ClearFeed

Personalized social feed where you’re not the product. Solution: Personalised social feed algorithm in an NFT that users store in their wallet. To get better content recommendation on Lens.

Comment 1

Inspiration

Feed algorithms rule society in a hidden black box manner. We want to make these algorithms more transparent and clear, thus we are ClearFeed.

Another inspiration is the sprit of open source and Web3. Additionally, current web2 social medias make a profit on selling their users data and provide no insight for the user to their content recommendation algorithms.

What it does

Lets users create their own feed algorithms as an NFT. Therefore, users own their algorithm, can change it however they want, sell it and so forth. This algorithm is based on the users social graph on the Lens protocol. Additionally, users can enhance their algorithm by doing a personality test and rating different posts. This data is fully private and users can share it however they please.

How we built it

Users can mint an NFT which stores a simple content recommendation algorithm. This consists of several different parameters. Such as the users social graph on Lens and sentiment analysis of posts. Content recommendation is remarkably improved by analysing a users second and third order followings. In short, this is done by:

  1. listing all the users second & third order follows
  2. and then ranking them based on how many common follows they have
  3. recommending content from the profiles with the highest ranking

The personality test is implemented within a zk-Snark. Users submit their data to the snark which trains a Passive-Agressive regression asynchronously. This lets users to enhance their feed with sensitive data without ever revealing that data to the public.

Different Lens front-ends can read the NFT and generate the feed belonging to that users. Currently we have implemented this on our site. Everything from minting to feed front-end is implemented.

Challenges we ran into

Our vision is to provide this NFT as an infrastructure block to Lens front-ends. The challenge is that the integration requires a lot of cooperation with all different front-ends and creating a standardised solution. This is not as straightforward as we initially thought -- as it's not only a technical problem but also a social and collaborative problem.

Another important challenge is querying the blockchain for data to train the content recommendation algorithm. Our current solution uses the Lens GraphQL API. It is however not efficient for graph traversing & graph calculus. The aim is to build our own blockchain indexer on AWS. Using their graph database to address the issue. However building this indexer has been a more cumbersome and time consuming than initially estimated.

Accomplishments that we're proud of

We are proud of the positive feedback from 10 lens users. They all minted their NFTs and noticed a significant improvement in the feed.

What we learned

It's impressive how large the improvements are with seemingly small enhancements to the feed algorithm. And additionally we've learned tons on how to build on Lens, index blockchains and develop zk-Snarks.

What's next for ClearFeed

  • Collaboration with a large player in the Lens ecosystem: Orb. We'll airdrop feed NFTs to Orb users, allowing them to enhance their feed on Orb.
  • This way we will achieve our goal of getting our 100 first users.
  • Complete building our own indexer

Built With

Share this project:

Updates