Scholarlink

<p>In today's era of big data, there is a growing need for computational techniques that can streamline people’s lives by generating, processing, and efficiently understanding textual data. Natural Language Processing (NLP), a branch of Artificial Intelligence, is continuously advancing in the development of deep learning models for these purposes. One crucial aspect of NLP is the detection of negation and speculation in text, particularly in a morphologically rich language like Arabic, as these phenomena can significantly alter the polarity and factuality of the text’s meaning. Addressing negation and speculation is crucial for improving the effectiveness of various NLP applications, including sentiment analysis, machine translation, and biomedical information retrieval. While many research studies have explored these challenges in English, Spanish, and Chinese, Arabic remains unexplored due to its complexity and lack of annotated corpora. This paper proposes a Multi-Task Learning (MTL) model that classifies sentiment analysis as the main task while using negation and speculation detection as auxiliary tasks to enhance contextual understanding of the main task. We trained, validated, and tested the proposed model using the Negation Speculation Arabic Review (NSAR) corpus, a pre-annotated corpus for negation and speculation. The experimental results demonstrate that our model achieves an enhancement of 5% and 3% in F1 over the baseline (F1 = 72%) for negation and speculation, respectively, when evaluated on a well-known benchmarked Arabic sentiment analysis dataset.</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

Improving Arabic sentiment analysis with negation and speculation auxiliary tasks

  • 期刊论文

  • 发表日期：

    2025年8月3日

查看全文
  • Ahmed Mahany,
  • Said Ghoniemy,
  • Heba Khaled

摘要

In today's era of big data, there is a growing need for computational techniques that can streamline people’s lives by generating, processing, and efficiently understanding textual data. Natural Language Processing (NLP), a branch of Artificial Intelligence, is continuously advancing in the development of deep learning models for these purposes. One crucial aspect of NLP is the detection of negation and speculation in text, particularly in a morphologically rich language like Arabic, as these phenomena can significantly alter the polarity and factuality of the text’s meaning. Addressing negation and speculation is crucial for improving the effectiveness of various NLP applications, including sentiment analysis, machine translation, and biomedical information retrieval. While many research studies have explored these challenges in English, Spanish, and Chinese, Arabic remains unexplored due to its complexity and lack of annotated corpora. This paper proposes a Multi-Task Learning (MTL) model that classifies sentiment analysis as the main task while using negation and speculation detection as auxiliary tasks to enhance contextual understanding of the main task. We trained, validated, and tested the proposed model using the Negation Speculation Arabic Review (NSAR) corpus, a pre-annotated corpus for negation and speculation. The experimental results demonstrate that our model achieves an enhancement of 5% and 3% in F1 over the baseline (F1 = 72%) for negation and speculation, respectively, when evaluated on a well-known benchmarked Arabic sentiment analysis dataset.