There seems to be a heavy bias on the Neutral expression. This could possibly be fixed by adding "weights" to emotions (between 0 and 1) that would be a multiplier of the emotion value returned by the API.
For example if the API returns 50% neutral and 50% surprise, if neutral has a weight of 0.5 and surprise has a weight of 1.0, neutral will have a "weighted" score of 25% (0.5 * 50) and surprise would have a score of 50% (1.0 * 50). Surprise would then be the highest scoring emotion and would be displayed.
There seems to be a heavy bias on the Neutral expression. This could possibly be fixed by adding "weights" to emotions (between 0 and 1) that would be a multiplier of the emotion value returned by the API.
For example if the API returns 50% neutral and 50% surprise, if neutral has a weight of 0.5 and surprise has a weight of 1.0, neutral will have a "weighted" score of 25% (0.5 * 50) and surprise would have a score of 50% (1.0 * 50). Surprise would then be the highest scoring emotion and would be displayed.