https://github.com/smaniu/oim
Algorithms for online influence maximization
https://github.com/smaniu/oim
algorithm graph influence-maximization research-project
Last synced: 6 months ago
JSON representation
Algorithms for online influence maximization
- Host: GitHub
- URL: https://github.com/smaniu/oim
- Owner: smaniu
- License: other
- Created: 2015-11-20T09:40:05.000Z (over 9 years ago)
- Default Branch: master
- Last Pushed: 2017-02-20T10:18:41.000Z (over 8 years ago)
- Last Synced: 2023-12-17T16:47:01.631Z (over 1 year ago)
- Topics: algorithm, graph, influence-maximization, research-project
- Language: C++
- Homepage:
- Size: 196 KB
- Stars: 19
- Watchers: 2
- Forks: 5
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
This folder contains the source code for the Online Influence Maximization with
Persistence algorithms, as described in this [paper][1] and the model-free
[extension][6]. The source code is header-only.
# Compiling
The *Makefile* is in the main folder, so simply execute *make*:
make clean; make
The output binary is *oim*.
## Dependencies
The *Makefile* requires GCC 4.9.0 (or superior) and uses C++14 features.
The code needs the Boost C++ library headers. It assumes the include files are
present in */usr/local/include*. If your Boost installation is someplace else,
you have to modify the *INCLUDE_DIRS* directive in *Makefile*. The binary
library does not need to be linked.
# Datasets
The networks used in the experimental section of [our paper][6] are available
[here](https://drive.google.com/open?id=0B6peHJJcM5GvQnZGdHU1U1owa0U). We
provide scripts to generate your own datasets in the `graph/` folder.
# Methods and usage
The program expects as input a tab delimited graph file of the following format:
node1 node2 prob
where *node1* and *node2* are the endpoints of a graph edge, and *prob* is the
influence probability.
The following methods are currently supported:
1. *exponentiated gradient*, which is run as follows:
./oim --eg [
]
2. *missing mass*, which runs as follows:
./oim --missing_mass
[ ]
3. *real graph*, which executes on the real graph:
./oim --real [
]
## Parameters
The parameters are set as follows:
* *graph* is the name of the graph file
* *alpha*, *beta* are the global prior on the edges of the graph
* *exploit*, *explore* can take any of the following values: **0** Random,
**1** Discountdegree, **2** Maxdegree, **3** [CELF][2], **4** [TIM][3],
**5** [SSA][4], **6** [PMC][5]
* *samples* is the number of spreads to estimate the expected value of chosen
seeds
* *trials* is the number of trials, *k* is the number of seeds in each trial
* *update* is **1** if the graph is updated, **0** otherwise
* *update_type* is the type of update: **0** local only, **1** least squares or
**2** maximum likelihood
* *reduction* can take the following values: **0** max cover, **1** highest
degree
* *policy* can take the following values: **0** random, **1** Good-UCB
* *model* can take the following values: **0** Linear Threshold, **1**
Independent Cascade
* *cascades* contains the path to the file containing **real** cascades (logs)
## Output
The different methods write on the standard output with the following format:
1. *exponentiated gradient*:
stage cum spread expected spread tselection tupdate
tround ttotal theta memory k model seeds
2. *missing mass*:
stage cumulative spread treduction tselection tupdate
tround ttotal memory k n_experts n_policy
n_reduction model seeds
3. *real graph*:
stage cumulative spread expected spread tround
ttotal k model seeds
# Contributors
* Paul Lagrée (Université Paris-Sud)
* Siyu Lei (University of Hong Kong) on code published in 2015
* Silviu Maniu (University of Hong-Kong, then Université Paris-Sud)
* Luyi Mo (University of Hoong-Kong) on code published in 2015
# License
The source code is provided as-is under an MIT License. If it is useful to you,
please cite the seminal [paper][1] and the model-free [extension][6].
[1]: "S. Lei, S. Maniu, L. Mo, R. Cheng, P. Senellart. Online Influence Maximization. KDD 2015"
[2]: "A. Goyal, W. Lu, L. Lakshmanan. CELF++: Optimizing the Greedy Algorithm for Influence Maximization in Social Networks. WWW 2011"
[3]: "Y. Tang, X. Xiao, and Y. Shi. Influence maximization: Near-optimal time complexity meets practical efficiency. SIGMOD 2014"
[4]: "H. T. Nguyen, M. T. Thai, and T. N. Dinh. Stop-and-Stare: Optimal Sampling Algorithms for Viral Marketing in Billion-scale Networks. SIGMOD 2016"
[5]: "N. Ohsaka, T. Akiba, Y. Yoshida and K. Kawarabayashi. Fast and Accurate Influence Maximization on Large Networks with Pruned Monte-Carlo Simulations. AAAI 2014"
[6]: "P. Lagrée, O. Cappé, B. Cautis, S. Maniu. Online Model-Free Influence Maximization with Persistence."