Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Don't Click the Bait: Title Debiasing News Recommendation via Cross-Field Contrastive Learning

Yijie Shu, Xiaokun Zhang, Youlin Wu, Bo Xu, Liang Yang, Hongfei Lin

TL;DR

The paper tackles the clickbait problem in news recommendations by showing that abstracts better reflect factual content than titles. It introduces the TDNR-C² framework, featuring a Multi-Field Knowledge Extraction (MFKE) module and a Cross-Field Contrastive Learning (C²) module, and leverages generated titles from abstracts to align title and abstract views using a $L = L_{Rec} + \lambda_{1} L_{CL}$ objective. Through experiments on the MIND-small dataset, TDNR-C² outperforms state-of-the-art baselines across $AUC$, $MRR$, and $NDCG$ metrics, with ablations confirming the value of MFKE and C² for mitigating title bias. The work demonstrates that emphasizing abstracts for debiasing can improve recommendation accuracy and reliability, with potential extensions to additional fields and interpretability improvements.

Abstract

News recommendation emerges as a primary means for users to access content of interest from the vast amount of news. The title clickbait extensively exists in news domain and increases the difficulty for news recommendation to offer satisfactory services for users. Fortunately, we find that news abstract, as a critical field of news, aligns cohesively with the news authenticity. To this end, we propose a Title Debiasing News Recommendation with Cross-field Contrastive learning (TDNR-C2) to overcome the title bias by incorporating news abstract. Specifically, a multi-field knowledge extraction module is devised to extract multi-view knowledge about news from various fields. Afterwards, we present a cross-field contrastive learning module to conduct bias removal via contrasting learned knowledge from title and abstract fileds. Experimental results on a real-world dataset demonstrate the superiority of the proposed TDNR-C2 over existing state-of-the-art methods. Further analysis also indicates the significance of news abstract for title debiasing.

Don't Click the Bait: Title Debiasing News Recommendation via Cross-Field Contrastive Learning

TL;DR

The paper tackles the clickbait problem in news recommendations by showing that abstracts better reflect factual content than titles. It introduces the TDNR-C² framework, featuring a Multi-Field Knowledge Extraction (MFKE) module and a Cross-Field Contrastive Learning (C²) module, and leverages generated titles from abstracts to align title and abstract views using a objective. Through experiments on the MIND-small dataset, TDNR-C² outperforms state-of-the-art baselines across , , and metrics, with ablations confirming the value of MFKE and C² for mitigating title bias. The work demonstrates that emphasizing abstracts for debiasing can improve recommendation accuracy and reliability, with potential extensions to additional fields and interpretability improvements.

Abstract

News recommendation emerges as a primary means for users to access content of interest from the vast amount of news. The title clickbait extensively exists in news domain and increases the difficulty for news recommendation to offer satisfactory services for users. Fortunately, we find that news abstract, as a critical field of news, aligns cohesively with the news authenticity. To this end, we propose a Title Debiasing News Recommendation with Cross-field Contrastive learning (TDNR-C2) to overcome the title bias by incorporating news abstract. Specifically, a multi-field knowledge extraction module is devised to extract multi-view knowledge about news from various fields. Afterwards, we present a cross-field contrastive learning module to conduct bias removal via contrasting learned knowledge from title and abstract fileds. Experimental results on a real-world dataset demonstrate the superiority of the proposed TDNR-C2 over existing state-of-the-art methods. Further analysis also indicates the significance of news abstract for title debiasing.
Paper Structure (25 sections, 15 equations, 4 figures, 1 table)

This paper contains 25 sections, 15 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. News Modeling in News Recommenders.
  4. Contrastive Learning for News Recommendation.
  5. Methodology
  6. Notations and Definitions
  7. Candidate News Encoder
  8. Representation Learning.
  9. Multi-Field Merging.
  10. User Encoder
  11. Multi-Field Knowledge Extraction.
  12. User Preference Extraction.
  13. Click Prediction
  14. Cross-Field Contrastive Learning
  15. Model Training
  16. ...and 10 more sections

Figures (4)

  • Figure 1: An example of news clickbait.
  • Figure 2: The proposed TDNR-C² model comprises two key components: (1) MFKE, which converts clicked news history into field-level history and captures multi-view knowledge of the clicked news history; (2) C², a cross-field contrastive learning module designed for clicked news history and candidate news to capture semantic relevance between title and abstract, alleviating the clickbait issue in news titles.
  • Figure 3: Results of ablation experiments.
  • Figure 4: Case study of the rankings of truly clicked news and clickbait news.