Deep_Learning_Final_Project

File tree

DEEP_LEARNING_FINAL_PROJECT
├── cleaned_data/
    ├── flited_data/
    ├── graph_data/
    └── splited_data/
├── code/
├── Picture/
├── yelp_dataset/
    └── yalp_academic_dataset_xxx.json
├── yelp_photos/photo_file
    ├── photos/
    └── photos.json
├── .gitignore
├── Proposal_pre_for_team_16.pdf
└── README.md

Required library

• requirements.txt: All libraries required for this project.

Workflow and Code Details

1. filter.ipynb and train-test-split.ipynb
   • Perform data filtering and splitting.
   • Output: Training and test sets in JSON files (*.json).
2. Node_data.ipynb
   • Select all nodes' data
3. Edge_n_weight.ipynb
   • Create edge data
   • Calculate weight
4. ndata.ipynb
   • Outputs processed node features in NumPy files (*.npy).
5. graph-construction.ipynb
   • Takes the two TXT files, node features, and the predicted tip ratings.
   • Outputs the binary file of the graph (*.bin).
6. a-gnn.ipynb
   • Takes the graph (*.bin) as input and outputs the predictions as *.npy.
7. fig3b.ipynb
   • Analyzes the output and plots Figure 3b.
8. all_models.ipynb
   • Takes as input:
     • Predictions of the GNN (*.npy).
     • node_type_ID.txt
     • The split *.json (training and test sets).
   • Prints evaluation metrics of the models.
   • Outputs predictions as *.npy.
9. Geographical Recommend.py
   • Sorts the nearest stores around a given location and recommends the stores based on a pre-ranked matrix for each user.
   • Example
     • one user has a row of 1,434 elements where each element represents the weight of how likely the user will visit that store.
     • Stores are selected based on the dist function.
     • Filtered stores are reranked based on weights using the function recall at 20.