Knowledge-Fusion-Based Iterative Graph Structure Learning Framework for Implicit Sentiment Identification.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Implicit sentiment identification is crucial for text analysis.
• Current Graph Neural Network (GNN) approaches struggle with limited structural information and noisy graph edges.
• These limitations hinder accurate capture of obscure sentiment expressions.

Purpose of the Study:

• To address limitations in current GNN-based implicit sentiment identification.
• To develop a framework that enhances structural information and optimizes graph topology.
• To improve the accuracy and robustness of implicit sentiment analysis.

Main Methods:

• Introduced a knowledge-fusion-based iterative graph structure learning framework (KIG).
• Constructed multi-view graph structures using co-occurrence statistics, cosine similarity, and syntactic dependency trees.
• Iteratively refined graph structures to better fit data and optimize sentiment analysis.

Main Results:

• KIG demonstrated superior performance compared to mainstream implicit sentiment identification methods.
• Achieved high accuracy (89.2%), recall (93.7%), and F1-score (91.1%) on the Pun of the Day dataset.
• Experimental results validate the effectiveness of the proposed method.

Conclusions:

• The proposed KIG framework effectively addresses limitations in existing GNN approaches for implicit sentiment analysis.
• Multi-view graph construction and iterative structure learning enhance the capture of subtle sentiment expressions.
• KIG offers a superior method for implicit sentiment identification tasks.