Twitter Bot Detection: via Retweet Correlation

Project uses unsupervised learning via clustering techniques to detect Twitter bot accounts.

Four Approachs:

• K-means++
  • Build Retweet matrix -> NMF -> K-means++ -> Detect high correlation clusters
• Spectral Clustering
  • Build Retweet matrix -> NMF -> Pairwise Correlation -> Spectral Clustering -> Detect high correlation clusters
• Clustering via Pairwise Correlation (using NMF)
  • Build Retweet matrix -> NMF -> Pairwise Correlation -> build graphs with only highly correlated users -> run BFS to cluster
• Clustering via Pairwise Correlation (using Node-to-vec)
  • Build Retweet matrix -> Node-to-vec -> Pairwise Correlation -> build graphs with only highly correlated users -> run BFS to cluster

Data and Evaluations

Clustering Results on Sample Dataset