How multiple emotion detection means better accuracy

“When the sadness leaves you broken in your bed, I will hold you in the depths of your despair, and it's all in the name of love.”

Wow…forgive me, I need a minute…

Thanks. We found this sentence in our never-ending search for testing data. It’s a lyric from a Martin Garrix and Bebe Rexha song titled “In the name of love,” And it’s one of the most gut-wrenching, heart tugging sentences our human coders have ever encountered.


As you can see, this sentence is packed full of emotions. It has all the “feels.” Anger, sadness, fear, and love are all apparent in the sentence provided. Each is uniquely communicating the sender’s emotions, yet together they combine to provide a profound expression.

This sentence illustrates the need for emotion detection that is sensitive enough to detect multiple discreet and distinct emotional clues. For what good is an emotion detector if it only detects one emotion per sentence? Especially given that we humans frequently use mixed emotions to communicate with one another. 

It certainly has sadness, as expressed directly in the use of “sadness,” and other clues. Fear is very apparent in “depths of your despair,” and anger is apparent by the sender indicating a state of unfairness. Love? That’s abundantly clear. In fact, this may be a good example of what the Greeks called “agape,” or “sacrificial love.”

How would sentiment analysis handle this phrase?

Most of the popular software available does so by averaging the signals, and providing a “neutral,” or “mixed,” result. Some more sophisticated sentiment analysis products get a little closer-some label the sentence as “fear,” or “sadness,” but none seem to grasp the complexity of the emotional statement.

At VERN™ AI, we believe that humans express themselves with multiple emotions, and often at the same time.

So we created detectors that would identify each emotion separately, as a possible (and increasingly likely) prediction of the sender’s intent. Because we see it every day, and experience it in every conversation. We’ve provided an output of the VERN AI API with this sentence run on it.

As the example above illustrates, you need to understand ALL of the emotions in communication. Not just one. Or not just an average of all of them. (Or, even worse, the system merely labels it as “mixed,” which is really just giving up).

Instead, VERN™ AI predicts that there is a likelihood of 90{b1ba36726a3bfcdc42af6e5ec24af305dbc6425c95dfb7052d7f2b4aabbf1a02} Fear, 66{b1ba36726a3bfcdc42af6e5ec24af305dbc6425c95dfb7052d7f2b4aabbf1a02} Sadness, 80{b1ba36726a3bfcdc42af6e5ec24af305dbc6425c95dfb7052d7f2b4aabbf1a02} Anger, and 80{b1ba36726a3bfcdc42af6e5ec24af305dbc6425c95dfb7052d7f2b4aabbf1a02} Love in the message.

And with VERN™ AI, detections above 80{b1ba36726a3bfcdc42af6e5ec24af305dbc6425c95dfb7052d7f2b4aabbf1a02} are generally considered high intensity with high confidence. This does confirm that the message is VERY emotional, and likely is meant to convey such a profound message of love.

In order to gain competitive advantage, one should look to adding emotional intelligence like VERN AI. Schedule a demonstration today.


VERN™ AI JSON output for example phrase

“text”: “When the sadness leaves you broken in your bed, I will hold you in the depths of your despair, and it’s all in the name of love”,

    “scores”: [


            “name”: “fear”,

            “value”: 90,

            “raw”: 90



            “name”: “sadness”,

            “value”: 66,

            “raw”: 66



            “name”: “anger”,

            “value”: 80,

            “raw”: 80



            “name”: “love”,

            “value”: 80,

            “raw”: 80



Leave a Reply