A Hybrid Recommendation Framework for Enhancing User Engagement in Local News

2025-08-26

AI Summary

Local news organizations face declining reader engagement amid intensifying competition from global media, while existing recommender systems struggle to jointly model users’ fine-grained local interests and global preferences. To address this, we propose a hybrid recommendation framework that unifies local and global preference modeling. Our approach introduces an adaptive ensemble strategy and a multi-stage training mechanism to jointly optimize localized and non-localized predictors within a single architecture. We further enhance generalization via LLM-based annotation, synthetic data augmentation, and multi-model ensembling. Experiments demonstrate that our method significantly outperforms single-model baselines in both accuracy and coverage, markedly improving recommendation relevance. Consequently, it drives measurable gains in user retention and subscription conversion. The framework offers a scalable, highly adaptable solution for personalized recommendation in local media contexts.

Technology Category

• Search and Optimization: Local Search
• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Local news organizations face an urgent need to boost reader engagement amid declining circulation and competition from global media. Personalized news recommender systems offer a promising solution by tailoring content to user interests. Yet, conventional approaches often emphasize general preferences and may overlook nuanced or eclectic interests in local news. We propose a hybrid news recommender that integrates local and global preference models to improve engagement. Building on evidence of the value of localized models, our method unifies local and non-local predictors in one framework. The system adaptively combines recommendations from a local model, specialized in region-specific content, and a global model that captures broader preferences. Ensemble strategies and multiphase training balance the two. We evaluated the model on two datasets: a synthetic set based on Syracuse newspaper distributions and a Danish dataset (EB-NeRD) labeled for local and non-local content with an LLM. Results show our integrated approach outperforms single-model baselines in accuracy and coverage, suggesting improved personalization that can drive user engagement. The findings have practical implications for publishers, especially local outlets. By leveraging both community-specific and general user interests, the hybrid recommender can deliver more relevant content, increasing retention and subscriptions. In sum, this work introduces a new direction for recommender systems, bridging local and global models to revitalize local news consumption through scalable, personalized user experiences.

Problem

• Enhancing user engagement in local news through personalized recommendations
• Integrating local and global preference models to improve content relevance
• Addressing declining readership by balancing community-specific and general interests

Innovation

• Hybrid model combining local and global preferences
• Ensemble strategies with multiphase training approach
• LLM-labeled datasets for local/non-local content classification