This is a code repogitory of PGT which is for a research paper of news recommender system. This project contains a preprocess code for the dataset (Adressa, Globo), and an implementation of competitor methods. This document includes a brief description of code, an environment setting, and an execution procedure. Please refer the 'code description document' to find the detailed information.
This project should be executed by python 3.6. You need to install the related python packages using the following command.
- pip install -r requirements.txt
You can find the sub-directory structure of the project in this section. The raw dataset files are not included in this repogitory because of the license issue.
Makefile: All procedures of project are executed by the dependency tree in the Makefile.src/: the path of directory for the source codes.data/: the path of raw dataset files (download links are listed below).cache/: the path of directory storing intermediate output files of the data preprocessing (this will be generated automatically during execution).
Please download the following datasets and put them to data/[dataset_name] folder of the project. The paths of the datasets should be:
Adressa (http://reclab.idi.ntnu.no/dataset)
- data/adressa/one_week/20170101 ... 20170107
- data/adressa/three_month/20170101 ... 20170331
- data/globo/clicks
- data/globo/articles_metadata.csv
All codes should be executed by the Makefile For example, you need to input the following command in the prompt to execute the pgt.
- make comp_pgt
This software may be used only for research evaluation purposes. For other purposes (e.g., commercial), please contact the authors. The authors are as follows:
- Bonhun Koo (darkgs@snu.ac.kr)
- Hyunsik Jeon (jeon185@snu.ac.kr)
- U Kang (ukang@snu.ac.kr) - corresponding author