This repository contains the code and a dataset example for a news recommendation system leveraging both DKN (Deep Knowledge-Aware Network) and NPA (Neural News Recommendation with Personalized Attention) models. The project aims to enhance user engagement by delivering personalized news recommendations.
-
Install Dependencies: Install the required packages using the command:
pip install recommenders
-
Download and Prepare Data: The MIND dataset is utilized for training and evaluation. Use the provided scripts to download and preprocess the data.
-
Model Implementation:
- DKN: Integrates knowledge graph embeddings with a knowledge-aware CNN to enrich news representations.
- NPA: Utilizes CNNs and personalized attention mechanisms to transform news titles and user interaction data into informative embeddings.
-
Future Improvements:
- Two-Tower Model: Implementing a Two-Tower model to process user and item data simultaneously for better embedding and recommendation precision.
-
Evaluation Metrics: The system's performance is evaluated using AUC, MRR, and nDCG metrics.