kg-summ-rec

Experiments with Knowledge Graph (KG) -based Summarization (Summ.) and Recommendation (Rec).

This is the code of the Knowledge Graph Summarization Impacts on Movie Recommendations in JIIS'21, which investigated the use of Graph Summarization (GS) as a Knowledge Graph (KG) preprocessing step of KG-based Recommender Systems (RS) and proposed KGE-K-Means Summarization, a GS method that combines KG Embedding (from Accenture/Ampligraph project) with node clustering (K-Means).

We summarize KG representing side information that enriches user-items interactions of KG-based RSs. Then, we evaluate summarized KGs in terms of reduction, RS model (from TaoMiner/joint-kg-recommender project) training efficiency and RS effectiveness (with metrics from caserec/CaseRecommender project). We adapt KG-based RS projects to evaluates effectiveness using CaseRec (see adapt folder).

Also, we provide exploratory data analisys (EDA) of original and summarized datasets using jupyter notebook.

Setup

Use the following steps in order to setup our project properly.

1. Run setup script.

$ setup.sh

Git folder should have the follow structure:

git
└─datasets
| └─ml-cao (with Cao's data)
| | └─cao-format
| | | └─ml1m
| | | | └─kg
| └─ml-sun (with Sun's data)
| | └─sun-format
└─joint-kg-recommender
└─kg-summ-rec
└─Recurrent-Knowledge-Graph-Embedding
└─results
| └─ml-cao
| └─ml-sun

4. Install cuda 7.5 from https://developer.nvidia.com/cuda-75-downloads-archive

5. Install Anaconda3, reopen terminal is required.

$ wget https://repo.anaconda.com/archive/Anaconda3-2020.02-Linux-x86_64.sh
$ bash -i Anaconda3-2020.02-Linux-x86_64.sh
$ conda update -n base -c defaults conda

6. Create python environment for each project.

$ bash -i util/create_envs.sh

Run

$ cd ~/git/kg-summ-rec
$ bash -i run.sh

OR

$ nohup bash -i run.sh </dev/null >nohup.out 2>nohup.err &
$ watch "ps -aux | grep 'python\|bash\|nohup'"
$ watch "ls -l"

Data and Results

We provide datasets and results of KGE-K-Means Summarization [1] from example JIIS-2021-revised in sacenti-jiis-2021. Note that these results were produced using JIIS2021 version of this project. To clone this specific version, please use the following command:

git clone --depth 1 --branch JIIS2021 https://github.com/juarezsacenti/kg-summ-rec.git

Reference

If you use our code, please cite our paper:

@inproceedings{sacenti2021knowledge,
  title={Knowledge Graph Summarization Impacts on Movie Recommendations},
  author={Sacenti, Juarez A. P. and Fileto, Renato and Willrich, Roberto},
  journal={J Intell Inf Syst},
  publisher={Springer},
  year={2021}
}