ZuSocial @Istanbul - Nov 15th Daily Roundup
1/ Data science & Machine learning 101
ZuSocial Hacker: Akira @akirawuc
What is data and machine learning?
Data is actually relative to different people. The data you choose to represent / simplify the world / task, directly reflect what you are seeking for. (Personally, data is piece of information with structured form.)
Machine learning involves the process of extracting/ simplifying the world into data, finding patterns through the data, and fitting it to the real world.
The DL/ ML/ AI universe and difference between rule-based system, classic ML, and representation learning
Deep learning, Machine learning, AI, etc. there are too many terms and below graph shows a clear relation between different concepts.
Compared to rule-based system, Classic ML mainly includes mapping from features before the output. While Representation learning system extracts features from raw data or use multiple layers of extracted features to replace hand-designed features.
What is a learning algorithm?
It is defined as a computer program that improves its performance at tasks in T, as measured by P, improves with experience E.
2/ XMTP workshop - The secure messaging network for Web3
Guest: Fabri Guespe @fabriguespe from XMTP @xmtp_
The problem XMTP tries to tackle
Developers can't reach users
Creators can't reach their audiences
Users can't reach each other
XMTP's solution and use cases
XMTP (Extensible Message Transport Protocol) is an open protocol, network, and standard for secure, private web3 messaging. It serves as a base protocol. Developers can build with XMTP SDKs to provide messaging between blockchain accounts in their apps.
XMTP's use cases include:
Interoperable inboxes
Coinbase wallet, Converse, Lenster, all use XMTP in the backend
Other use cases
CRM tools for creators to reach NFT holders
Community-driven social media super app, e.g. Orb
Bots that can handle complex interactions like smart contract
Global payments over messaging, e.g. Coinbase
Timely alerts that drive action, e.g. Snapshot
Matchmaking based on on-chain interactions, e.g. Converse
Working on protocol v3
Some features under development include:
Double-ratchet messaging for foward secrecy & post-compromise security
libxmtp for transparent protocol development
Group chats
Consent features
Decentralization
Relevant link
Official site: https://xmtp.org/
3/ Lens workshop - Building full stack DeSo apps
Guest: Nader Dabit @dabit3 from @LensProtocol
What is Lens?
Lens offers a protocol and a suite of tools and APIs for building social apps or integrating social features into existing apps. Social app involves creating a profile, following, publishing, viewing feed of other comments, and recommendation algo
Lens V2
Lens launched V2 on Nov 13th and V2 will include several new features:
Publication metadata: V2 introduces a native way to enable Quoted Publications
Separate profiles and handles: ERC 6551 powers profile as a wallet and profile becomes the core identity for all actions
Smart posts (previously called Open actions): V2 facilitates smart contract interaction on any protocol, smart contract, or network with an existing publication, incl. buying nft, staking, voting etc.
New referral system: V2 allows rewarding those that helped to discover a publication, reward original posters for any activity that happens after, and reward apps and UIs used to interact with Lens, and help discovery
Profile manager: V2 allows delegation of social actions to a different wallet
How does Lens work
Set of smart contracts
Deployed on polygon
GraphQL API for quering capabilities
Gasless transactions via relayer for whitelisted front ends
Dispatcher for seamless interaction
High quality dev experience
Relevant link
Launch of Lens V2: https://mirror.xyz/lensprotocol.eth/Kc1AkZYWX1kR6XCsOf0bpU3pbXi0Juayt1tPyPicv4c
Official site: https://www.lens.xyz/
4/ Recommendation system workshop
ZuSocial Hacker: Yassime @YassineLanda
Recommendation system in current social media
Benefit of the recommendation system: It helps grow the network more dense in a faster way
Disadvantage: It can be very addictive for users.
How does feed ranking actually work?
2010 formula:
Priority (user, item) = affinity (user, poster) * weight [item, type] / item.age
2017 formula:
MSI (user, item) = affinity (user, poster) ∑ P (user, item, int-type) weight [int - type]
MSI: meaningful social interaction, which prioritizes interactions, such as comments and likes, between friends and family. The idea was to give more weight to the posts and engagements of people that social media platform thought are closest to users.
Personalization best practice
It's all about interactions! Best practice includes:
Quantify interactions
Predicting interest
Personalization
Limitation of data-based algorithm
There are certain aspects of the impact/ value, especially long-term ones, difficult to be represented by data currently, thus hard to measure.