Introduction

This repository is the implementation of KRED: Knowledge-Aware Document Representation for News Recommendations

Model description

KRED is a knowledge enhanced framework which enhance a document embedding with knowledge information for multiple news recommendation tasks. The framework mainly contains two part: representation enhancement part(left) and multi-task training part(right).

Data description:

We use MIND dataset in this repo, due to MIND dataset doesn't contain location information, so in this experiments we do not use the local news detection task.

Running the code

$ python main.py  (this will run KRED for user2item single task training set by default parameters)

we also offer an example of our extensions in this notebook: KRED_full_pipeline.ipynb

Environment

The code has been tested running under Python 3.6.10, with the following packages installed (along with their dependencies):

• numpy==1.19.1
• pytorchtools==0.0.2
• scikit-learn==0.23.2
• scipy==1.5.2
• torch==1.4.0

Extensions

In this project, we have applied 2 extensions: Ablation study and New dataset application - Amazon Books Recommendation.

Ablation study

We train and test the model by removing one of the layers and with multiple tasks: User to Item recommendation, Item to Item recommendation, and Article Category Classification.

Amazon Books Recommendation

We build a new knowledge graph with the dataset: Amazon Books, and train it on our model.