Scholarlink

In contemporary times, as financial content proliferates across the internet and social networks, accurately predicting future trends has become an everyday necessity for providing optimal investment strategies. Sentiment Analysis (SA), a prominent subject in artificial intelligence, is pivotal in revealing people’s emotions and opinions on specific matters. This paper aims to leverage text-mining algorithms to categorize a text-based financial dataset through sentiment analysis. Furthermore, a novel hybrid feature selection model is introduced to enhance accuracy and performance when studying economic text. Initially, a widely recognized financial text dataset (FiQA) was chosen. After applying preprocessing techniques encompassing data cleansing and feature extraction, the feature pool is reduced by utilizing ANOVA, RFI, and CHI2 algorithms. Subsequently, the features are refined using the Particle Swarm Optimization (PSO) approach. In the subsequent stages, the text is classified by the Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), K-Nearest Neighbours (KNN), Naïve Bayes, and Support Vector Machine (SVM) algorithms, all of which yield notable performance outcomes. The results show that the ANOVA-PSO hybrid model for LSTM classification achieves an accuracy rate of 75%, superior to other Feature selection models.

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 a Hybrid Feature Selection Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ray Shams,
  • Javad Khosravian,
  • Parnia Samimi,
  • Faisal Saeed

摘要

In contemporary times, as financial content proliferates across the internet and social networks, accurately predicting future trends has become an everyday necessity for providing optimal investment strategies. Sentiment Analysis (SA), a prominent subject in artificial intelligence, is pivotal in revealing people’s emotions and opinions on specific matters. This paper aims to leverage text-mining algorithms to categorize a text-based financial dataset through sentiment analysis. Furthermore, a novel hybrid feature selection model is introduced to enhance accuracy and performance when studying economic text. Initially, a widely recognized financial text dataset (FiQA) was chosen. After applying preprocessing techniques encompassing data cleansing and feature extraction, the feature pool is reduced by utilizing ANOVA, RFI, and CHI2 algorithms. Subsequently, the features are refined using the Particle Swarm Optimization (PSO) approach. In the subsequent stages, the text is classified by the Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), K-Nearest Neighbours (KNN), Naïve Bayes, and Support Vector Machine (SVM) algorithms, all of which yield notable performance outcomes. The results show that the ANOVA-PSO hybrid model for LSTM classification achieves an accuracy rate of 75%, superior to other Feature selection models.