Diamond DAO knowledge graph specialist Omar outlines a framework for applying network analysis algorithms to the Chainverse knowledge graph using Neo4J’s Graph Data Science (GDS) library.

Over the past few weeks, we’ve seen growing interest in applying network analysis to uncover “clusters” of Web3 communities.

https://twitter.com/rafathebuilder/status/1531876726120275968?s=20&t=UrPLwacH-JRJe_wd55faTg

Internally, Diamond DAO community members have also expressed interest in leveraging Neo4J’s Graph Data Science (GDS) framework to experiment with advanced analytic techniques, ranging from label propagation algorithms to node embeddings.

Generally, analysts start with a research question (i.e. what are the most central DeFi DAOs) and then choose the right tools for the job.

In our case, where community stakeholders have many different research questions, it makes more sense to start with an analytic framework that allows researchers to quickly test the network analysis techniques appropriate for their research question/s.

To start, the Chainverse Network Analysis (CNA) framework will generally contain four steps:

1. Identify features of DAOs (eg.,. membership overlap, voting) to feed into our network analysis algorithm.

2. Create a “projection” of the Chainverse knowledge graph which removes everything except the set of features we want the network analysis algorithm to analyze (e.g., DAOs and their membership overlap with other DAOs)

3. Identify the appropriate Neo4J GDS algorithms and write code to run them

4. Interpret results of Step 3, iterate based on findings.

Right now, Chainverse covers two networks of DAOs: DAOs that use the DAOhaus framework and DAOs that use Snapshot for voting.

DAOHaus and Snapshot provide us two separate operating models for DAOs which is reflected in the separate data model for each. Although slightly different, the two networks are able to provide us methods for analyzing DAOs across the two networks with the same lens depending on the features that we want to model.

DAOhaus Data Model

Snapshot Data Model

Our initial “feature” for analysis will be a membership list capturing overlap across DAOs.

Here are the Cypher (Neo4J query language) queries we used to create those relationships:

For initial analysis let’s focus on the DAOs with the highest centrality. We should see DAOs that everyone is pretty familiar with at the top of this list

It’s interesting to see Metacartel, despite its relatively small number of members, appear as one of the most central DAOs in the Web3 ecosystem.

Next, we run the Louvain community detection algorithm to identify “clusters” of DAOs. We see that the DAO landscape is extremely fragmented, with 5,523 DAOs breaking down into 2,190 clusters.

Let’s check out these community clusters.

The size of the communities drastically drops off and even the 10th largest community seems only have 14 DAOs:

Let’s look into the clusters with the most members.

ENS voters dominate the largest cluster, which isn’t surprising -- a lot of people claimed the ENS airdrop.

The second largest cluster is dominated by members of Polygon ecosystem DAOs, as well as some DeFi (i.e. Aave, Balancer) and public goods (i.e. Gitcoin) focused DAOs.

The third largest cluster is dominated by members of “generic” DeFi DAOs.

The fourth largest cluster is dominated by Metacartel ecosystem DAOs and prominent social token DAOs.

An interesting start!

There are lots of opportunities to refine and iterate on this analysis:

• We generate more “thematic” clusters by including tag entered by Chainverse users (i.e. “public-goods”, “developer-tooling”) in our feature set

• We could identify key DAO contributors by adding member wallets to our graph projection and re-running the centrality algorithm and including relevant attributes (i.e.whether they had summoned DAOs, were multi-sig signers, had passed proposals, etc) as features

• We could add NFT transfer data to experiment with recommendation algorithms

Lots to explore!

If you’re interested in the intersection of network analysis and Web3 communities, reach out to @ozhar or @c_lemp on Twitter.

Diamond DAO knowledge graph specialist Omar outlines a framework for applying network analysis algorithms to the Chainverse knowledge graph using Neo4J’s Graph Data Science (GDS) library.

Over the past few weeks, we’ve seen growing interest in applying network analysis to uncover “clusters” of Web3 communities.

https://twitter.com/rafathebuilder/status/1531876726120275968?s=20&t=UrPLwacH-JRJe_wd55faTg

Internally, Diamond DAO community members have also expressed interest in leveraging Neo4J’s Graph Data Science (GDS) framework to experiment with advanced analytic techniques, ranging from label propagation algorithms to node embeddings.

Generally, analysts start with a research question (i.e. what are the most central DeFi DAOs) and then choose the right tools for the job.

In our case, where community stakeholders have many different research questions, it makes more sense to start with an analytic framework that allows researchers to quickly test the network analysis techniques appropriate for their research question/s.

To start, the Chainverse Network Analysis (CNA) framework will generally contain four steps:

1. Identify features of DAOs (eg.,. membership overlap, voting) to feed into our network analysis algorithm.

2. Create a “projection” of the Chainverse knowledge graph which removes everything except the set of features we want the network analysis algorithm to analyze (e.g., DAOs and their membership overlap with other DAOs)

3. Identify the appropriate Neo4J GDS algorithms and write code to run them

4. Interpret results of Step 3, iterate based on findings.

Right now, Chainverse covers two networks of DAOs: DAOs that use the DAOhaus framework and DAOs that use Snapshot for voting.

DAOHaus and Snapshot provide us two separate operating models for DAOs which is reflected in the separate data model for each. Although slightly different, the two networks are able to provide us methods for analyzing DAOs across the two networks with the same lens depending on the features that we want to model.

DAOhaus Data Model

Snapshot Data Model

Our initial “feature” for analysis will be a membership list capturing overlap across DAOs.

Here are the Cypher (Neo4J query language) queries we used to create those relationships:

For initial analysis let’s focus on the DAOs with the highest centrality. We should see DAOs that everyone is pretty familiar with at the top of this list

It’s interesting to see Metacartel, despite its relatively small number of members, appear as one of the most central DAOs in the Web3 ecosystem.

Next, we run the Louvain community detection algorithm to identify “clusters” of DAOs. We see that the DAO landscape is extremely fragmented, with 5,523 DAOs breaking down into 2,190 clusters.

Let’s check out these community clusters.

The size of the communities drastically drops off and even the 10th largest community seems only have 14 DAOs:

Let’s look into the clusters with the most members.

ENS voters dominate the largest cluster, which isn’t surprising -- a lot of people claimed the ENS airdrop.

The second largest cluster is dominated by members of Polygon ecosystem DAOs, as well as some DeFi (i.e. Aave, Balancer) and public goods (i.e. Gitcoin) focused DAOs.

The third largest cluster is dominated by members of “generic” DeFi DAOs.

The fourth largest cluster is dominated by Metacartel ecosystem DAOs and prominent social token DAOs.

An interesting start!

There are lots of opportunities to refine and iterate on this analysis:

• We generate more “thematic” clusters by including tag entered by Chainverse users (i.e. “public-goods”, “developer-tooling”) in our feature set

• We could identify key DAO contributors by adding member wallets to our graph projection and re-running the centrality algorithm and including relevant attributes (i.e.whether they had summoned DAOs, were multi-sig signers, had passed proposals, etc) as features

• We could add NFT transfer data to experiment with recommendation algorithms

Lots to explore!

If you’re interested in the intersection of network analysis and Web3 communities, reach out to @ozhar or @c_lemp on Twitter.