A Go implementation of community detection for social graphs. Identifies clusters of similar users based on their follow relationships, with a focus on grouping users around influencers.

SimClusters analyzes a social network's follow graph to:

• Detect influencers (high-follower accounts)
• Group users into communities based on who they follow
• Use MinHash sketches for efficient similarity computation

go build ./cmd/simmy

To start, you will need a follow graph in the following csv format:

id,author,subject
1234,2345,3456

Which says the follow relation with the ID 1234 is from user 2345 to user 3456. The program works best when this data is split up across multiple files so it can be processed in parallel. The ID field is not super important for the actual clustering but does help for deduplicating new follows added to the dataset later.

./simmy run --checkpoint ./checkpoints data/*.csv

Resume from a previous checkpoint:

./simmy run --from-checkpoint ./checkpoints

Interactively explore user communities:

./simmy query --mapping mapping.json --influencers influencer_communities.json

./simmy export-csv --mapping mapping.json --output clusters.csv

Find posts similar to cluster centroids:

./simmy cluster-posts --limit 100 --days 5

The clustering produces two main files:

• mapping.json - Maps users to their cluster memberships
• influencer_communities.json - Maps influencers to their communities