The emotion model
The emotion model learned from a large collection of images that people labelled with the emotion each one made them feel. It reads your original creative and predicts the strength of eight emotions. Emotional Engagement and Valence are derived from those eight values.
The emotion model predicts which emotions your creative is likely to evoke in a typical viewer, and how strongly. It is the source of the emotional profile, Emotional Engagement and Valence.
What it measures
The model scores eight emotions: four pleasant ones (amusement, awe, contentment, excitement) and four unpleasant ones (anger, disgust, fear, sadness).
It measures the feeling the content evokes in the viewer, not the expressions of the people shown in it. A photo of a smiling person and a photo of a breathtaking landscape can both evoke awe or contentment; what matters is how the content is likely to make the viewer feel.
What it learned from
The model learned from a large collection of images, each labelled by people with the emotion it evoked in them.
From those human judgements it learned which visual and narrative cues tend to evoke excitement, fear, sadness and the other emotions: a celebration, a threat, a loss, a joke, a spectacular view.
What feeds it
When you run an analysis, the emotion model receives two things:
- Your original creative: never the platform mock-up. If you choose a platform, a small, fixed contribution from its interface is added to the results afterwards.
- A short description of its content. The emotion model also uses a short description of the creative's content as context, so it reads both the visuals and what they depict.
Sound is not analysed directly; music and sound effects do not affect the scores. Spoken words can be reflected indirectly when the content description mentions them.
How it works
For each image (or each video frame) the model returns a strength for every one of the eight emotions. Each emotion is scored independently: several emotions can be high at once (a mixed reaction, such as excitement and fear in a thriller trailer), or all can be low (calm, neutral content). The eight values don't add up to 100.
Two KPIs are then derived from the eight values:
- Emotional Engagement summarises how intense the overall reaction is, whatever its tone. It reflects how activating the predicted emotions are: higher-energy emotions count more than calmer ones. The result is ranked against Foresight's reference creatives. Polarity itself plays no part: a strong negative reaction is comparable to a strong positive one of similar energy. See Emotional Engagement.
- Valence summarises the tone, from −100 (clearly negative) to +100 (clearly positive). Pleasant emotions push it up, unpleasant ones push it down. It describes the creative itself rather than a ranking. See Valence.
Both are derived the same way for every creative.
Scale & direction
| Output | Scale | Direction |
|---|---|---|
| Each of the eight emotions | 0–100 | Higher = that emotion is more strongly predicted. Not compared with a benchmark. |
| Emotional Engagement | 0–100 | Higher = a more intense reaction; around 50 is typical. |
| Valence | −100 to +100 | Higher = more positive tone. Not "better" in itself, it depends on your goal. |
Emotional Engagement is positioned against Foresight's reference creatives of the same media type: an image among images, a video among videos. For videos, each moment on the timeline is positioned against individual frames from the reference videos. See how scores are positioned.
Image vs video
- Image: one prediction gives one emotional profile.
- Video: every frame is predicted, so the emotional timeline is continuous. The video's overall profile, Emotional Engagement and Valence summarise the whole timeline. See How videos are analysed.