TCEM: Two-Phase Co-exposure maximization Algorithm
Contact author: Sijing Tu sijing@kth.se
-
Citation information: Tu, S., Aslay, C., Gionis, A. (2020). Co-exposure Maximization in Online Social Networks. In Advances in Neural Information Processing Systems (NeurIPS)
-
Copyright: Redistribution and use in source and binary forms, with or without modifications, are permitted for academic purposes, provided that the proper acknowledgements are done.
Acknowledgement: Cigdem Aslay.
Compilation
make -f Makefile
Configuration of input files and parameters
config.txt:: specify the output folder, set &epsilon = 0.2, and l = 1. It is possible to modify these parameters.
Input file of the graph
Input directed graph file with one line per arc followed by two probabilities, the node ids should be mapped to 0 to n-1 where n is the total number of nodes.
format: node_u node_v p^1_uv p^2_uv
compare file
Indicate the selected red nodes ids and blue nodes ids. The file should contain two lines; the first
line begins with red nodes: , and the second line begins with blue nodes: . The node ids should separate by commas.
format:
red nodes: x1 x2 ... xk_r
blue nodes: y1 y2 ... yk_b
Running from command line
There are two possible command types, they have the common part e.g. ./main_TCEM -c config.txt -x data/karate.txt. The flag -c is followed by a configure file, in this project, it is config.txt; the flag -x is followed by the file of input graph.
For the first one you need to specify the number of red nodes and blue nodes; the number of red nodes and blue nodes are followed by -r and
-b respectively.
./main_TCEM -c config.txt -x data/karate.txt -r 2 -b 4
For the second one you need to specify the compare file, which is followed by -p.
./main_TCEM -c config.txt -x data/karate.txt -p comparefolder/compare.txt
Note: The implementation also contains the code for baselines used. Comment out line 88-93 of allocator.cc to receive also the results for baselines.
Reference
Aslay, C., Galbrun, E., Matakos, A., Gionis, A. (2018). Maximizing the diversity of exposure in a social network. IEEE International Conference on Data Mining (ICDM).
Note
Update in Oct 2020,
We made a modification on lambda in our code, changing from
theta = (2 + 2/3.0 * epsilon) * n * (ell * log(n) + log(2) + logP(n,k_r,tao)) / (epsilon * epsilon * lb);
to
theta = (8 + 4/3.0 * epsilon) * n * (ell * log(n) + log(2) + logP(n,k_r,tao)) / (epsilon * epsilon * lb);.