A combination of TEXTCNN model and Bayesian classifier for microblog sentiment analysis

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• Microblog sentiment analysis is crucial for understanding public opinion.
• Existing TEXTCNN models lack interpretability and struggle with word polysemy.
• There's a need for improved methods to analyze emotional content in short texts.

Purpose of the Study:

• To develop an interpretable and extensible microblog sentiment analysis method.
• To address the limitations of standard TEXTCNN and word embedding in capturing context.
• To improve the accuracy of emotion detection in microblog comments.

Main Methods:

• Utilized word2vec for initial word embeddings.
• Integrated ELMo (Embeddings from Language Models) to generate context-aware word vectors.
• Employed TEXTCNN (Text Convolutional Neural Network) for feature extraction via convolution and pooling layers.
• Combined extracted features with a Bayes classifier for final emotion classification.

Main Results:

• The proposed TEXTCNN and Bayes model achieved high performance on the Stanford Sentiment Classification Corpus (SST) dataset.
• Achieved Accuracy of 0.9813, Precision of 0.9821, Recall of 0.9804, and F1-score of 0.9812.
• Outperformed standard TEXTCNN, LSTM, and LSTM-TEXTCNN models in sentiment analysis tasks.

Conclusions:

• The developed method effectively enhances microblog sentiment analysis accuracy.
• The integration of ELMo and Bayes classifier with TEXTCNN provides superior feature representation and classification.
• The model is suitable for accurate emotional analysis and event identification in microblogs.