Models
Overview
For predicted reactions, the Adverteyes API provides a set of models that predict the attention and emotional reactions of an audience to a video or image advertisement. The models are based on a combination of machine learning and computer vision techniques. The models are trained on a large dataset of videos and images with ground truth data on attention and emotional reactions. The models are designed to predict the attention and emotional reactions of an audience to a video or image advertisement based on the visual and auditory content of the advertisement. The models are designed to be used in real-time to provide feedback on the performance of an advertisement and to help advertisers optimize their creative content to maximize the impact on their target audience.
Attention Potential is also known as Sales Impact.
Model Name |
Applicable for |
Status |
|---|---|---|
reactions-v1.0 |
Video |
|
reactions-v1.1-composite |
Video |
|
reactions-v2.0 |
Video and Audio |
|
reactions-v2.1-composite |
Video and Audio |
|
reactions-v3.0 |
Video and Audio |
|
reactions-v3.2 |
Video and Audio |
Experimental |
reactions-v3.3 |
Video and Audio |
|
reactions-v3.4 |
Video and Audio |
Default for video and image |
reactions-v4.0 |
Video and Audio |
|
reactions-v4.1 |
Video and Audio |
Image API endpoint Request Predicted Reactions for Images
Video API endpoint Request Predicted Reactions for Video
Best audience detection
Image and video prediction requests can optionally find the best target audience for any subset of age_range,
gender_distribution, and platform_format. Add detect_best_audience to
POST /v1/{images|videos}/predicted-reactions/{file_id} as an array of the dimensions to optimize:
{
"model_version": "reactions-v3.4",
"model_params": {
"country": "US"
},
"detect_best_audience": [
"age_range",
"gender_distribution",
"platform_format"
]
}
The default is an empty array. Omitting the field, or passing [], disables best-audience detection and preserves the
existing response shape.
After processing completes, GET /v1/{images|videos}/predicted-reactions/{file_id} includes a top-level best_audience
object. Only values for requested dimensions are returned:
{
"state": "processed",
"data": { "...": "existing prediction output" },
"model_version": "reactions-v3.4",
"model_params": { "...": "existing model parameters" },
"norms_dimensions": { "...": "existing norms dimensions" },
"request_id": "request-id",
"best_audience": {
"age_range": "18-29",
"gender_distribution": "female",
"platform_format": "feed"
}
}
best_audience is additive and is omitted when detect_best_audience is empty, so existing consumers do not need to
change.
Best-audience values are selected from the audience-candidate artifact generated by Core. When detection was requested
but the artifact is not available yet, the endpoint continues to return state as processing. It returns processed
only after the requested best-audience values can be resolved. If a finished prediction is missing the Core job ID
needed to locate that artifact, the endpoint returns state as failed because polling cannot repair the metadata.
Models
reactions-v1.0
Attention Potential is Attention Mean Index.
Only video channel is analyzed (fit for video and image).
Example API request body:
{
"model_version": "reactions-v1.0"
}
reactions-v1.1-composite
Attention Potential is a weighted average of attention and emotion reactions indexes, chosen based on their correlation to sales signals.
Default weights of Attention Potential score are: Attention (0.65), Happiness (0.2), Confusion(0.03), Contempt (0.03) and Surprise (0.10)
Only video channel is analyzed (fit for video and image).
Example API request body:
{
"model_version": "reactions-v1.1-composite"
}
reactions-v2.0
Attention Potential is Attention Mean Index.
Video and Audio channels are analyzed (fit for video and audio).
Example API request body:
{
"model_version": "reactions-v2.0"
}
reactions-v2.1-composite
Attention Potential is a weighted average of attention and emotion reactions indexes, chosen based on their correlation to sales signals.
Default weights of Attention Potential score are: Attention (0.65), Happiness (0.2), Confusion(0.03), Contempt (0.03) and Surprise (0.10)
Video and Audio channels are analyzed (fit for video and audio).
Current default model for videos and images.
Example API request body:
{
"model_version": "reactions-v2.1-composite"
}
reactions-v3.0
Experimental model.
Attention Potential is a weighted average of attention and emotion reactions indexes, chosen based on their correlation to sales signals.
Default weights of Attention Potential score are: Attention (0.65), Happiness (0.2), Confusion(0.03), Contempt (0.03) and Surprise (0.10)
Engaged Attention is introduced in reactions-v3.0. The metric is computed as the percentage of people that was attentive and emotional engaged at the same time.
Video and Audio channels are analyzed (fit for video and audio).
Predictions are refined with target population details (age, gender, country/region) and channel specifics (ad format, device proportion).
In case a parameter is not set from the parameters list, the model will use the default value for that parameter.
Example API request body with the default parameters:
{
"model_version": "reactions-v3.0",
"model_params": {
"ad_format": "tv",
"age_range": "30-49",
"device_type": "desktop",
"gender_distribution": "mix",
"country": "US",
"view_month": "Sep",
"view_time": "morning"
}
}
For a general assessment of creative potential, it is recommended to include the country (based on asset language if applicable) and use default parameters for general population and placements.
parameters list
Param Name |
Param Value |
Default |
|---|---|---|
|
The target placing for the advertisement |
|
|
The age group of the target audience |
|
|
The device type where the advertisement is displayed |
|
|
The gender distribution of the target audience |
|
|
The country where the advertisement is displayed |
|
|
The month when the advertisement is displayed |
|
|
The daytime of the advertisement is displayed |
|
reactions-v3.2 (Experimental)
Experimental model.
Attention Potential is a weighted average of attention and emotion reactions indexes, chosen based on their correlation to sales signals.
Default weights of Attention Potential score are: Attention (0.65), Happiness (0.2), Confusion(-0.025), Contempt (-0.025) and Surprise (-0.10)
Engaged Attention is introduced in reactions-v3.0. The metric is computed as the percentage of people that was attentive and emotional engaged at the same time.
Video and Audio channels are analyzed (fit for video and audio).
Predictions are refined with target population details (age, gender, country/region) and channel specifics (ad format, device proportion).
In case a parameter is not set from the parameters list, the model will use the default value for that parameter.
Example API request body with the default parameters:
{
"model_version": "reactions-v3.2",
"model_params": {
"platform_format": "tv",
"age_range": "30-49",
"device_type": "desktop",
"gender_distribution": "mix",
"country": "US",
"view_month": "Sep",
"view_time": "morning"
}
}
For a general assessment of creative potential, it is recommended to include the country (based on asset language if applicable) and use default parameters for general population and placements.
parameters list
Param Name |
Param Value |
Default |
|---|---|---|
|
The target placing for the advertisement |
|
|
The age group of the target audience |
|
|
The device type where the advertisement is displayed |
|
|
The gender distribution of the target audience |
|
|
The country where the advertisement is displayed |
|
|
The month when the advertisement is displayed |
|
|
The daytime of the advertisement is displayed |
|
reactions-v3.3
Experimental model.
Attention Potential is a weighted average of attention and emotion reactions indexes, chosen based on their correlation to sales signals.
Default weights of Attention Potential score are: Attention (0.65), Happiness (0.2), Confusion(0.03), Contempt (0.03) and Surprise (0.10)
Engaged Attention is introduced in reactions-v3.0. The metric is computed as the percentage of people that was attentive and emotional engaged at the same time.
Video and Audio channels are analyzed (fit for video and audio).
Predictions are refined with target population details (age, gender, country/region) and channel specifics (ad format, device proportion).
In case a parameter is not set from the parameters list, the model will use the default value for that parameter.
Example API request body with the default parameters:
{
"model_version": "reactions-v3.3",
"model_params": {
"platform_format": "tv",
"age_range": "30-49",
"device_type": "desktop",
"gender_distribution": "mix",
"country": "US",
"view_month": "Sep",
"view_time": "morning"
}
}
For a general assessment of creative potential, it is recommended to include the country (based on asset language if applicable) and use default parameters for general population and placements.
parameters list
Param Name |
Param Value |
Default |
|---|---|---|
|
The target placing for the advertisement |
|
|
The age group of the target audience |
|
|
The device type where the advertisement is displayed |
|
|
The gender distribution of the target audience |
|
|
The country where the advertisement is displayed |
|
|
The month when the advertisement is displayed |
|
|
The daytime of the advertisement is displayed |
|
reactions-v3.4 (Default for video and image)
Experimental model.
Differences compared to reactions-v3.3:
Improvements on the aggregation algorithm on the media level.
parameters list
For the parameters and details see reactions-v3.3.
Example API request body with the default parameters:
{
"model_version": "reactions-v3.4",
"model_params": {
"platform_format": "tv",
"age_range": "30-49",
"device_type": "desktop",
"gender_distribution": "mix",
"country": "US",
"view_month": "Sep",
"view_time": "morning"
}
}
reactions-v4.0 / reactions-v4.1
Experimental model.
Differences compared to reactions-v3.4:
Improved norm logic applied to create indexes.
parameters list
For the parameters and details see reactions-v3.3.
Example API request body with the default parameters:
{
"model_version": "reactions-v4.0",
"model_params": {
"platform_format": "tv",
"age_range": "30-49",
"device_type": "desktop",
"gender_distribution": "mix",
"country": "US",
"view_month": "Sep",
"view_time": "morning"
}
}