Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Three-Class Text Sentiment Analysis Based on LSTM

Yin Qixuan

TL;DR

This paper tackles three-class sentiment analysis of Chinese Weibo comments using an LSTM-based architecture to capture long-range contextual dependencies and nuanced emotional cues. It describes a pipeline with tokenization, embedding, and a Softmax-based classifier, and reports high performance with an accuracy of 98.31% and F1 of 98.28%, surpassing traditional machine learning and other deep learning baselines. The work demonstrates the effectiveness of LSTM for social-media sentiment tasks and discusses practical limitations such as computational cost and handling sarcasm, proposing future directions like integrating pre-trained models or enhanced feature engineering. Overall, the approach provides a robust, data-driven solution for fine-grained sentiment classification on large-scale Weibo comment data with potential applicability to other Chinese social media corpora.

Abstract

Sentiment analysis is a crucial task in natural language processing (NLP) with applications in public opinion monitoring, market research, and beyond. This paper introduces a three-class sentiment classification method for Weibo comments using Long Short-Term Memory (LSTM) networks to discern positive, neutral, and negative sentiments. LSTM, as a deep learning model, excels at capturing long-distance dependencies in text data, providing significant advantages over traditional machine learning approaches. Through preprocessing and feature extraction from Weibo comment texts, our LSTM model achieves precise sentiment prediction. Experimental results demonstrate superior performance, achieving an accuracy of 98.31% and an F1 score of 98.28%, notably outperforming conventional models and other deep learning methods. This underscores the effectiveness of LSTM in capturing nuanced sentiment information within text, thereby enhancing classification accuracy. Despite its strengths, the LSTM model faces challenges such as high computational complexity and slower processing times for lengthy texts. Moreover, complex emotional expressions like sarcasm and humor pose additional difficulties. Future work could explore combining pre-trained models or advancing feature engineering techniques to further improve both accuracy and practicality. Overall, this study provides an effective solution for sentiment analysis on Weibo comments.

Three-Class Text Sentiment Analysis Based on LSTM

TL;DR

This paper tackles three-class sentiment analysis of Chinese Weibo comments using an LSTM-based architecture to capture long-range contextual dependencies and nuanced emotional cues. It describes a pipeline with tokenization, embedding, and a Softmax-based classifier, and reports high performance with an accuracy of 98.31% and F1 of 98.28%, surpassing traditional machine learning and other deep learning baselines. The work demonstrates the effectiveness of LSTM for social-media sentiment tasks and discusses practical limitations such as computational cost and handling sarcasm, proposing future directions like integrating pre-trained models or enhanced feature engineering. Overall, the approach provides a robust, data-driven solution for fine-grained sentiment classification on large-scale Weibo comment data with potential applicability to other Chinese social media corpora.

Abstract

Sentiment analysis is a crucial task in natural language processing (NLP) with applications in public opinion monitoring, market research, and beyond. This paper introduces a three-class sentiment classification method for Weibo comments using Long Short-Term Memory (LSTM) networks to discern positive, neutral, and negative sentiments. LSTM, as a deep learning model, excels at capturing long-distance dependencies in text data, providing significant advantages over traditional machine learning approaches. Through preprocessing and feature extraction from Weibo comment texts, our LSTM model achieves precise sentiment prediction. Experimental results demonstrate superior performance, achieving an accuracy of 98.31% and an F1 score of 98.28%, notably outperforming conventional models and other deep learning methods. This underscores the effectiveness of LSTM in capturing nuanced sentiment information within text, thereby enhancing classification accuracy. Despite its strengths, the LSTM model faces challenges such as high computational complexity and slower processing times for lengthy texts. Moreover, complex emotional expressions like sarcasm and humor pose additional difficulties. Future work could explore combining pre-trained models or advancing feature engineering techniques to further improve both accuracy and practicality. Overall, this study provides an effective solution for sentiment analysis on Weibo comments.

Paper Structure

This paper contains 17 sections, 10 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Sentiment Analysis Methods Based on Sentiment Dictionaries/Rules
  4. Sentiment Analysis Methods Based on Traditional Machine Learning
  5. Sentiment Analysis Methods Based on Deep Learning
  6. Model Design and Construction
  7. LSTM Model Architecture
  8. Input Layer Structure
  9. LSTM Layer
  10. Output Layer
  11. Experiments and Results Analysis
  12. Data Preparation
  13. Experimental Environment Setup
  14. Parameter Settings
  15. Evaluation Metrics
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Diagram of RNN and LSTM cell structures
  • Figure 2: Forget Gate
  • Figure 3: ct
  • Figure 4: Output Gate
  • Figure 5: LSTM