SoMin.ai: Bridging Psychographics and Generative AI for the Future of Advertising
SoMin.ai: Personality-Driven Content Generation Platform
The paper introduces SoMin.ai, the world's first personality-driven platform for marketing content generation. It integrates a "BiGRU+VGG" multi-view personality profiling framework with StyleGAN-based image generation to create tailored social media advertisements based on MBTI personality traits.
TL;DR
SoMin.ai is a pioneering platform that automates the creation of marketing content by aligning it with the viewer's personality. By combining a BiGRU+VGG profiling framework with StyleGAN generation, it moves beyond mass-market ads to hyper-personalized creatives tailored to MBTI traits, significantly reducing manual design effort while increasing ad relevance.
The Scalability Gap in Digital Marketing
In the modern programmatic advertising landscape, platforms like Facebook and Google support Dynamic Content Optimization (DCO)—a feature that can handle thousands of ad variations. However, human creative teams are a bottleneck. They rely on "gut feeling" and can only produce a handful of assets, leaving the potential of DCO untapped.
Furthermore, most AI generators focus on "what" to generate (e.g., a car in a city) rather than "how" to appeal to a specific person (e.g., a car image that appeals to an Introverted/Thinking type vs. an Extroverted/Feeling type).
Methodology: The "BiGRU+VGG" Fusion
SoMin.ai uses a two-pronged technical approach to solve the personalization and generation challenges.
1. Multi-View Personality Profiling
The system analyzes a user's Instagram or Twitter history to predict their MBTI (Myers-Briggs Type Indicator).
- Text Processing: A Bi-directional Gated Recurrent Unit (BiGRU) processes post captions to capture linguistic nuances.
- Image Processing: A VGG16 network extracts visual features from the user's most recent photos.
- Fusion: The latent representations are combined through a direct-product operation to form a comprehensive psychographic profile.
2. Personality-Driven Generation
Once a user's MBTI is inferred, the system performs cross-domain recommendation. It retrieves successful content styles preferred by similar personality types within a specific industry (e.g., Fashion, Fast Food, or Automobile). These styles are then used as parameters for a pre-trained StyleGAN to generate new, branded variations that maintain industry compliance while maximizing personal appeal.
Experimental Results and Feedack Loops
The platform's profiling framework was tested against standard machine learning models like SVM and LightGBM. The results demonstrated that fusing text and images is superior to using either alone.
| Model | EI (Introversion/Extroversion) | SN (Sensing/iNtuition) | TF (Thinking/Feeling) | JP (Judging/Perceiving) |
|---|---|---|---|---|
| Our Approach (BiGRU+VGG) | 0.83 | 0.57 | 0.64 | 0.60 |
A unique feature of SoMin.ai is its Feedback Loop. Users can rate generated content, and these ratings act as a reward/penalty mechanism. Successful "styles" are prioritized and augmented in the next training cycle, while unsuccessful ones are pruned, allowing the AI to evolve with real-world consumer tastes.
Critical Insight & Future Outlook
SoMin.ai represents a shift from Demographic Targeting (age, location) to Psychographic Personalization. By automating the production of thousands of ad variations, it allows brands to leverage the full power of DCO.
Limitations: While powerful, MBTI is a simplified psychological model. Future iterations could benefit from using the Big Five (OCEAN) traits or more fluid behavioral embeddings. Additionally, moving from GANs to modern Latent Diffusion Models could significantly improve the visual fidelity and textual coherence of the generated ads.
In conclusion, SoMin.ai proves that AI's greatest value in marketing isn't just "generating more stuff," but "generating the right stuff" for the right mind.