Python实现基于图神经网络的异构图表示学习和推荐算法研究

资源下载地址：https://download.csdn.net/download/sheziqiong/85978304
资源下载地址：https://download.csdn.net/download/sheziqiong/85978304

目录结构

GNN-Recommendation/gnnrec/ 算法模块顶级包hge/异构图表示学习模块kgrec/基于图神经网络的推荐算法模块data/ 数据集目录（已添加.gitignore）model/模型保存目录（已添加.gitignore）img/图片目录academic_graph/ Django项目模块rank/ Django应用manage.py Django管理脚本

安装依赖

Python 3.7

CUDA 11.0

pip install -r requirements_cuda.txt

CPU

pip install -r requirements.txt

异构图表示学习（附录）

基于对比学习的关系感知异构图神经网络(Relation-aware Heterogeneous Graph Neural Network with Contrastive Learning, RHCO)

实验

见 readme

基于图神经网络的推荐算法(附录)

基于图神经网络的学术推荐算法(Graph Neural Network based Academic Recommendation Algorithm, GARec)

实验

见 readme

Django 配置

MySQL 数据库配置

创建数据库及用户

CREATE DATABASE academic_graph CHARACTER SET utf8mb4;CREATE USER 'academic_graph'@'%' IDENTIFIED BY 'password';GRANT ALL ON academic_graph.* TO 'academic_graph'@'%';

在根目录下创建文件.mylogin.cnf

[client]host = x.x.x.xport = 3306user = usernamepassword = passworddatabase = databasedefault-character-set = utf8mb4

创建数据库表

python manage.py makemigrations --settings=academic_graph.settings.prod rankpython manage.py migrate --settings=academic_graph.settings.prod

导入 oag-cs 数据集

python manage.py loadoagcs --settings=academic_graph.settings.prod

注：由于导入一次时间很长（约 9 小时），为了避免中途发生错误，可以先用 data/oag/test 中的测试数据调试一下

拷贝静态文件

python manage.py collectstatic --settings=academic_graph.settings.prod

启动 Web 服务器

export SECRET_KEY=xxxpython manage.py runserver --settings=academic_graph.settings.prod 0.0.0.0:8000

系统截图

搜索论文

论文详情

搜索学者

学者详情

附录

基于图神经网络的推荐算法

数据集

oag-cs – 使用 OAG 微软学术数据构造的计算机领域的学术网络（见 readme）

预训练顶点嵌入

使用 metapath2vec（随机游走 +word2vec）预训练顶点嵌入，作为 GNN 模型的顶点输入特征

随机游走

python -m gnnrec.kgrec.random_walk model/word2vec/oag_cs_corpus.txt

训练词向量

python -m gnnrec.hge.metapath2vec.train_word2vec --size=128 --workers=8 model/word2vec/oag_cs_corpus.txt model/word2vec/oag_cs.model

召回

使用微调后的 SciBERT 模型（见 readme 第 2 步）将查询词编码为向量，与预先计算好的论文标题向量计算余弦相似度，取 top k

python -m gnnrec.kgrec.recall

召回结果示例：

graph neural network

0.9629Aggregation Graph Neural Networks0.9579Neural Graph Learning: Training Neural Networks Using Graphs0.9556Heterogeneous Graph Neural Network0.9552Neural Graph Machines: Learning Neural Networks Using Graphs0.9490On the choice of graph neural network architectures0.9474Measuring and Improving the Use of Graph Information in Graph Neural Networks0.9362Challenging the generalization capabilities of Graph Neural Networks for network modeling0.9295Strategies for Pre-training Graph Neural Networks0.9142Supervised Neural Network Models for Processing Graphs0.9112Geometrically Principled Connections in Graph Neural Networks

recommendation algorithm based on knowledge graph

0.9172Research on Video Recommendation Algorithm Based on Knowledge Reasoning of Knowledge Graph0.8972An Improved Recommendation Algorithm in Knowledge Network0.8558A personalized recommendation algorithm based on interest graph0.8431An Improved Recommendation Algorithm Based on Graph Model0.8334The Research of Recommendation Algorithm based on Complete Tripartite Graph Model0.8220Recommendation Algorithm based on Link Prediction and Domain Knowledge in Retail Transactions0.8167Recommendation Algorithm Based on Graph-Model Considering User Background Information0.8034A Tripartite Graph Recommendation Algorithm Based on Item Information and User Preference0.7774Improvement of TF-IDF Algorithm Based on Knowledge Graph0.7770Graph Searching Algorithms for Semantic-Social Recommendation

scholar disambiguation

0.9690Scholar search-oriented author disambiguation0.9040Author name disambiguation in scientific collaboration and mobility cases0.8901Exploring author name disambiguation on PubMed-scale0.8852Author Name Disambiguation in Heterogeneous Academic Networks0.8797KDD Cup 2013: author disambiguation0.8796A survey of author name disambiguation techniques: 2010–20160.8721Who is Who: Name Disambiguation in Large-Scale Scientific Literature0.8660Use of ResearchGate and Google CSE for author name disambiguation0.8643Automatic Methods for Disambiguating Author Names in Bibliographic Data Repositories0.8641A brief survey of automatic methods for author name disambiguation

精排

构造 ground truth

（1）验证集

从 AMiner 发布的 AI 2000 人工智能全球最具影响力学者榜单抓取人工智能 20 个子领域的 top 100 学者

pip install scrapy>=2.3.0cd gnnrec/kgrec/data/preprocessscrapy runspider ai2000_crawler.py -a save_path=/home/zzy/GNN-Recommendation/data/rank/ai2000.json

与 oag-cs 数据集的学者匹配，并人工确认一些排名较高但未匹配上的学者，作为学者排名 ground truth 验证集

export DJANGO_SETTINGS_MODULE=academic_graph.settings.commonexport SECRET_KEY=xxxpython -m gnnrec.kgrec.data.preprocess.build_author_rank build-val

（2）训练集

参考 AI 2000 的计算公式，根据某个领域的论文引用数加权求和构造学者排名，作为 ground truth 训练集

计算公式：

即：假设一篇论文有 n 个作者，第 k 作者的权重为 1/k，最后一个视为通讯作者，权重为 1/2，归一化之后计算论文引用数的加权求和

python -m gnnrec.kgrec.data.preprocess.build_author_rank build-train

（3）评估 ground truth 训练集的质量

python -m gnnrec.kgrec.data.preprocess.build_author_rank eval

nDGC@100=0.2420 Precision@100=0.1859Recall@100=0.2016nDGC@50=0.2308Precision@50=0.2494 Recall@50=0.1351nDGC@20=0.2492Precision@20=0.3118 Recall@20=0.0678nDGC@10=0.2743Precision@10=0.3471 Recall@10=0.0376nDGC@5=0.3165 Precision@5=0.3765Recall@5=0.0203

（4）采样三元组

从学者排名训练集中采样三元组(t, ap, an)，表示对于领域 t，学者 ap 的排名在 an 之前

python -m gnnrec.kgrec.data.preprocess.build_author_rank sample

训练 GNN 模型

python -m gnnrec.kgrec.train model/word2vec/oag-cs.model model/garec_gnn.pt data/rank/author_embed.pt

异构图表示学习

数据集

ACM – ACM 学术网络数据集
DBLP – DBLP 学术网络数据集
ogbn-mag – OGB 提供的微软学术数据集
oag-venue – oag-cs 期刊分类数据集

数据集	顶点数	边数	目标顶点	类别数
ACM	11246	34852	paper	3
DBLP	26128	239566	author	4
ogbn-mag	1939743	21111007	paper	349
oag-venue	4235169	34520417	paper	360

Baselines

R-GCN
HGT
HGConv
R-HGNN
C&S
HeCo

R-GCN (full batch)

python -m gnnrec.hge.rgcn.train --dataset=acm --epochs=10python -m gnnrec.hge.rgcn.train --dataset=dblp --epochs=10python -m gnnrec.hge.rgcn.train --dataset=ogbn-mag --num-hidden=48python -m gnnrec.hge.rgcn.train --dataset=oag-venue --num-hidden=48 --epochs=30

（使用 minibatch 训练准确率就是只有 20% 多，不知道为什么）

预训练顶点嵌入

使用 metapath2vec（随机游走 +word2vec）预训练顶点嵌入，作为 GNN 模型的顶点输入特征

python -m gnnrec.hge.metapath2vec.random_walk model/word2vec/ogbn-mag_corpus.txtpython -m gnnrec.hge.metapath2vec.train_word2vec --size=128 --workers=8 model/word2vec/ogbn-mag_corpus.txt model/word2vec/ogbn-mag.model

HGT

python -m gnnrec.hge.hgt.train_full --dataset=acmpython -m gnnrec.hge.hgt.train_full --dataset=dblppython -m gnnrec.hge.hgt.train --dataset=ogbn-mag --node-embed-path=model/word2vec/ogbn-mag.model --epochs=40python -m gnnrec.hge.hgt.train --dataset=oag-venue --node-embed-path=model/word2vec/oag-cs.model --epochs=40

HGConv

python -m gnnrec.hge.hgconv.train_full --dataset=acm --epochs=5python -m gnnrec.hge.hgconv.train_full --dataset=dblp --epochs=20python -m gnnrec.hge.hgconv.train --dataset=ogbn-mag --node-embed-path=model/word2vec/ogbn-mag.modelpython -m gnnrec.hge.hgconv.train --dataset=oag-venue --node-embed-path=model/word2vec/oag-cs.model

R-HGNN

python -m gnnrec.hge.rhgnn.train_full --dataset=acm --num-layers=1 --epochs=15python -m gnnrec.hge.rhgnn.train_full --dataset=dblp --epochs=20python -m gnnrec.hge.rhgnn.train --dataset=ogbn-mag model/word2vec/ogbn-mag.modelpython -m gnnrec.hge.rhgnn.train --dataset=oag-venue --epochs=50 model/word2vec/oag-cs.model

C&S

python -m gnnrec.hge.cs.train --dataset=acm --epochs=5python -m gnnrec.hge.cs.train --dataset=dblp --epochs=5python -m gnnrec.hge.cs.train --dataset=ogbn-mag --prop-graph=data/graph/pos_graph_ogbn-mag_t5.binpython -m gnnrec.hge.cs.train --dataset=oag-venue --prop-graph=data/graph/pos_graph_oag-venue_t5.bin

HeCo

python -m gnnrec.hge.heco.train --dataset=ogbn-mag model/word2vec/ogbn-mag.model data/graph/pos_graph_ogbn-mag_t5.binpython -m gnnrec.hge.heco.train --dataset=oag-venue model/word2vec/oag-cs.model data/graph/pos_graph_oag-venue_t5.bin

（ACM 和 DBLP 的数据来自 https://github.com/ZZy979/pytorch-tutorial/tree/master/gnn/heco ，准确率和 Micro-F1 相等）

RHCO

基于对比学习的关系感知异构图神经网络(Relation-aware Heterogeneous Graph Neural Network with Contrastive Learning, RHCO)

在 HeCo 的基础上改进：

网络结构编码器中的注意力向量改为关系的表示（类似于 R-HGNN）
正样本选择方式由元路径条数改为预训练的 HGT 计算的注意力权重、训练集使用真实标签
元路径视图编码器改为正样本图编码器，适配 mini-batch 训练
Loss 增加分类损失，训练方式由无监督改为半监督
在最后增加 C&S 后处理步骤

ACM

python -m gnnrec.hge.hgt.train_full --dataset=acm --save-path=model/hgt/hgt_acm.ptpython -m gnnrec.hge.rhco.build_pos_graph_full --dataset=acm --num-samples=5 --use-label model/hgt/hgt_acm.pt data/graph/pos_graph_acm_t5l.binpython -m gnnrec.hge.rhco.train_full --dataset=acm data/graph/pos_graph_acm_t5l.bin

DBLP

python -m gnnrec.hge.hgt.train_full --dataset=dblp --save-path=model/hgt/hgt_dblp.ptpython -m gnnrec.hge.rhco.build_pos_graph_full --dataset=dblp --num-samples=5 --use-label model/hgt/hgt_dblp.pt data/graph/pos_graph_dblp_t5l.binpython -m gnnrec.hge.rhco.train_full --dataset=dblp --use-data-pos data/graph/pos_graph_dblp_t5l.bin

ogbn-mag（第 3 步如果中断可使用–load-path 参数继续训练）

python -m gnnrec.hge.hgt.train --dataset=ogbn-mag --node-embed-path=model/word2vec/ogbn-mag.model --epochs=40 --save-path=model/hgt/hgt_ogbn-mag.ptpython -m gnnrec.hge.rhco.build_pos_graph --dataset=ogbn-mag --num-samples=5 --use-label model/word2vec/ogbn-mag.model model/hgt/hgt_ogbn-mag.pt data/graph/pos_graph_ogbn-mag_t5l.binpython -m gnnrec.hge.rhco.train --dataset=ogbn-mag --num-hidden=64 --contrast-weight=0.9 model/word2vec/ogbn-mag.model data/graph/pos_graph_ogbn-mag_t5l.bin model/rhco_ogbn-mag_d64_a0.9_t5l.ptpython -m gnnrec.hge.rhco.smooth --dataset=ogbn-mag model/word2vec/ogbn-mag.model data/graph/pos_graph_ogbn-mag_t5l.bin model/rhco_ogbn-mag_d64_a0.9_t5l.pt

oag-venue

python -m gnnrec.hge.hgt.train --dataset=oag-venue --node-embed-path=model/word2vec/oag-cs.model --epochs=40 --save-path=model/hgt/hgt_oag-venue.ptpython -m gnnrec.hge.rhco.build_pos_graph --dataset=oag-venue --num-samples=5 --use-label model/word2vec/oag-cs.model model/hgt/hgt_oag-venue.pt data/graph/pos_graph_oag-venue_t5l.binpython -m gnnrec.hge.rhco.train --dataset=oag-venue --num-hidden=64 --contrast-weight=0.9 model/word2vec/oag-cs.model data/graph/pos_graph_oag-venue_t5l.bin model/rhco_oag-venue.ptpython -m gnnrec.hge.rhco.smooth --dataset=oag-venue model/word2vec/oag-cs.model data/graph/pos_graph_oag-venue_t5l.bin model/rhco_oag-venue.pt

消融实验

python -m gnnrec.hge.rhco.train --dataset=ogbn-mag --model=RHCO_sc model/word2vec/ogbn-mag.model data/graph/pos_graph_ogbn-mag_t5l.bin model/rhco_sc_ogbn-mag.ptpython -m gnnrec.hge.rhco.train --dataset=ogbn-mag --model=RHCO_pg model/word2vec/ogbn-mag.model data/graph/pos_graph_ogbn-mag_t5l.bin model/rhco_pg_ogbn-mag.pt

实验结果

顶点分类

参数敏感性分析

消融实验

资源下载地址：https://download.csdn.net/download/sheziqiong/85978304
资源下载地址：https://download.csdn.net/download/sheziqiong/85978304

Python实现基于图神经网络的异构图表示学习和推荐算法研究

目录结构

安装依赖

CUDA 11.0

CPU

异构图表示学习（附录）

实验

基于图神经网络的推荐算法(附录)

实验

Django 配置

MySQL 数据库配置

拷贝静态文件

启动 Web 服务器

系统截图

附录

基于图神经网络的推荐算法

数据集

预训练顶点嵌入

召回

精排

构造 ground truth

训练 GNN 模型

异构图表示学习

数据集

Baselines

R-GCN (full batch)

预训练顶点嵌入

HGT

HGConv

R-HGNN

C&S

HeCo

RHCO

实验结果

相关文章

最新关注

热文推荐