Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Long-Term Ad Memorability: Understanding & Generating Memorable Ads

Harini SI, Somesh Singh, Yaman K Singla, Aanisha Bhattacharyya, Veeky Baths, Changyou Chen, Rajiv Ratn Shah, Balaji Krishnamurthy

TL;DR

The work tackles long-term ad memorability by introducing LAMBDA, the first large-scale LT memorability dataset for multimodal ads, and presents Henry, a multimodal memorability model that fuses visual, cognitive, and world knowledge via LLM grounding. It demonstrates strong state-of-the-art performance across multiple datasets and strong zero-shot generalization, while identifying that short-term memory can benefit from STM signals for LT memorability. To push beyond prediction, the authors introduce SEED and UltraLAMBDA to enable scalable memorability-driven ad generation, achieving a 44% average improvement in memorability over originals and showing GPT-4/3.5 baselines lag behind dedicated Henry-SEED generations. The open-source LAMBDA and UltraLAMBDA datasets, together with the Henry framework, provide a foundation for future research in memorability-aware advertising and synthetic data-driven content generation with practical marketing impact.

Abstract

Despite the importance of long-term memory in marketing and brand building, until now, there has been no large-scale study on the memorability of ads. All previous memorability studies have been conducted on short-term recall on specific content types like action videos. On the other hand, long-term memorability is crucial for the advertising industry, and ads are almost always highly multimodal. Therefore, we release the first memorability dataset, LAMBDA, consisting of 1749 participants and 2205 ads covering 276 brands. Running statistical tests over different participant subpopulations and ad types, we find many interesting insights into what makes an ad memorable, e.g., fast-moving ads are more memorable than those with slower scenes; people who use ad-blockers remember a lower number of ads than those who don't. Next, we present a model, Henry, to predict the memorability of a content. Henry achieves state-of-the-art performance across all prominent literature memorability datasets. It shows strong generalization performance with better results in 0-shot on unseen datasets. Finally, with the intent of memorable ad generation, we present a scalable method to build a high-quality memorable ad generation model by leveraging automatically annotated data. Our approach, SEED (Self rEwarding mEmorability Modeling), starts with a language model trained on LAMBDA as seed data and progressively trains an LLM to generate more memorable ads. We show that the generated advertisements have 44% higher memorability scores than the original ads. We release this large-scale ad dataset, UltraLAMBDA, consisting of 5 million ads. Our code and the datasets, LAMBDA and UltraLAMBDA, are open-sourced at https://behavior-in-the-wild.github.io/memorability.

Long-Term Ad Memorability: Understanding & Generating Memorable Ads

TL;DR

The work tackles long-term ad memorability by introducing LAMBDA, the first large-scale LT memorability dataset for multimodal ads, and presents Henry, a multimodal memorability model that fuses visual, cognitive, and world knowledge via LLM grounding. It demonstrates strong state-of-the-art performance across multiple datasets and strong zero-shot generalization, while identifying that short-term memory can benefit from STM signals for LT memorability. To push beyond prediction, the authors introduce SEED and UltraLAMBDA to enable scalable memorability-driven ad generation, achieving a 44% average improvement in memorability over originals and showing GPT-4/3.5 baselines lag behind dedicated Henry-SEED generations. The open-source LAMBDA and UltraLAMBDA datasets, together with the Henry framework, provide a foundation for future research in memorability-aware advertising and synthetic data-driven content generation with practical marketing impact.

Abstract

Despite the importance of long-term memory in marketing and brand building, until now, there has been no large-scale study on the memorability of ads. All previous memorability studies have been conducted on short-term recall on specific content types like action videos. On the other hand, long-term memorability is crucial for the advertising industry, and ads are almost always highly multimodal. Therefore, we release the first memorability dataset, LAMBDA, consisting of 1749 participants and 2205 ads covering 276 brands. Running statistical tests over different participant subpopulations and ad types, we find many interesting insights into what makes an ad memorable, e.g., fast-moving ads are more memorable than those with slower scenes; people who use ad-blockers remember a lower number of ads than those who don't. Next, we present a model, Henry, to predict the memorability of a content. Henry achieves state-of-the-art performance across all prominent literature memorability datasets. It shows strong generalization performance with better results in 0-shot on unseen datasets. Finally, with the intent of memorable ad generation, we present a scalable method to build a high-quality memorable ad generation model by leveraging automatically annotated data. Our approach, SEED (Self rEwarding mEmorability Modeling), starts with a language model trained on LAMBDA as seed data and progressively trains an LLM to generate more memorable ads. We show that the generated advertisements have 44% higher memorability scores than the original ads. We release this large-scale ad dataset, UltraLAMBDA, consisting of 5 million ads. Our code and the datasets, LAMBDA and UltraLAMBDA, are open-sourced at https://behavior-in-the-wild.github.io/memorability.
Paper Structure (36 sections, 1 equation, 12 figures, 8 tables)

This paper contains 36 sections, 1 equation, 12 figures, 8 tables.

Table of Contents

  1. Introduction
  2. LAMBDA Protocol, Study & Insights
  3. Video Collection
  4. Annotation Protocol
  5. What makes an Ad memorable?
  6. Predicting Ad Memorability
  7. Encoding Multimodal Content
  8. Two-Stage Training
  9. Results and Discussion
  10. Ablation of Data and Architectural Choices on Memorability Prediction
  11. Generating Memorable Ads
  12. Evaluation
  13. Results
  14. Ablation of Data and Architectural Choices on Memorable Ad Generation
  15. Conclusion
  16. ...and 21 more sections

Figures (12)

  • Figure 1: Correlations between content factors (a-d), interaction factors (e-g), and customer behavior factors (h-j) with memorability on LAMBDA samples. While emotion has a high correlation with memory, other content factors do not have much correlation. Further, while there is little correlation between the order of videos seen and memorability; with time, participants' memory of the videos shows a forgetting trend. Video popularity, as measured by YouTube likes/views, shows a slight positive correlation with memory. Average brand relevance has a strong positive correlation with memory, with top sectors being remembered as food, entertainment, and tech. Speech, silence and music have little effect with silence having the highest positive correlation with recall. Silence ratio is measured as the percentage of silence in a video, similarly for music and speech.
  • Figure 2: The graph shows the relationship of the year the ad is uploaded on youtube vs the recall.
  • Figure 3: Predicting memorability by encoding visual information (via visual encoder EVA-CLIP), cognitive concepts (via verbalization module), and world knowledge (through fine-tuned Llama). We instruction fine-tune the combined model end to end to predict user memorability. Snowflake and fire symbols denote the frozen and unfrozen parts of the architecture.
  • Figure 4: Henry-SEED Prompt: Generate the detailed description of a 30-second memorable advertisement titled "Brainly Keep Learning 30sec Final 16x9" for the brand Brainly. Link to the original ad: https://www.youtube.com/watch?v=kytRXyWXivU Original Memorability score: 85. Memorability score of Generated Ad: 99.
  • Figure 5: Henry-SEED Prompt: Generate the detailed description of a 50 second memorable advertisement titled "Shining a Light on Women’s Rights | The Truth Has a Voice | The New York Times" for the brand The New York Times Link to the original ad: https://www.youtube.com/watch?v=bPblzhUzTeg Original memorability score: 65. Memorability score of Generated Ad: 91.
  • ...and 7 more figures