Deep Learning-Based Text Emotion Analysis for Legal Anomie

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Affective Computing

Background:

• Anomie behaviors require nuanced analysis, often expressed through text.
• Existing text emotion analysis methods may not fully capture complex emotional expressions.

Purpose of the Study:

• To propose a novel text emotion analysis framework (BCDF) for analyzing emotions in anomie behaviors.
• To enhance sentence representation by integrating word embeddings and part-of-speech tagging.
• To improve emotion classification accuracy using advanced deep learning techniques.

Main Methods:

• Developed the Bi-direction Convolutional Word Embedding Classification Framework (BCDF).
• Incorporated an emotional parallel learning mechanism using Bi-LSTM with gating.
• Utilized a Convolutional Long Short-Term Memory (ConvLSTM) network for feature extraction.

Main Results:

• The BCDF model demonstrated superior performance compared to traditional LSTM.
• Achieved a 3.3% F1 score increase on psychological text datasets.
• Achieved a 10.9% F1 score increase on news text datasets.

Conclusions:

• The proposed BCDF model, integrating Bi-LSTM and ConvLSTM, offers significant advancements in text emotion analysis.
• The framework shows practical value for analyzing anomie behaviors through text.
• The model effectively captures sentence components crucial for emotion expression.