https://github.com/reshalfahsi/web-traffic-prediction

Web Traffic Prediction via Temporal Graph Neural Network
https://github.com/reshalfahsi/web-traffic-prediction

graph-neural-networks pytorch-geometric-temporal pytorch-lightning temporal-data temporal-graph-networks web-traffic-forecasting

Last synced: about 2 months ago
JSON representation

Web Traffic Prediction via Temporal Graph Neural Network

• Host: GitHub
• URL: https://github.com/reshalfahsi/web-traffic-prediction
• Owner: reshalfahsi
• Created: 2023-08-10T14:07:32.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2024-01-11T00:45:18.000Z (12 months ago)
• Last Synced: 2024-01-11T01:47:37.204Z (12 months ago)
• Topics: graph-neural-networks, pytorch-geometric-temporal, pytorch-lightning, temporal-data, temporal-graph-networks, web-traffic-forecasting
• Language: Jupyter Notebook
• Homepage:
• Size: 355 KB
• Stars: 1
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Web Traffic Prediction via Temporal Graph Neural Network

 


    

    


 


Temporal Graph Neural Network or Temporal GNN is one of the variants of the GNN which handles the spatio-temporal data. The term "spatio-" refers to the nature of the graph that is closely related to the spatial relationship that exists in the image data (recall that an image is basically a graph), and the term "temporal" here indicates the data may be progressively changing in a sequence of time. In this project, the [Chebysev GCNN+LSTM](https://arxiv.org/pdf/1612.07659.pdf) module and the Wiki Maths dataset are leveraged, which are provided by [PyTorch Geometric Temporal](https://github.com/benedekrozemberczki/pytorch_geometric_temporal). The complete Temporal GNN model contains the Chebysev GCNN+LSTM module, followed by a fully connected layer. Here, the model is trained to predict the daily user visits to Wikipedia's vital mathematics articles (represented by nodes/vertices). The graph's characteristic in the dataset is non-heterogenous and static. The details of the dataset can be seen [here](https://pytorch-geometric-temporal.readthedocs.io/en/latest/modules/dataset.html#torch_geometric_temporal.dataset.wikimath.WikiMathsDatasetLoader). 

## Experiment

To try out the experiment, the [Google Colab notebook](https://github.com/reshalfahsi/web-traffic-prediction/blob/master/Web_Traffic_Prediction.ipynb) is provided.

## Result

## Quantitative Result

The quantitative performance of the model is given in the table below.

Metrics | Score |

------------ | ------------- |

Loss | 0.452 |

MAPE | 4.110 |

MAE | 0.438 |

MSE | 0.519 |

## Evaluation Metric Curve

  
 Loss curve of the training and validation of the Temporal GNN model. 


  
 MAPE curve of the training and validation of the Temporal GNN model. 


  
 MAE curve of the training and validation of the Temporal GNN model. 


  
 MSE curve of the training and validation of the Temporal GNN model. 


## Qualitative Result

The figure below shows the qualitative result of the model.

  
 The visitor prediction at one of the vital mathematics articles on Wikipedia. 


## Credit

- [Structured Sequence Modeling with Graph Convolutional Recurrent Networks](https://arxiv.org/pdf/1612.07659.pdf)

- [PyTorch Geometric Temporal](https://github.com/benedekrozemberczki/pytorch_geometric_temporal)

- [PyTorch Lightning](https://lightning.ai/docs/pytorch/latest/)