Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

News Recommendation with Category Description by a Large Language Model

Yuki Yada, Hayato Yamana

TL;DR

The paper tackles information overload in online news by enriching content features with automatically generated category descriptions produced by a large language model. It integrates these LLM-generated descriptions with existing PLM-based news encoders, using a concatenation strategy with a SEP token to form enhanced representations. Experiments on the MIND dataset demonstrate consistent AUC gains of up to about 5.8% over strong baselines, underscoring the value of category-level semantic enrichment for recommendations. The approach shows promise for improving content understanding in news systems, though it notes limitations where generated descriptions may occasionally misalign with category scope and calls for future improvements in description quality. Code for reproduction is provided, enabling broader adoption and evaluation.

Abstract

Personalized news recommendations are essential for online news platforms to assist users in discovering news articles that match their interests from a vast amount of online content. Appropriately encoded content features, such as text, categories, and images, are essential for recommendations. Among these features, news categories, such as tv-golden-globe, finance-real-estate, and news-politics, play an important role in understanding news content, inspiring us to enhance the categories' descriptions. In this paper, we propose a novel method that automatically generates informative category descriptions using a large language model (LLM) without manual effort or domain-specific knowledge and incorporates them into recommendation models as additional information. In our comprehensive experimental evaluations using the MIND dataset, our method successfully achieved 5.8% improvement at most in AUC compared with baseline approaches without the LLM's generated category descriptions for the state-of-the-art content-based recommendation models including NAML, NRMS, and NPA. These results validate the effectiveness of our approach. The code is available at https://github.com/yamanalab/gpt-augmented-news-recommendation.

News Recommendation with Category Description by a Large Language Model

TL;DR

The paper tackles information overload in online news by enriching content features with automatically generated category descriptions produced by a large language model. It integrates these LLM-generated descriptions with existing PLM-based news encoders, using a concatenation strategy with a SEP token to form enhanced representations. Experiments on the MIND dataset demonstrate consistent AUC gains of up to about 5.8% over strong baselines, underscoring the value of category-level semantic enrichment for recommendations. The approach shows promise for improving content understanding in news systems, though it notes limitations where generated descriptions may occasionally misalign with category scope and calls for future improvements in description quality. Code for reproduction is provided, enabling broader adoption and evaluation.

Abstract

Personalized news recommendations are essential for online news platforms to assist users in discovering news articles that match their interests from a vast amount of online content. Appropriately encoded content features, such as text, categories, and images, are essential for recommendations. Among these features, news categories, such as tv-golden-globe, finance-real-estate, and news-politics, play an important role in understanding news content, inspiring us to enhance the categories' descriptions. In this paper, we propose a novel method that automatically generates informative category descriptions using a large language model (LLM) without manual effort or domain-specific knowledge and incorporates them into recommendation models as additional information. In our comprehensive experimental evaluations using the MIND dataset, our method successfully achieved 5.8% improvement at most in AUC compared with baseline approaches without the LLM's generated category descriptions for the state-of-the-art content-based recommendation models including NAML, NRMS, and NPA. These results validate the effectiveness of our approach. The code is available at https://github.com/yamanalab/gpt-augmented-news-recommendation.
Paper Structure (12 sections, 5 figures, 3 tables)

This paper contains 12 sections, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. News Recommendations
  4. Use of LLM-Generated Texts
  5. Proposed Method
  6. Generation of Category Descriptions
  7. Integration of Category Descriptions into News Recommendation Models
  8. Experiments and Results
  9. Experimental Setting
  10. Results
  11. Limitations
  12. Conclusion

Figures (5)

  • Figure 1: Neural News Recommendation Model
  • Figure 2: Overview of our proposed method
  • Figure 3: Prompt to generate a category description for tv-golden-glove as an example.
  • Figure 4: Category description for tv-golden-globes via GPT-4
  • Figure 5: Category description for the tunedin category