Scholarlink

<p>Financial results briefings are an important corporate disclosure for Japanese companies, and dialogue-style Q&amp;A sessions provide valuable information. However, their conversational and colloquial style is different from the typical targets of financial sentiment analysis, which presents challenges for sentiment classification. In this paper, we propose an approach that leverages large language models (LLMs) to enhance sentiment analysis of Q&amp;A sessions. We segment the Q&amp;A transcripts into conversational chunks and use ChatGPT to summarize and standardize their content, which enables more accurate sentiment classification with a BERT model fine-tuned for financial contexts. Using an event study on stock returns, we find that summary-based sentiment is significantly related to abnormal returns (ARs). We further compare BERT-based sentiment with sentiment scores directly assigned by ChatGPT, and find that both explain ARs. Our findings indicate that LLMs can effectively enhance sentiment analysis across various financial documents with different writing styles, including dialogue-style transcripts.</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

Extracting information and sentiment analysis on dialogue in financial results briefing

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Hiromasa Nakatsuka,
  • Yoshiyuki Suimon

摘要

Financial results briefings are an important corporate disclosure for Japanese companies, and dialogue-style Q&A sessions provide valuable information. However, their conversational and colloquial style is different from the typical targets of financial sentiment analysis, which presents challenges for sentiment classification. In this paper, we propose an approach that leverages large language models (LLMs) to enhance sentiment analysis of Q&A sessions. We segment the Q&A transcripts into conversational chunks and use ChatGPT to summarize and standardize their content, which enables more accurate sentiment classification with a BERT model fine-tuned for financial contexts. Using an event study on stock returns, we find that summary-based sentiment is significantly related to abnormal returns (ARs). We further compare BERT-based sentiment with sentiment scores directly assigned by ChatGPT, and find that both explain ARs. Our findings indicate that LLMs can effectively enhance sentiment analysis across various financial documents with different writing styles, including dialogue-style transcripts.