Connecting Text Classification with Image Classification: A New Preprocessing Method for Implicit Sentiment Text...

Area of Science:

• Natural Language Processing
• Machine Learning
• Computer Vision

Background:

• Sentiment analysis is crucial in Natural Language Processing (NLP).
• Implicit sentiment analysis, lacking explicit emotion words, poses significant challenges for existing methods.
• Accurate identification of sentiment polarity in implicit text remains a difficult task.

Purpose of the Study:

• To propose a new preprocessing method for Chinese implicit sentiment text classification.
• To improve the performance of implicit sentiment analysis by highlighting sentiment differences.
• To address the limitations of current text-processing methods in identifying implicit sentiment polarity.

Main Methods:

• Developed a novel preprocessing technique named Text To Picture (TTP).
• Converted original text data into word frequency maps using deep learning.
• Utilized the visual differences in word frequency maps to identify sentiment clues.

Main Results:

• Experimental tests on SMP2019 and EWECT datasets demonstrated significant accuracy improvements.
• Accuracy on the SMP2019 dataset increased by 4.55-7.06%.
• Accuracy on the EWECT dataset improved by 1.81-3.95%.

Conclusions:

• The proposed Text To Picture (TTP) preprocessing method effectively enhances Chinese implicit sentiment text classification.
• TTP offers a promising approach to overcome the challenges in implicit sentiment analysis.
• The method achieves superior classification results compared to existing techniques.