Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Co-NAML-LSTUR: A Combined Model with Attentive Multi-View Learning and Long- and Short-term User Representations for News Recommendation

Minh Hoang Nguyen, Thuat Thien Nguyen, Minh Nhat Ta, Tung Le, Huy Tien Nguyen

TL;DR

Co-NAML-LSTUR tackles the dual challenges of rich multi-view news representation and dual-scale user modeling under resource constraints. It combines an attentive multi-view News Encoder with a LSTUR-inspired User Encoder that fuses long-term embeddings and short-term sequences, and offers a flexible Click Predictor (dot-product or neural) trained with Noise Contrastive Estimation. The approach leverages DistilBERT-based embeddings to improve semantic encoding while maintaining efficiency, and introduces MIND-tiny as a compact training corpus to enable development on limited hardware. Empirical results on MIND-small and MIND-large show consistent improvements over strong baselines such as NRMS and NAML, with competitive performance relative to MINER and a clear emphasis on practical resource usage. The work contributes a modular, scalable framework and public code that supports efficient deployment and extension, including potential integration of external knowledge or multi-modal content in future work.

Abstract

News recommendation systems play a critical role in alleviating information overload by delivering personalized content. A key challenge lies in jointly modeling multi-view representations of news articles and capturing the dynamic, dual-scale nature of user interests-encompassing both short- and long-term preferences. Prior methods often rely on single-view features or insufficiently model user behavior across time. In this work, we introduce Co-NAML-LSTUR, a hybrid news recommendation framework that integrates NAML for attentive multi-view news encoding and LSTUR for hierarchical user modeling, designed for training on limited data resources. Our approach leverages BERT-based embeddings to enhance semantic representation. We evaluate Co-NAML-LSTUR on two widely used benchmarks, MIND-small and MIND-large. Results show that our model significantly outperforms strong baselines, achieving improvements over NRMS by 1.55% in AUC and 1.15% in MRR, and over NAML by 2.45% in AUC and 1.71% in MRR. These findings highlight the effectiveness of our efficiency-focused hybrid model, which combines multi-view news modeling with dual-scale user representations for practical, resource-limited resources rather than a claim to absolute state-of-the-art (SOTA). The implementation of our model is publicly available at https://github.com/MinhNguyenDS/Co-NAML-LSTUR

Co-NAML-LSTUR: A Combined Model with Attentive Multi-View Learning and Long- and Short-term User Representations for News Recommendation

TL;DR

Co-NAML-LSTUR tackles the dual challenges of rich multi-view news representation and dual-scale user modeling under resource constraints. It combines an attentive multi-view News Encoder with a LSTUR-inspired User Encoder that fuses long-term embeddings and short-term sequences, and offers a flexible Click Predictor (dot-product or neural) trained with Noise Contrastive Estimation. The approach leverages DistilBERT-based embeddings to improve semantic encoding while maintaining efficiency, and introduces MIND-tiny as a compact training corpus to enable development on limited hardware. Empirical results on MIND-small and MIND-large show consistent improvements over strong baselines such as NRMS and NAML, with competitive performance relative to MINER and a clear emphasis on practical resource usage. The work contributes a modular, scalable framework and public code that supports efficient deployment and extension, including potential integration of external knowledge or multi-modal content in future work.

Abstract

News recommendation systems play a critical role in alleviating information overload by delivering personalized content. A key challenge lies in jointly modeling multi-view representations of news articles and capturing the dynamic, dual-scale nature of user interests-encompassing both short- and long-term preferences. Prior methods often rely on single-view features or insufficiently model user behavior across time. In this work, we introduce Co-NAML-LSTUR, a hybrid news recommendation framework that integrates NAML for attentive multi-view news encoding and LSTUR for hierarchical user modeling, designed for training on limited data resources. Our approach leverages BERT-based embeddings to enhance semantic representation. We evaluate Co-NAML-LSTUR on two widely used benchmarks, MIND-small and MIND-large. Results show that our model significantly outperforms strong baselines, achieving improvements over NRMS by 1.55% in AUC and 1.15% in MRR, and over NAML by 2.45% in AUC and 1.71% in MRR. These findings highlight the effectiveness of our efficiency-focused hybrid model, which combines multi-view news modeling with dual-scale user representations for practical, resource-limited resources rather than a claim to absolute state-of-the-art (SOTA). The implementation of our model is publicly available at https://github.com/MinhNguyenDS/Co-NAML-LSTUR

Paper Structure

This paper contains 29 sections, 18 equations, 6 figures, 5 tables.

Table of Contents

  1. I. Introduction
  2. II. Related Work
  3. 2.1 News Recommendation
  4. 2.3 Neural News Recommendation
  5. III. Our Method
  6. 3.1 News Encoder
  7. 3.2 User Encoder
  8. 3.3 Click Predictor
  9. Dot Product-Based Predictor.
  10. Neural Network-Based Predictor.
  11. 3.4 Model Training
  12. IV. MIND-tiny
  13. 4.1 Overview of the MIND Dataset
  14. 4.2 Motivation and Construction of MIND-tiny
  15. 4.3 Dataset Statistics and Analysis
  16. ...and 14 more sections

Figures (6)

  • Figure 1: Examples of two news articles that a user has read. Image adapted from studiesbib1.
  • Figure 2: An illustration of long-term and short-term user interests in news reading. Image adapted from studiesan2019neural.
  • Figure 3: The framework of our Co-NAML-LSTUR approach for News Recommendation, which consists of a multi-view news encoder, a long- and short-term user encoder, and a click predictor.
  • Figure 4: Distributional statistics of categories, title lengths, and abstract lengths in MIND-tiny.
  • Figure 5: The news click history of a sampled user, who has various interests including entertainment, lifestyle and food.
  • ...and 1 more figures