Hey there! Ever thought about what it would be like if all our Twitter feeds looked exactly the same? Imagine Netflix without personalized top picks, or Amazon not suggesting items based on your purchase history. Would you have binge watched those shows that you didn't know existed or bought those items that you didn’t know you needed? Personalization plays a significant role in user engagement. This lack of personalization might explain why some platforms, like Threads, don’t engage us as much as Twitter does.

The secret sauce behind the success of platforms like Netflix and Amazon is their powerful AI models, which tailor content to keep us hooked. So why do many Web3 applications, such as NFT marketplaces, show the same content to everyone?

Enter Persona. With rich blockchain data, Persona has developed personalization models based on a wallet’s on-chain activity. As part of Eth Denver Hackathon, we wanted to demonstrate the power of personalization. We experimented with two versions of an NFT marketplace—one with and one without personalization. The difference? Night and day.

Personalization in Action: We focused on two functionalities 

1. Top Picks For You‍
When the wallet is signed in, the "Top Picks For You" section predicts NFTs similar to those you've previously owned. To achieve this, Persona used Natural Language Processing (NLP) models on NFT descriptions. We tried two approaches:

SBERT Model: This method applied sentence transformers to NFT descriptions and computed cosine similarities.
TFIDF Approach: This method created a TFIDF (term frequency-inverse document frequency) matrix, used a TFIDFVectorizer, and computed cosine similarities.

The TFIDF approach outperformed the SBERT model, providing more relevant recommendations by emphasizing unique terms over common ones.

‍
‍2. People Who Bought This Also Bought That

When the user clicks on an NFT, the user sees not only the NFT’s details but also recommendations based on what others bought alongside it. Persona’s model tracked how often pairs of NFTs were owned by the same wallet, ranking these pairs to suggest the most frequently bought-together NFTs.

Building these models wasn't without hurdles. The two main challenges were:
‍
1. Data Availability: Lacking labels for accuracy testing, Persona scraped category data from popular NFT marketplaces to create a ground truth table, comparing collections within the same category.
‍2. Data Size: With more than half a million wallets on the Ethereum transaction table, managing this vast dataset was daunting. Persona downsampled the data by focusing on recent transactions by active wallets.

Why does personalization matter for an application? 

- Enhanced User Experience: Users see content relevant to their interests, making it easier to find appealing NFTs.
- Increased Engagement/Sales: A personalized experience keeps users on the platform longer, potentially leading to higher sales.
- Efficient Search: Personalization improves search and discovery, offering more relevant results based on user behavior.
- Brand Loyalty: A tailored experience builds stronger relationships with users, fostering loyalty.
- Competitive Advantage: Personalized experiences set platforms apart, attracting more users and establishing market leadership.

To elevate user engagement in Web3, applications must embrace personalization, just like our favorite mainstream platforms do. Leveraging AI technology can transform the user experience in the Web3 ecosystem, making it as captivating and user-friendly as Instagram, Twitter, Netflix, or Amazon.

Persona has mastered building AI models trained on blockchain data for ad targeting. We extended our AI expertise to personalization in Web3 as well, setting a new standard in AI technology in web3.