Skip to content


Repository files navigation


This repository contains code that is associated with our paper accepted for publication at the IEEE/WIC/ACM International Joint Conference On Web Intelligence And Intelligent Agent Technology (WI-IAT '20). The paper is titled HyCoNN: Hybrid Cooperative Neural Networks for Personalized News Discussion Recommendation. It is based on a Master's thesis by Victor Künstler, co-supervised by Julian Risch and Ralf Krestel. A video presentation is available on YouTube.


If you use our work, please cite our upcoming paper HyCoNN: Hybrid Cooperative Neural Networks for Personalized News Discussion Recommendation accepted for publication at WI-IAT'20 as follows:

title = {HyCoNN: Hybrid Cooperative Neural Networks for Personalized News Discussion Recommendation},
author = {Risch, Julian and K{\"u}nstler, Victor and Krestel, Ralf},
booktitle = {Proceedings of the International Joint Conferences on Web Intelligence and Intelligent Agent Technologies (WI-IAT)},
pages = {41--48},
year = {2020},
publisher = {IEEE Computer Society},
doi = {10.1109/WIIAT50758.2020.00011},
url = {},

There are this, a requirements.txt and seven subdirectories:

  • baselines_evaluation contains code for all experiments, including the rank fusion ensemble
  • category_crawler is a small scraper to retrieve keywords (categories) of news articles
  • comment_explorer is a comment exploration tool only used for debugging purposes and not described in the paper
  • data_exploration contains code for exploratory data analysis, including association rule mining
  • models contains the implementation of the neural network models and the community graph
  • preprocessing contains the preprocessing methods, such as the selection of negative samples or the tf-idf model
  • torchtrainer-master is used for callbacks during the training process: torchtrainer


We provide the comment ids of the training, validation and test datasets. There are positive and negative samples describing the state of discussions (at a particular point in time) where a particular user did or did not comment.

Example row from one of the csv files: author_id,article_id,max_timestamp,comment_ids 44041,72032,2017-06-12 19:08:29+00:00,"[100252639, 100239810, 100224652, 100180729]"

author_id identifies the user, article_id identifies the news article discussion, max_timestamp is the point in time described by the row, comment_ids is the list of comment ids posted in the discussion until the point in time given by max_timestamp.

The csv file contains either only positive or only negative samples. The example is from the file with negative samples. Therefore, the row describes a situation where the user with id author_id did not comment on the discussion on article article_id with the comments comment_ids. The discussion where the user did comment is in the file with the positive samples.

For easier processing, the files are split into partitions, e.g., partition-0_val.csv, partition-2_val.csv, partition-3_val.csv, ...

The zipped files can be downloaded here (1.5GB):


No description, website, or topics provided.






No releases published


No packages published