SUBSIM

This project implements the SUBSIM and HIST algorithms for the following paper:

• Qintian Guo, Sibo Wang, Zhewei Wei, and Ming Chen. 2020. Influence Maximization Revisited: Efficient Reverse Reachable Set Generation with Bound Tightened. In SIGMOD.

Please cite the paper if you would like to use the code.

The code is developed on the basis of the project: https://github.com/tangj90/OPIM.

Author: Qintian Guo ([email protected])

Compile

make

The option -std=c++1x should be included.

Options

• -func=string

  Specify the function to execute. Possible values are in the following:

  • format: format the graph
  • im: influence maximization

• -gname=string

  Specify the name of the input graph.

• -dir=string

  Specify the directory of the input graphs. The default directory is graphInfo in the current directory.

• -outpath=string

  Specify the directory for saving the output. The default directory is result in the working directory

The following options are used with -func=format.

• -pdist=string

  Specify the probability distribution.

  • wc [default]: WC setting (default). The weight of the edge (u, v) is set to p(u, v) = 1/indeg(v).
  • uniform: Uniform setting. All the edges are assigned a given weight specified by -pedge=float.
  • skewed: Skewed distribution. The weights follow a skewed distribution.

• -wcvariant=float

  Specify the weight sum in the wc variant setting. For example, if -wcvariant=1.2, it means the weight of the edge (u, v) is set as min{1, 1.2/indegree(v)}. This option works for -pdist=wc only.

• -pedge=float

  Specify the probability for all the edges in the uniform setting. The default value is 0.1. This option works for -pdist=uniform only.

• -skew=string

  Specify which skewed distribution is used. This option works for -pdist=skewed only.

  • exp[default]: Exponential distribution.
  • weibull: Weibull distribution.

The following options are used with -func=im.

• -seedsize=integer

  Specify the size of the seed set. The default value is 50.

• -eps=float

  Specify the error threshold such that subsim returns (1-1/e-eps)-approximate solution. The default value is 0.1.

• -delta=float

  Specify the failure probability. It means that the returned solution is valid with at least 1 - delta probability. The default value is 1/n.

• -vanilla=integer

  Specify the RR set generation method.

  • 0 [default]: the RR set is generated by the subsim method.
  • 1: the RR set is generated by the vanilla version.

• -hist=integer

  Specify whether the HIST algorithm is invoked.

  • 0 [default]: the HIST algorithm is not invoked.
  • 1: the HIST algorithm is invoked.

Format the graph

Before running influence maximization algorithms, please format the graph first.

Examples:

//the wc setting and wc variant setting
$ ./subsim -func=format -gname=facebook -pdist=wc
$ ./subsim -func=format -gname=facebook -pdist=wc -wcvariant=1.2

//the uniform setting
$ ./subsim -func=format -gname=facebook -pdist=uniform -pedge=0.01

//the skewed setting with exponential or weibull distribution
$ ./subsim -func=format -gname=facebook -pdist=skewed -skew=exp
$ ./subsim -func=format -gname=facebook -pdist=skewed -skew=weibull

Influence maximization

Examples:

//use the subsim method to sample RR sets
$ ./subsim -func=im -gname=facebook -seedsize=100 -eps=0.01

//use the vanilla method to sample RR sets
$ ./subsim -func=im -gname=facebook -seedsize=100 -eps=0.01 -vanilla=1

HIST

In highly influential scenarios, the average size of random RR sets is usually very large. The HIST algorithm is invoked to handle this problem. For example, if the edge probability in the uniform setting is large, the network is likely to be highly influential.

Example

//format the input graph in the uniform setting with a high probability
$ ./subsim -func=format -gname=facebook -pdist=uniform -pedge=0.5

//influence maximization
$ ./subsim -func=im -gname=facebook -seedsize=100 -eps=0.01 -hist=1