Scholarlink

<p>In U.S. media coverage of China’s debt practices, the “blame game” functions as a primary discursive mechanism for assigning responsibility. However, existing research lacks a unified methodology to detect subtle, recurrent blame viewpoints within news texts. This study introduces a novel computational framework that operationalizes the “blame game” as a seven-dimensional, sentence-level construct. We employed large language models (LLMs) to extract blame viewpoints from news across three major U.S. publications (2017–2024), implementing a rigorous two-stage validation process. The analysis reveals systematic construction patterns: an oppositional stance was dominant (82.61%) while sentiment remained balanced (50.20% neutral), indicating a strategic decoupling of stance and sentiment. Furthermore, a persistent role asymmetry emerged, contrasting implicit accusers (68%) with consistently explicit accused entities, primarily China (49%). Frequent itemset mining demonstrated that a limited set of structural configurations accounted for the majority of blame instances, revealing underlying narrative templates. This study contributes a reproducible methodology that transforms the abstract “blame game” concept into measurable units, bridging macro-level framing theories with micro-linguistic analysis while enhancing the interpretability of LLM-based discourse analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LLM-Based Viewpoint Mining in the “Blame Game”: How U.S. Media Frame China’s Debt Debate

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Rende Li,
  • Ying Jiang

摘要

In U.S. media coverage of China’s debt practices, the “blame game” functions as a primary discursive mechanism for assigning responsibility. However, existing research lacks a unified methodology to detect subtle, recurrent blame viewpoints within news texts. This study introduces a novel computational framework that operationalizes the “blame game” as a seven-dimensional, sentence-level construct. We employed large language models (LLMs) to extract blame viewpoints from news across three major U.S. publications (2017–2024), implementing a rigorous two-stage validation process. The analysis reveals systematic construction patterns: an oppositional stance was dominant (82.61%) while sentiment remained balanced (50.20% neutral), indicating a strategic decoupling of stance and sentiment. Furthermore, a persistent role asymmetry emerged, contrasting implicit accusers (68%) with consistently explicit accused entities, primarily China (49%). Frequent itemset mining demonstrated that a limited set of structural configurations accounted for the majority of blame instances, revealing underlying narrative templates. This study contributes a reproducible methodology that transforms the abstract “blame game” concept into measurable units, bridging macro-level framing theories with micro-linguistic analysis while enhancing the interpretability of LLM-based discourse analysis.