# deepcommunitydetection

**Repository Path**: dyjch/deepcommunitydetection

## Basic Information

- **Project Name**: deepcommunitydetection
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-11-16
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README


PyDCD: A Deep Learning-Based Community Detection Software in Python for Large-scale Networks
=========================================================

DCD (Deep learning-based Community Detection) is designed to apply state-of-the-art deep learning technologies to identify communities for large-scale networks. Compared with existing community detection methods, DCD offers a unified solution for many variations of community detection problems.  

![DCD logo](pics/frameworks.png)

DCD provides implementation of 4 community detection algorithms, 1 evaluation, and two types of networked data:


| Function      | Description       | Input | Output |
|------------|-------------------------------|-----------|---------|
| K-Means     | Baseline (1) | -Network node file <br/> -Network edge file <br/>-Performance evaluation flag <br/> -K | <node id, community id> |
| MM      | Baseline (2) | -Network node file <br/> -Network edge file | <node id, community id> |
| DCD     | DCD | -Network node file <br/> -Network edge file <br/>-Performance evaluation flag <br/> -Node attribute flag<br/> -K | <node id, community id> |
| Random network Generation | Generate random network datasets | -Network size <br/> -Community size <br/> -Probability of edges within communities <br/> -Probability of edges between communities <br/> -Directed network flag | <node id, community id> <br/> Network node file <br/> Network edge file |
| Load Dataset | Load Facebook, citation or user-provided datasets | Dataset name| Facebook dataset <br/>[Citation] dataset |

[Citation]: https://snap.stanford.edu/data/cit-HepTh.html

<!--- Performance
------------

Performance comparison on four random networks. Note: numbers in parentheses are running time (seconds).

| Network size    | Community size  | K-Means | Modularity <br/> Maximization| DCD   |
|-----------------|-----------------|---------|------------------------|---------------|
| 100       | 10  | 0.561<br/>(0.07) | 0.922<br/>(0.01) |0.826<br/>(0.01)|
| 1,000     | 100 | 0.699<br/>(1.04) | 0.807<br/>(1.11) |0.935<br/>(0.11)|
| 10,000    | 100 | 0.726<br/>(199.90) | 0.633<br/>(338.82) |0.845<br/>(62.30)|
| 20,000    | 100 | 0.709<br/>(807.56) | 0.702<br/>(1666.59) |0.814<br/>(444.12)|


Performance comparison on two real-world networks. Note: numbers in parentheses are running time (seconds).

| Network   | Community size  | K-Means | Modularity <br/> Maximization| DCD  | DCD+|
|-----------------|-----------------|---------|------------------------|------|-----|
|       | 50  | 0.451<br/>(82.72) |  /   |0.503<br/>(38.46)| 0.532<br/>(39.01) |
|       | 100 | 0.427<br/>(103.91) |   /  |0.519<br/>(37.86)| 0.520<br/>(38.95) |
| Facebook <br/> weighted and undirected <br/>network with node attributes | 150 | 0.406<br/>(118.58) |   /  |0.532<br/>(37.87)| 0.525<br/>(38.92) |
|       | 200 | 0.383<br/>(144.77) |  /   |0.521<br/>(37.87)| 0.530<br/>(39.50)|
|       | 33(mm)| 0.464<br/>(75.47) |  0.516<br/>(64.70) |0.521<br/>(38.05)| 0.538<br/>(39.20) |
|       |       |   |   |   |   |
|       | 100 | 0.438<br/>(446.27) |  /   |0.897<br/>(216.36)||
|       | 200 | 0.447<br/>(596.94) |   /  |0.916<br/>(216.37)||
| Facebook <br/> weighted and undirected <br/>network with node attributes | 500 | 0.561<br/>(1096.14) |   /  |0.927<br/>(216.59)| No node attributes |
|       | 1,000 | 0.611<br/>(1843.03) |  /   |0.940<br/>(217.32)| |
|       | 2,078 (mm)  | 0.660<br/>(3219.14) |  0.790<br/>(715.26) |0.939<br/>(217.56)| |
-->

Requirements
------------

Generally, the library is compatible with Python 3.6/3.7.<br/>
NetworkX >= 2.3


Installation
------------

### From PIP ###

```bash
pip3 install pydcd
```

Quick Start
-----------

Here is a quick-start example.

```bash
Python 3.7.3 (default, January 01 2020, 09:00:00) 
[Clang 4.0.1 (tags/RELEASE_401/final)] :: Anaconda, Inc. on darwin
Type "help", "copyright", "credits" or "license" for more information.

>>> from pydcd import DCD, KM, MM
>>> kmeans_detector = KM(10)
>>> kmeans_detector.km_detect_community('fb_nodes.txt','fb_edges.txt','N') # N means no evaluation

>>> mm_detector = MM()
>>> mm_detector.mm_detect_community('fb_nodes.txt','fb_edges.txt','Y') # Y means showing evaluation

>>> dcd_detector = DCD() # using default setting for initialization, or
>>> dcd_detector = DCD(128,64,128,50) # set the neurons for three hidden layers and the output dimension
>>> dcd_detector.dcd_detect_community('fb_nodes_withattributes.txt','fb_edges.txt','Y','N') # Y means nodes having attributes
>>> dcd_detector.dcd_detect_community('fb_nodes_noattributes.txt','fb_edges.txt','N','N') # The first N means nodes no attributes

>>> rn = RandNet() # to generate random networks
>>> rn.generate_random_networks(1000,100,0.2,0.05) # undirected network with 1000 nodes and 100 communities
>>> rn.generate_random_networks(1000,100,0.2,0.05,directed=True) # directed network with 1000 nodes and 100 communities

```

Input Examples
------------------

node file without attributes:

```bash
node_id_1
node_id_2
node_id_3
...
node_id_n
```

node file with attributes:

```bash
node_id_1 <tab> value_for_attribute_1 value_for_attribute_2 ... value_for_attribute_m
node_id_2 <tab> value_for_attribute_1 value_for_attribute_2 ... value_for_attribute_m
node_id_3 <tab> value_for_attribute_1 value_for_attribute_2 ... value_for_attribute_m
...
node_id_n <tab> value_for_attribute_1 value_for_attribute_2 ... value_for_attribute_m
```

edge file:

```bash
node_id_1 node_id_2
...
node_id_i node_id_j
...
node_id_m node_id_k
```

Development Team
----------------

PyDCD is developed by Prof. [Kunpeng Zhang], Prof. [Shaokun Fan], and Prof. [Bruce Golden].

[Kunpeng Zhang]: http://www.terpconnect.umd.edu/~kpzhang/
[Shaokun Fan]: https://business.oregonstate.edu/users/shaokun-fan
[Bruce Golden]: http://scholar.rhsmith.umd.edu/bgolden/home

Citation
--------

If you find this useful for your research or development, please cite our work.