisons: Repustate, OpenDover, Lymbix, Skyttle, uClassify, Aiaioo Labs, Daedalus. (If you know of any other that we may have missed, please let us know!)

As a first step, we have started the test with a relatively simple sentence:

"The food was not bad, but the service was rather bad".

This sentence contains two different opinions (one about the "food" , another one about the "service"), with different polarities (positive about "food"; negative about "service").

What is the result of the analysis for this sentence in the different API demos?

Some trends we have detected:

  • Most APIs provide a general score/value for the sentence (or the document), not identifying the actual topics affected by sentiment, “food” and “service”.
  • Most APIs identify the expression “not bad” as negative, rather than positive. Probably due to a problem with detection of negation.
  • Most APIs don’t identify polarity modifiers or intensifiers like “rather” (in “rather bad”).

Check it for yourself!


Subscribe Here!