Data Augmented Graph Neural Networks for Personality Detection

• DOI: 10.1609/aaai.v38i1.27823
• Publication Date: 2024-03-25T08:48:26Z
• Authors: Yangfu Zhu; Yue Xia; Meiling Li; Tingting Zhang; Bin Wu

Personality detection is a fundamental task for user psychology research. One of the biggest challenges in personality lies quantitative limitation labeled data collected by completing questionnaire, which very time-consuming and labor-intensive. Most existing works are mainly devoted to learning rich representations posts based on data. However, they still suffer from inherent weakness amount labels, potentially restricts capability model deal with unseen In this paper, we construct heterogeneous graph each unlabeled develop novel psycholinguistic augmented neural network detect semi-supervised manner, namely Semi-PerGCN. Specifically, our first explores supervised Graph Neural Network (PGNN) refine representation graph. For remaining massive users, utilize empirical psychological knowledge Linguistic Inquiry Word Count (LIWC) lexicon multi-view augmentation perform unsupervised consistent constraints parameters shared PGNN. During process finite noise-invariant large scale users combined enhance generalization ability. Extensive experiments three real-world datasets, Youtube, PAN2015, MyPersonality demonstrate effectiveness Semi-PerGCN detection, especially scenarios limited users.