This article explains how our Memory metric has been developed.
The Neurons Memory score shows how likely an ad is to be remembered after a quick glance. Unlike traditional metrics, it considers an implicit component of time to recall to give more reliable insights on what sticks.
This article will cover everything you need to know about the Neurons Memory feature, including:
- Methodology: How the score is calculated and what makes it valuable.
- Validation: The scientific approaches that lead to high accuracy and reliability.
AI Model: An overview of the advanced Neurons Memory AI Model that powers the score and provides actionable recommendations.
Feature Overview
The Memory prediction model is designed to predict how memorable an image or video asset will be to viewers. By using an advanced deep learning model trained on the Neurons' extensive database, which includes variations of visual stimuli across different media and contexts, this AI tool provides a Memory score that predicts ad recall probability. The Neurons Memory AI is essential for marketers looking to optimize their visual content for better ad memorability.
The Neurons Memory feature includes:
- AI recommendations
- Image & video insights
- Benchmarks
- Downloadable reports
- Heatmaps (for images)
- AOI level metrics (for images)
- Second-by-second view for videos
The Accuracy
The Memory score of the AI predicts the likelihood of an asset being remembered with a range from 0 to 100. This measurement considers both how accurately participants recall an asset and how fast they do so, introducing an implicit component to the explicit recall metric. The AI model achieved an accuracy of 90% during testing, ensuring a reliable prediction of ad recall despite some limitations.
The Methodology
The Neurons Memory AI utilizes a data-driven approach derived from the Visual Memory Game methodology developed by MIT. The original paper and subsequent pa
pers with this methodology have provided a solid research foundation and initial validation for our metric.
The study involves exposing participants to a rapid sequence of images. Participants must indicate when they've seen an image for the second time. This rigorous methodology captures both reaction times and correctness, thereby integrating cognitive engagement factors into the memory score.
The datasets used to train the model comprise a wide array of advertising materials, capturing various industry use cases — from print and Out-Of-Home (OOH) displays to digital platforms like websites and social media.
Neurons gathered data from over 7000 participants, age 18-55, split equally men and women. This dataset is enriched continuously to ensure that the model is always capturing the latest participant behavior and trends in advertising.
The Model
At the core of the Neurons Memory AI is a computer vision deep learning model employing the EfficientNetB2V2 framework. It boasts approximately 8.8 million trainable parameters, trained on over thousands of images with corresponding ground truth memory scores. By utilizing the GradCAM methodology, the AI generates heatmaps that highlight the elements most affecting the prediction, offering valuable insights into what aspects of visual content are most likely to stick in consumers' minds. Through its robust architecture, Neurons Memory AI provides actionable intelligence for crafting more memorable and impactful advertisements.
Predicting Memory on Video
For video assets, each frame is scored individually based on the image prediction model, which operates with the constraint of not utilizing sequential data from video flow or audio influence.
To further ensure that our memory metric is robust, we correlated the predicted scores on video with ground truth data from over 100 videos with scores for ad recall from more than 1000 participants, from our Memory Recall Test. This methodology is based on the most reliable methodologies for testing memory functions, adapted from proven approaches used to diagnose Alzheimer’s.
The Neurons AI Memory score shows strong alignment with the human testing results, suggesting that sustained memory engagement throughout the ad is a strong predictor of ad recall and overall memory performance. This alignment between human testing and AI predictions indicates that the Neurons AI Memory score is a reliable proxy for ad memorability in videos.
Validation of the memory score
The development of new metrics typically goes through several important steps, including rigorous validation tests. Our newly developed memory metrics underwent three separate validation experiments, at the end of which our memory prediction demonstrated:
- Reliability and validity in effectively assessing ad memory;
- Specificity and sensitivity, ensuring that each newly developed metric differs from one another, thus averting collinearity;
- Discriminatory ability, signifying its capacity to generate distinct scores that elucidate clear disparities between the various tested ads.