Text-based automatic personality prediction using KGrAt-Net: a knowledge graph attention network classifier.

Area of Science:

• Computational Linguistics
• Artificial Intelligence
• Psychological Informatics

Background:

• Human communication increasingly occurs online, creating vast amounts of text data.
• Automatic Personality Prediction (APP) from text can improve interpersonal understanding and relationships.
• Existing methods often lack deep semantic understanding of textual content.

Purpose of the Study:

• To propose KGrAt-Net, a novel Knowledge Graph Attention Network (KGAT) for text classification.
• To apply KGAT for the first time to Automatic Personality Prediction (APP) based on the Big Five personality traits.
• To enhance the accuracy and semantic richness of personality prediction models.

Main Methods:

• Developed KGrAt-Net, a text classifier integrating knowledge graphs and attention mechanisms.
• Constructed knowledge graphs from input text to capture conceptual and semantic relationships.
• Employed an attention mechanism to focus on relevant parts of the knowledge graph for prediction.
• Utilized knowledge graph embedding to further enrich the classification process.

Main Results:

• KGrAt-Net achieved an average accuracy of 70.26% in Automatic Personality Prediction.
• Incorporating knowledge graph embedding further improved prediction accuracy to an average of 72.41%.
• The model demonstrated significant improvements over existing methods in personality prediction tasks.

Conclusions:

• KGrAt-Net offers a robust and effective method for Automatic Personality Prediction using text.
• The integration of knowledge graphs and attention mechanisms provides deeper semantic insights.
• This approach holds promise for applications requiring accurate personality assessment from digital communications.