HG-PerCon: Cross-view contrastive learning for personality prediction

Area of Science:

• Computational Social Science
• Psychological Informatics

Background:

• Personality prediction is crucial for understanding user behavior in fields like psychology and behavioral economics.
• Current methods primarily analyze user posts or psychological data, but often overlook personality's long-term consistency and cross-view clue integration.

Purpose of the Study:

• To develop an efficient and effective model for personality prediction that addresses the limitations of existing approaches.
• To capture consistent, long-lasting personality-related information across diverse user data views.

Main Methods:

• Proposed HG-PerCon model utilizing user representations from historical semantic information and psychological knowledge.
• Employed a transformer-based module for long-term personality information extraction from user posts.
• Leveraged a psychological knowledge graph with language styles for knowledge-guided user representations.
• Utilized cross-view contrastive learning to ensure consistency of personality clues.

Main Results:

• The HG-PerCon model demonstrated superior performance in personality prediction.
• Achieved significant reductions in Root Mean Square Error (RMSE) compared to baseline methods, with improvements of 2%, 4%, and 6%.

Conclusions:

• HG-PerCon effectively integrates historical semantic data and psychological knowledge for robust personality prediction.
• The cross-view contrastive learning approach enhances the capture of consistent personality traits over time.