DocNet: Semantic Structure in Inductive Bias Detection Models

2024-06-16

arXiv.org

AI Summary

This work addresses political bias detection in news articles under low-resource settings. We propose DocNet, a novel model that first represents news documents as document-level graphs—capturing their semantic structure—and employs graph neural networks to learn structured embeddings via inductive learning for cross-document generalization. Our key insight is the high consistency in syntactic-semantic organizational patterns across news articles of differing political stances; leveraging this, DocNet constructs a lightweight, structure-aware classifier. Experiments demonstrate that DocNet significantly outperforms large language model (LLM) baselines with orders-of-magnitude more parameters under few-shot settings. Moreover, its structural inductive bias enhances both robustness and interpretability, enabling fine-grained, graph-based analysis of bias manifestations. This establishes a new structured paradigm for political bias detection, bridging linguistic form and ideological function in low-data regimes.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Structured Learning
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Social Networks and Social Media: Fairness and bias in social network and social media analysis

Abstract

News will have biases so long as people have opinions. It is increasingly important for informed citizens to be able to identify bias as social media becomes the primary entry point for news and partisan differences increase. If people know the biases of the news they are consuming, they will be able to take action to avoid polarizing echo chambers. In this paper, we explore an often overlooked aspect of bias detection in documents: the semantic structure of news articles. We present DocNet, a novel, inductive, and low-resource document embedding and bias detection model that outperforms large language models. We also demonstrate that the semantic structure of news articles from opposing partisan sides, as represented in document-level graph embeddings, have significant similarities. These results can be used to advance bias detection in low-resource environments. Our code, data, and the corresponding datasheet are made available at: https://anonymous.4open.science/r/DocNet/.

Problem

• Political Bias
• News Articles
• Organizational Structure

Innovation

• DocNet
• Political Bias Detection
• Simplified Linguistic Representation