We have published a new analysis of the different Sentiment Analysis services in the market (those that provide a free online evaluation API).
This time we have analyzed the results these services return for conditional sentences, and how this affects sentiment analysis. Current example is
"If the iPhone had a better price, then I would buy one".
At first sight we could state that this sentence contains a positive opinion about the iPhone. In fact, most tools will tag it that way, because of the "better price" part. But this positive expression appears within a conditional clause. The author isn't stating that the iPhone has a better price than any other; he's saying that if it had a better price... So, it's not correct to consider this a positive reference. It's much safer to tag it as neutral (we could even consider it negative, the implication being that it does not have a good price now).
What is the result of the analysis for this sentence in the different API Demos?
The main trends we have detected:
- Most APIs don't take into account the conditional structure nor manage it in any special way.
- Because of that, most APIs mark the sentence as positive with respect to the iPhone (as if the "if" expression didn't exist at all) and not as neutral.
This is another example about how a Sentiment Analysis service must be able to detect special linguistic structures (such as conditional sentences) in order to extract all the relevant Sentiment information from a text.
Check it for yourself!