/Social-network-codes

Primary LanguagePython

Social network codes

This has all the codes used in Social Network paper. Working with undirected networks. D is symmetric. Bounds for the missing entries are also calculated using observed entries. The computed lower and upper bounds help Low-rank Matrix Completion (LMC) to estimate within the given range instead of -inf to +inf. Missing entries in the matrix are recovered using Low-rank Matrix Completion.Results are compared for with and without bounds scenario.

Approach

D >> VC >> delete a percentage of entries >> Recover VC
Where
D: complete distance matrix
VC: Virtual coordinate matrix = Only a few columns are selected from the NxN distance matrix
Percentage deletion: 20, 40, 60, 80, 90% of entries are deleted from the VC matrix
We use Low-rank Matrix Completion integrated with bounds for missing entries to estimate missing node pair distances.

Performance evaluation metrics:

  1. Absolute hop distance error
  2. Mean error
  3. Count of non exact entries in the predicted distance matrix
  4. Clustering coefficient (after recovery)
  5. Average node degree (after recovery)

Data used

Undirected social networks were used:

  1. Facebook (file name: Original_Dist_nw1)
  2. Collaboration (file name: Original_Dist_nw2)
  3. Enron Email (file name: Original_Dist_nw3) Distance matrices of these networks are given in this repository.