Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LL4G: Self-Supervised Dynamic Optimization for Graph-Based Personality Detection

Lingzhi Shen, Yunfei Long, Xiaohao Cai, Guanming Chen, Yuhan Wang, Imran Razzak, Shoaib Jameel

TL;DR

This paper tackles MBTI personality detection from social-media text, addressing data sparsity and static graph limitations. It introduces LL4G, which uses LLMs to extract deep semantic features and to construct adaptive graphs with explicit and implicit edges, including a user node, and then applies DGCN-based representation learning. The model is trained with node reconstruction, edge prediction, and contrastive learning alongside supervised MBTI classification, with the objective L_final = L_class + λ (α L_node + β L_edge + γ L_contrastive). On Kaggle and Pandora, LL4G outperforms state-of-the-art baselines, demonstrating the value of integrating semantic reasoning with dynamic graph optimization for robust personality inference.

Abstract

Graph-based personality detection constructs graph structures from textual data, particularly social media posts. Current methods often struggle with sparse or noisy data and rely on static graphs, limiting their ability to capture dynamic changes between nodes and relationships. This paper introduces LL4G, a self-supervised framework leveraging large language models (LLMs) to optimize graph neural networks (GNNs). LLMs extract rich semantic features to generate node representations and to infer explicit and implicit relationships. The graph structure adaptively adds nodes and edges based on input data, continuously optimizing itself. The GNN then uses these optimized representations for joint training on node reconstruction, edge prediction, and contrastive learning tasks. This integration of semantic and structural information generates robust personality profiles. Experimental results on Kaggle and Pandora datasets show LL4G outperforms state-of-the-art models.

LL4G: Self-Supervised Dynamic Optimization for Graph-Based Personality Detection

TL;DR

This paper tackles MBTI personality detection from social-media text, addressing data sparsity and static graph limitations. It introduces LL4G, which uses LLMs to extract deep semantic features and to construct adaptive graphs with explicit and implicit edges, including a user node, and then applies DGCN-based representation learning. The model is trained with node reconstruction, edge prediction, and contrastive learning alongside supervised MBTI classification, with the objective L_final = L_class + λ (α L_node + β L_edge + γ L_contrastive). On Kaggle and Pandora, LL4G outperforms state-of-the-art baselines, demonstrating the value of integrating semantic reasoning with dynamic graph optimization for robust personality inference.

Abstract

Graph-based personality detection constructs graph structures from textual data, particularly social media posts. Current methods often struggle with sparse or noisy data and rely on static graphs, limiting their ability to capture dynamic changes between nodes and relationships. This paper introduces LL4G, a self-supervised framework leveraging large language models (LLMs) to optimize graph neural networks (GNNs). LLMs extract rich semantic features to generate node representations and to infer explicit and implicit relationships. The graph structure adaptively adds nodes and edges based on input data, continuously optimizing itself. The GNN then uses these optimized representations for joint training on node reconstruction, edge prediction, and contrastive learning tasks. This integration of semantic and structural information generates robust personality profiles. Experimental results on Kaggle and Pandora datasets show LL4G outperforms state-of-the-art models.

Paper Structure

This paper contains 12 sections, 15 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Node Semantic Features Extraction
  4. Graph Construction and Dynamic Update
  5. Graph Representation Learning
  6. Personality Prediction
  7. Experiments
  8. Experimental Setup
  9. Overall Results
  10. Ablation Study
  11. Case Study
  12. Conclusions

Figures (3)

  • Figure 1: The architecture diagram of the proposed LL4G. It illustrates the workflow pipeline, encompassing key stages such as node semantic extraction, graph construction and dynamic updates, graph representation learning, and personality prediction.
  • Figure 2: Performance comparison between the LL4G framework and various graph-based models across four dimensions and their overall average in the form of the stacked bar chart in terms of Macro-F1 (%) on the Kaggle dataset.
  • Figure 3: Performance comparison bewteen the LL4G framework and various graph-based models across four dimensions and their overall average (Avg) in the form of line chart in terms of Macro-F1 (%) on the Pandora dataset.