LogiRec++

This repository contains a implementation of our "Logical Relation Modeling and Mining in Hyperbolic Space for Recommendation".

Environment Setup

1. torch 1.13.0
2. Python 3.8.13
3. geoopt ($ pip install git+https://github.com/geoopt/geoopt.git)
4. numpy
5. scipy
6. pandas
7. tqdm

Guideline

data

We provide one dataset, ciao.

adj_csr.npz adj matrix built for training gcn.

item_tag_matrix.npz items attributes matrix.

train.pkl train set.

test.pkl test set.

user_item_list.pkl user-item dict for the complete dataset.

implication.pt the hierarchies between tags.

exclusion.pt the exclusions between tags.

models

The implementation of model(model.py). code to implement Hyperbolic gcn (encoders.py, hyp_layers.py).

utils

data_generator.py read and organize data.

helper.py some method for helping preprocess data or set seeds and devices.

sampler.py a parallel sampler to sample batches for training.

train_utils.py read and parse the config arguments.

Example to run the codes

python run.py
python run.py > logs/ciao.log