Documentation Guide for VERN

(For model description, click here)

Making a Request

V.E.R.N. can be accessed via an API by POST request to http://vern.stage.vernai.com/analyze

V.E.R.N. requires the following Header configuration. Replace the Authorization value with your API Key.

Header: {
'Authorization': 'XXXXX:XXXXXXXXXXX',
'Content-Type': 'application/json',
'Accept': 'application/json'
}

Send the message you would like analyzed via the “text” key in the POST form data.

Body: {
text: "Sample message"
}

Response

V.E.R.N. returns the following JSON response. It includes the message that was analyzed along with a breakdown of the confidence rating on a scale of 0.00 to 100.00

{
"text": "Sample message",
"scores": [
{
"name": "sadness",
"value": 0
},
{
"name": "anger",
"value": 12.5
},
{
"name": "humor",
"value": 0
}
]
}

Understanding the Results

V.E.R.N. scores the results of an analysis on a percent confidence scale between 0.00 – 100.00. The higher the confidence rating, the more likely the detected emotion exists within the message. Our current standard is that a rating of 51% shows a statistical significant likelihood the emotion is present.

 

The output is a single score per enabled emotion analyser. That score is a confidence level in the range 0-100 against the specified emotion. 

 

The following is a general guide to the interpretation of these scores:

 

ScoreInterpretation
70-100High confidence the text reflects the emotion.

The higher the score the more intense the emotion is felt by a receive

50-70Moderate confidence text reflects the emotion.

May contain other emotions,

0-50No meaningful information, no interpretation.

 

The model is designed such that it is increasingly hard for a sample of text to increase a score. This can be compared to “logarithmic scaling”. As such a score in the range 60-80 is considered to be a clear signal. Scores above 80 should be investigated as this can mean the input text is highly construed or atypical in structure.

 

Note that a high score from a selected emotion analyzer does not preclude a high score from a different emotion analyzer.

A single message may contain multiple emotions cues as we tend to package a lot of data when we communicate in order to convey complex emotions. More complex emotions may be found through the combinations of core emotives (Humor, Anger, Sadness, Happiness).

The applications of V.E.R.N. can vary based on the context being applied. They can either be used to help score communication for emotional content or be used within the context of a decision matrix where specific responses are triggered if an emotional threshold is reached.

Example of how to use VERN

Examples of how to use VERN

Here's a look at the side-by-side comparison of VERN to popular databases in sentiment and emotional analysis. Click to see the Praveengovi Dataset 1

Here's a look at the side-by-side comparison of VERN to popular databases in sentiment and emotional analysis. Click to see the Praveengovi Dataset 2

Since most of us alive now know the smash hit "Friends" (which is enjoying a much-deserved renaissance), we thought we'd show a small snippet of how VERN would analyze an episode. 

V.E.R.N. emotional recognition AI enables system understanding of human emotions.

Newsletter Signup

Contacts