NFT Recommendation System

Proposals Round 4
#1

Proposal in One Sentence:

Utilize ML techniques such as collaborative filtering on our team’s dataset (NFT metadata and image similarity data from millions of NFTs) to create a Proof-of-Concept for NFT recommendation based on NFTs held in a dApp user’s wallet.

Proposed PoC

NFTs held on public blockchains contain a wealth of digital asset data which lends itself to ML-based analysis.

One interesting use case for this analysis is the clustering of similar data to build an NFT Recommendation system. Such a system could employ well-established ML models like collaborative filtering and similar techniques to exploit the hidden connections between NFTs and improve NFT asset and project discovery.

The model can then be used to perform inference thusly:

  • Compute recommendations within the same type of NFTs
    Eg. given a Profile Picture collection, it will be able to recommend other collections relevant to the user.

  • Compute recommendations across data types
    Eg. Owners of a Netflix subscription utility token could be interested in a “Stranger things” inspired collections

Proposed Approach

The overall idea is simple and aligns with consolidated ML workflows:

  1. Build a dataset by dumping the whole ledger of a specific blockchain, creating a wallet to tokens (i.e. NFT, FTs, etc.) map
  2. Pre processing data (filter out unwanted data and noise)
  3. Train and test model
  4. Offer recommendations based on tokens held in a user’s wallet

As we’re building a multi-chain suite of services, an important requirement is to ensure that model pre-processing should be built in a blockchain agnostic way.

As newcomers to the Algovera community who have nonetheless put a great deal of work and consideration into solving problems around NFT authentication and discovery (https://argusnft.medium.com/; Argus NFT · GitHub ) our interest is to collaborate with a data scientist (or scientists) in the community to test our approaches and presumptions with regard to the creation of the NFT recommendation model. We can and will self-fund our current team’s efforts on the proposal; we’ll achieve the proposal deliverables regardless of whether a DS from the community makes significant contributions to the effort; and any grant funding awarded will go exclusively to compensate a DS from the community for their efforts in helping to create the model. And of course, depending on their level of contribution, we can discuss additional compensation and/or continued collaboration beyond the scope of this proposal.

Essentially, we see this grant as an opportunity to:

  1. Create a PoC for a novel and useful solution within the NFT space
  2. Attempt to identify and develop a relationship with an individual data scientist (or squad of data scientists) who enjoys the space we’re working in, and is eager to experiment on different approaches to this challenge. Ideally, we’d like to form a solid working relationship so we can collaborate on additional iterations and models in the future.

The members of the ArgusNFT team supporting this proposal are:

  • Giovanni Gargiulo - CEO and ML Engineer. Principal Software Engineer with 15+ years of commercial experience. DevOps advocate with strong experience in Machine Learning, distributed systems and cloud. Open Source Developer.

  • Anthony Capitan - COO. Entrepreneur with 7 years business development experience, 15 years sales, marketing and project management experience.

  • Jaime Caso - Full Stack and Mobile Engineer. Cardano Wallet Developer.

  • Kurt Hartmann - Lead Backend Engineer. 20+ years experience as a developer and architect working as an IT consultant for companies in the financial, health, and retail industries. Areas of expertise include modernizing and integrating legacy systems, ETL batch jobs, building API data sources, and implementing workflow patterns using micro-services and messaging. Kurt also has a solid background in database design and web services.

1 Like