# Link Prediction using GCN on pytorch
## Project explanation
This project is to predict whether patent's cpc nodes are linked or not. To accomplish this project, general GCN model from Kipf are used on pytorch. The patents are crawled in the Mobile Payment Industry.
## Framework
1) Search 'Mobile Payment' in google patent advanced search and get patent numbers.
2) Crawl all patents by using the patent numbers from 1).
3) Create adjacency matrix and feature matrix
4) Remove links and split data into train set and validation set.
5) Get new node features from GCN layers.
6) Calculate similarities of node pairs.
7) Minimize loss with the labels and update weights.
## Usage
1) ```python crawling.py```
2) ```python removelinks.py```
3) ```python features.py```
4) ```python train.py```
* Best epochs was 44~46
## Reference
- kenyonke/LinkPredictionGCN
- tkipf/pygcn
```
@article{kipf2016semi,
title={Semi-Supervised Classification with Graph Convolutional Networks},
author={Kipf, Thomas N and Welling, Max},
journal={arXiv preprint arXiv:1609.02907},
year={2016}
}
```
易三叨
- 粉丝: 47
- 资源: 4609
最新资源
- 共创在线考试系统(JSP+SERVLET)130223.rar
- 俄罗斯方块(java+applet)130228.rar
- 会员管理系统(struts+hibernate+spring)130226.rar
- 会员管理系统(struts+hibernate+spring).rar
- 基于J2EE在分布式环境下的底层结构(外文翻译+文献综述).rar
- 基于JSP的畅想空间电子商务系统.rar
- 基于JSP的网上购物系统的设计与实现(源代码+论文).rar
- 基于JSP的房产中介系统的设计与实现(源代码+论文).rar
- 基于JSP电子商务系统?(论文).rar
- 基于jsp的新闻发布系统(论文).rar
- 基于matlab的数字图像分割技术研究及实现(论文).rar
- 基于jsp网上书店(源代码+论文).rar
- 计算器(java+applet)130228.rar
- 家庭理财系统(java+applet)130227.rar
- 教材订购系统(jsp+servlet+mysql).rar
- 教材订购系统(jsp+servlet+mysql)130226.rar
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈