Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sentence-level Media Bias Analysis with Event Relation Graph

Yuanyuan Lei, Ruihong Huang

TL;DR

This work tackles sentence-level media bias detection by introducing an event relation graph that encodes cross-event associations through coreference, temporal, causal, and subevent relations. A two-step framework combines an event-aware language model trained with soft labels from the graph and a relation-aware graph encoder trained with hard labels to update sentence embeddings, enabling precise bias identification. Evaluations on BASIL and BiasedSents show consistent improvements in precision and recall, with ablations confirming the complementary value of soft and hard labels and all four relation types. The approach advances bias analysis by leveraging document-wide event structure, offering more nuanced detection and potential for interpretable bias reasoning, while highlighting future work in handling implicit relations and enhancing interpretability.

Abstract

Media outlets are becoming more partisan and polarized nowadays. In this paper, we identify media bias at the sentence level, and pinpoint bias sentences that intend to sway readers' opinions. As bias sentences are often expressed in a neutral and factual way, considering broader context outside a sentence can help reveal the bias. In particular, we observe that events in a bias sentence need to be understood in associations with other events in the document. Therefore, we propose to construct an event relation graph to explicitly reason about event-event relations for sentence-level bias identification. The designed event relation graph consists of events as nodes and four common types of event relations: coreference, temporal, causal, and subevent relations. Then, we incorporate event relation graph for bias sentences identification in two steps: an event-aware language model is built to inject the events and event relations knowledge into the basic language model via soft labels; further, a relation-aware graph attention network is designed to update sentence embedding with events and event relations information based on hard labels. Experiments on two benchmark datasets demonstrate that our approach with the aid of event relation graph improves both precision and recall of bias sentence identification.

Sentence-level Media Bias Analysis with Event Relation Graph

TL;DR

This work tackles sentence-level media bias detection by introducing an event relation graph that encodes cross-event associations through coreference, temporal, causal, and subevent relations. A two-step framework combines an event-aware language model trained with soft labels from the graph and a relation-aware graph encoder trained with hard labels to update sentence embeddings, enabling precise bias identification. Evaluations on BASIL and BiasedSents show consistent improvements in precision and recall, with ablations confirming the complementary value of soft and hard labels and all four relation types. The approach advances bias analysis by leveraging document-wide event structure, offering more nuanced detection and potential for interpretable bias reasoning, while highlighting future work in handling implicit relations and enhancing interpretability.

Abstract

Media outlets are becoming more partisan and polarized nowadays. In this paper, we identify media bias at the sentence level, and pinpoint bias sentences that intend to sway readers' opinions. As bias sentences are often expressed in a neutral and factual way, considering broader context outside a sentence can help reveal the bias. In particular, we observe that events in a bias sentence need to be understood in associations with other events in the document. Therefore, we propose to construct an event relation graph to explicitly reason about event-event relations for sentence-level bias identification. The designed event relation graph consists of events as nodes and four common types of event relations: coreference, temporal, causal, and subevent relations. Then, we incorporate event relation graph for bias sentences identification in two steps: an event-aware language model is built to inject the events and event relations knowledge into the basic language model via soft labels; further, a relation-aware graph attention network is designed to update sentence embedding with events and event relations information based on hard labels. Experiments on two benchmark datasets demonstrate that our approach with the aid of event relation graph improves both precision and recall of bias sentence identification.
Paper Structure (21 sections, 20 equations, 3 figures, 6 tables)

This paper contains 21 sections, 20 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Event Relation Graph
  4. Event Relation Graph Components
  5. Event Relation Graph Training
  6. Event Relation Graph Construction
  7. Bias Sentences Identification
  8. Event-Aware Language Model
  9. Event Relation Graph Encoder
  10. Experiments
  11. Datasets
  12. Evaluation of Event Relation Graph
  13. Experimental Settings
  14. Baselines
  15. Experimental Results
  16. ...and 6 more sections

Figures (3)

  • Figure 1: An example article containing bias sentences, and its corresponding event relation graph. Bias sentences are highlighted in red. Events words are shown in bold text. Event relation graph consists of events as nodes and four types of event relations: coreference, temporal, causal, and subevent relation.
  • Figure 2: An illustration of sentence-level media bias identification based on event relation graph
  • Figure 3: Example of our method succeed in solving false negative error, and a failing example. Bias sentences are highlighted in red. Events words are shown in bold text. The solid arrows in the event relation graphs represent the successfully extracted event relations, and the dashed arrow means the missing event relation.