Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CROWN: A Novel Approach to Comprehending Users' Preferences for Accurate Personalized News Recommendation

Yunyong Ko, Seongeun Ryu, Sang-Wook Kim

TL;DR

CROWN tackles personalized news recommendation by jointly modeling manifold article intents, post-read preferences, and cold-start users. It introduces a category-guided intent disentangled news encoder, a consistency-based aggregation, and a GNN-enhanced hybrid user encoder, with an auxiliary category-prediction task to improve intent disentanglement. Extensive experiments on MIND-small and Adressa show consistent, substantial gains over 12 baselines across AUC, MRR, and nDCG, and ablations confirm the effectiveness of each component. The framework advances accurate personalization under realistic constraints and provides reproducible code for the community.

Abstract

Personalized news recommendation aims to assist users in finding news articles that align with their interests, which plays a pivotal role in mitigating users' information overload problem. Although many recent works have been studied for better personalized news recommendation, the following challenges should be explored more: (C1) Comprehending manifold intents coupled within a news article, (C2) Differentiating varying post-read preferences of news articles, and (C3) Addressing the cold-start user problem. To tackle the aforementioned challenges together, in this paper, we propose a novel personalized news recommendation framework (CROWN) that employs (1) category-guided intent disentanglement for (C1), (2) consistency-based news representation for (C2), and (3) GNN-enhanced hybrid user representation for (C3). Furthermore, we incorporate a category prediction into the training process of CROWN as an auxiliary task, which provides supplementary supervisory signals to enhance intent disentanglement. Extensive experiments on two real-world datasets reveal that (1) CROWN provides consistent performance improvements over ten state-of-the-art news recommendation methods and (2) the proposed strategies significantly improve the accuracy of CROWN.

CROWN: A Novel Approach to Comprehending Users' Preferences for Accurate Personalized News Recommendation

TL;DR

CROWN tackles personalized news recommendation by jointly modeling manifold article intents, post-read preferences, and cold-start users. It introduces a category-guided intent disentangled news encoder, a consistency-based aggregation, and a GNN-enhanced hybrid user encoder, with an auxiliary category-prediction task to improve intent disentanglement. Extensive experiments on MIND-small and Adressa show consistent, substantial gains over 12 baselines across AUC, MRR, and nDCG, and ablations confirm the effectiveness of each component. The framework advances accurate personalization under realistic constraints and provides reproducible code for the community.

Abstract

Personalized news recommendation aims to assist users in finding news articles that align with their interests, which plays a pivotal role in mitigating users' information overload problem. Although many recent works have been studied for better personalized news recommendation, the following challenges should be explored more: (C1) Comprehending manifold intents coupled within a news article, (C2) Differentiating varying post-read preferences of news articles, and (C3) Addressing the cold-start user problem. To tackle the aforementioned challenges together, in this paper, we propose a novel personalized news recommendation framework (CROWN) that employs (1) category-guided intent disentanglement for (C1), (2) consistency-based news representation for (C2), and (3) GNN-enhanced hybrid user representation for (C3). Furthermore, we incorporate a category prediction into the training process of CROWN as an auxiliary task, which provides supplementary supervisory signals to enhance intent disentanglement. Extensive experiments on two real-world datasets reveal that (1) CROWN provides consistent performance improvements over ten state-of-the-art news recommendation methods and (2) the proposed strategies significantly improve the accuracy of CROWN.
Paper Structure (21 sections, 3 equations, 7 figures, 9 tables)

This paper contains 21 sections, 3 equations, 7 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Definition
  4. CROWN: Proposed Framework
  5. Module 1: News Encoder
  6. Module 2: User Encoder
  7. Modules 3-4: Click and Category Predictors
  8. Experimental Validation
  9. Experimental Setup
  10. EQ1. Model Accuracy
  11. EQ2. News Representation
  12. EQ3. User Representation
  13. EQ4. Auxiliary Task
  14. EQ5. Cold Start User Problem
  15. Conclusion
  16. ...and 6 more sections

Figures (7)

  • Figure 1: Challenges of personalized news recommendation: (C1) news articles are usually created with various intents, (C2) users have varying post-read preferences to news articles, and (C3) some users have only a few clicked news.
  • Figure 2: (Left) Different intent distributions of news articles according to their categories in MIND wu2020mind and (Right) Distribution of Pearson correlation coefficients between the title-content consistency and users' content reading time in Adressa gulla2017adressa.
  • Figure 3: Overview of CROWN: two encoding modules (Modules 1-2) and two prediction modules (Modules 3-4).
  • Figure 4: The impact of the auxiliary task on the accuracy according to the control weight $\beta$. The auxiliary task with $0.3\leq\beta\leq0.5$ is beneficial to improving the accuracy.
  • Figure 5: Visualization of the effect of the auxiliary task of CROWN. News embeddings (a) without and (b) with incorporating the auxiliary task in CROWN.
  • ...and 2 more figures