Public Service Algorithm: towards a transparent, explainable, and scalable content curation for news content based on editorial values

TL;DR

This paper presents the Public Service Algorithm (PSA), a framework using Large Language Models to automate transparent, value-driven news content curation at scale, aligning with human editorial judgments and addressing disinformation challenges.

Contribution

The paper introduces PSA, a novel LLM-based framework for scalable, transparent news curation based on public service values, validated with European news datasets and expert comparisons.

Findings

LLMs show promising alignment with human editorial judgments.

PSA enables scalable, transparent content curation.

Potential to improve trustworthiness in news dissemination.

Abstract

The proliferation of disinformation challenges traditional, unscalable editorial processes and existing automated systems that prioritize engagement over public service values. To address this, we introduce the Public Service Algorithm (PSA), a novel framework using Large Language Models (LLMs) for scalable, transparent content curation based on Public Service Media (PSM) inspired values. Utilizing a large multilingual news dataset from the 'A European Perspective' project, our experiment directly compared article ratings from a panel of experienced editors from various European PSMs, with those from several LLMs, focusing on four criteria: diversity, in-depth analysis, forward-looking, and cross-border relevance. Utilizing criterion-specific prompts, our results indicate a promising alignment between human editorial judgment and LLM assessments, demonstrating the potential of LLMs to automate value-driven curation at scale without sacrificing transparency. This research constitutes a first step towards a scalable framework for the automatic curation of trustworthy news content.