Scholarlink

This study investigates media framing in the coverage of the Israeli-Palestinian conflict across several decades, using advanced natural language processing (NLP) techniques to analyze sentiment, emotion, named entity recognition (NER), and semantic bias. Using a multi-method approach—integrating lexicon-based (VADER, TextBlob), transformer-based (RoBERTa), and word embedding (Word2Vec) models—we examine sentiment patterns, emotional valence, entity-level sentiment, and framing differences between headlines and article content. Additionally, we incorporate a transformer-based large language model (LLM) for emotion detection to complement the NRC Emotion Lexicon analysis. Results reveal that media tends to exhibit more negative sentiment towards Palestinian mentions than Israeli ones, with gaps increasing during conflict escalations (e.g., 2014 Gaza escalation, 2023 Hamas-Israel escalation). Emotion analysis highlights fear as the dominant emotion for both groups, with NRC showing fear 8.1% higher in Israeli contexts and LLM showing a smaller difference of 6.3%. Anger shows divergent trends, with NRC indicating 9.5% higher prevalence in Palestinian contexts, while LLM shows almost no difference (–0.2%). Sadness is consistently higher in Israeli contexts according to NRC (+71.2%), but LLM shows it as more prevalent in Palestinian contexts (–10.3%). These complementary findings demonstrate the robustness of combining lexicon-based and transformer-based methods for emotion detection. Headlines display greater sentiment polarity than article bodies, with a negative bias in 59.5% of cases, amplifying conflict-oriented framing. Word embeddings indicate stronger associations of violence-related terms, with the conflict escalations periods. Named Entity Analysis reveals that sentiment towards prominent individuals—such as political leaders and military figures—shifts significantly with conflict intensity, offering insight into how public figures are emotionally framed in media narratives. The findings highlight systematic framing differences that may influence public perception.

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

Media Bias in the Guardian: Sentiment and Semantics in Israeli-Palestinian Coverage

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ahmed Gamea,
  • Ahmed Hasan,
  • Abdelaziz Serour,
  • Moustafa Rezk,
  • Walid Gomaa

摘要

This study investigates media framing in the coverage of the Israeli-Palestinian conflict across several decades, using advanced natural language processing (NLP) techniques to analyze sentiment, emotion, named entity recognition (NER), and semantic bias. Using a multi-method approach—integrating lexicon-based (VADER, TextBlob), transformer-based (RoBERTa), and word embedding (Word2Vec) models—we examine sentiment patterns, emotional valence, entity-level sentiment, and framing differences between headlines and article content. Additionally, we incorporate a transformer-based large language model (LLM) for emotion detection to complement the NRC Emotion Lexicon analysis. Results reveal that media tends to exhibit more negative sentiment towards Palestinian mentions than Israeli ones, with gaps increasing during conflict escalations (e.g., 2014 Gaza escalation, 2023 Hamas-Israel escalation). Emotion analysis highlights fear as the dominant emotion for both groups, with NRC showing fear 8.1% higher in Israeli contexts and LLM showing a smaller difference of 6.3%. Anger shows divergent trends, with NRC indicating 9.5% higher prevalence in Palestinian contexts, while LLM shows almost no difference (–0.2%). Sadness is consistently higher in Israeli contexts according to NRC (+71.2%), but LLM shows it as more prevalent in Palestinian contexts (–10.3%). These complementary findings demonstrate the robustness of combining lexicon-based and transformer-based methods for emotion detection. Headlines display greater sentiment polarity than article bodies, with a negative bias in 59.5% of cases, amplifying conflict-oriented framing. Word embeddings indicate stronger associations of violence-related terms, with the conflict escalations periods. Named Entity Analysis reveals that sentiment towards prominent individuals—such as political leaders and military figures—shifts significantly with conflict intensity, offering insight into how public figures are emotionally framed in media narratives. The findings highlight systematic framing differences that may influence public perception.