Scholarlink

This paper discusses the application of pre-trained language models for the analysis of sentiments in the financial domain. We define the methodology using pre-trained language models, which were implemented using the encoder part of the transformer network. In this work, we analyzed BERT, ALBERT, DistilBERT, and RoBERTa. The proposed methodology, designed using the aforementioned language models, involves a fine-tuning step followed by a few-shot learning step to output the label using the trained knowledge. We evaluated our approach using the FiQA and Financial PhraseBank datasets. The effectiveness of the proposed methodology outperformed the state-of-the-art model in the positive category (0.958, an improvement of approximately 10.25%) and the neutral category (0.887, an improvement of approximately 4.23%). However, the proposed method’s performance is comparable in the negative class.

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

Enhancing Financial Sentiment Analysis with Pre-trained BERT-Based Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • V. Harish Vijay,
  • B. Premjith

摘要

This paper discusses the application of pre-trained language models for the analysis of sentiments in the financial domain. We define the methodology using pre-trained language models, which were implemented using the encoder part of the transformer network. In this work, we analyzed BERT, ALBERT, DistilBERT, and RoBERTa. The proposed methodology, designed using the aforementioned language models, involves a fine-tuning step followed by a few-shot learning step to output the label using the trained knowledge. We evaluated our approach using the FiQA and Financial PhraseBank datasets. The effectiveness of the proposed methodology outperformed the state-of-the-art model in the positive category (0.958, an improvement of approximately 10.25%) and the neutral category (0.887, an improvement of approximately 4.23%). However, the proposed method’s performance is comparable in the negative class.