github.com/yusanshi/NewsRecommendation ↗
Implementations of some methods in news recommendation.
Open this visualization on its own page →
Contributors
2
Lines of Code
421
From
2020-05-19
To
2020-10-14
About yusanshi/NewsRecommendation
This repository provides implementations of several neural network models designed for news recommendation systems. The project includes six models from published papers—NRMS, NAML, LSTUR, DKN, Hi-Fi Ark, and TANR—each using different approaches like multi-head self-attention, multi-view learning, and knowledge-aware networks to predict which news articles users will engage with. Additionally, it contains an experimental model that combines techniques from NRMS with category information and ensemble methods.
The repository is designed to be practical and reproducible, offering a complete pipeline from dataset download and preprocessing through model training and evaluation. Users configure their experiments through a central config file that supports both general settings and model-specific parameters, and they can monitor training progress using TensorBoard with custom remarks for run organization. Pre-trained checkpoints are available for verification of results.
The project targets researchers and practitioners working on news recommendation systems. It uses the Microsoft News Dataset (MIND) and evaluates models using standard metrics including AUC, MRR, and normalized discounted cumulative gain at different cutoff points. The codebase is written in Python and structured to make it straightforward to compare different recommendation approaches on the same benchmark.