Research map / Label-efficient and robust learning / Semi-supervised learning

Data Augmented Graph Neural Networks for Personality Detection

AAAI 2024 · Label-efficient and robust learning › Semi-supervised learning · approach: Self-training theory · 10 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Semi-PerGCN builds heterogeneous user-word-LIWC graphs and trains a GCN for Big Five personality regression on labeled users, adding a consistency loss over LIWC-based augmented graphs of unlabeled users.

Problem

Personality detection from social media posts is limited by scarce labeled data, since Big Five questionnaires are time-consuming to collect, so models generalize poorly to unseen users.

Approach

Combine a supervised personality graph neural network over user, word, and LIWC-category nodes with an unsupervised consistency loss on unlabeled users, where augmentations (LIWC synonym replacement and random word deletion) are derived from psycholinguistic knowledge.

Key results

• Semi-PerGCN reaches average MAE 64.21% on Youtube, 12.98% on PAN2015, and 57.13% on MyPersonality, the best of all compared methods.
• Relative to the strongest baseline GRU+Attn, Semi-PerGCN improves average MAE by 1.51% (Youtube), 3.49% (PAN2015), and 1.32% (MyPersonality).
• Removing the consistency regularization raises Youtube average MAE by 1.27%, and removing LIWC nodes raises it by 1.65%.

Measured effects reported in the paper

Surprising findings

• LIWC features alone outperform the CNN+LIWC combined model on Youtube and the other datasets.
• Adding more unlabeled data shows diminishing returns, with an upper limit on what can be extracted.

Limitations

• Gains are modest (about 1-3.5% MAE), and the paper gives no significance testing.
• Only two augmentation types (LIWC synonym replacement and random deletion) are explored.
• Authors note contrastive learning and other metric learning methods are left for future work.

Experimental setup

Datasets: Youtube Personality PAN2015 MyPersonality LIWC 2015

Benchmarks: Youtube Personality PAN2015 MyPersonality

Metrics: MAE

Baselines: LIWC TextCNN TextCNN+LIWC 2CLSTM BERT Attn TrigNet D-DGCN

Builds on: TrigNet D-DGCN Graph Convolutional Networks UDA BERT

Closest papers by meaning

• Knowledge-Enhanced Hierarchical Heterogeneous Graph for Personality Identification with Limited Training Data (AAAI 2025 · similarity 0.90)
• LLM vs Small Model? Large Language Model Based Text Augmentation Enhanced Personality Detection Model (AAAI 2024 · similarity 0.87)
• Ask, Answer, and Detect: Role-Playing LLMs for Personality Detection with Question-Conditioned Mixture-of-Experts (NeurIPS 2026 · similarity 0.87)
• Leveraging the Dual Capabilities of LLM: LLM-Enhanced Text Mapping Model for Personality Detection (AAAI 2025 · similarity 0.87)
• PersonaX: Multimodal Datasets with LLM-Inferred Behavior Traits (ICLR 2026 · similarity 0.83)
• Theoretical and Empirical Insights into the Origins of Degree Bias in Graph Neural Networks (NeurIPS 2024 · similarity 0.82)