Pre-trained Prompt-driven Community Search

2025-05-18

AI Summary

Existing semi-supervised community detection methods typically generalize from known communities and cannot guarantee that the output community contains a given query node, rendering them unsuitable for query-oriented community search. To address this, we propose the first “pre-training + prompting” framework for semi-supervised community search, comprising three stages: GNN-based node encoding, structural-similarity-driven initial community sampling, and prompt-guided fine-tuning. Our key contributions are: (1) the first prompt-learning community search model explicitly ensuring query-node membership; (2) a structure-aware sample generation mechanism leveraging topological proximity; and (3) an end-to-end prompt-guided community prediction pipeline. Extensive experiments on five real-world datasets demonstrate significant improvements over state-of-the-art baselines in both accuracy and efficiency—up to 37% faster search speed—while ablation studies confirm the effectiveness of each component.

Technology Category

• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Search and Optimization: Learning to Search
• Machine Learning: Semi-Supervised Learning

Application Category

• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Social Networks and Social Media: Social mining and social search on the Web

Abstract

The"pre-train, prompt"paradigm is widely adopted in various graph-based tasks and has shown promising performance in community detection. Most existing semi-supervised community detection algorithms detect communities based on known ones, and the detected communities typically do not contain the given query node. Therefore, they are not suitable for searching the community of a given node. Motivated by this, we adopt this paradigm into the semi-supervised community search for the first time and propose Pre-trained Prompt-driven Community Search (PPCS), a novel model designed to enhance search accuracy and efficiency. PPCS consists of three main components: node encoding, sample generation, and prompt-driven fine-tuning. Specifically, the node encoding component employs graph neural networks to learn local structural patterns of nodes in a graph, thereby obtaining representations for nodes and communities. Next, the sample generation component identifies an initial community for a given node and selects known communities that are structurally similar to the initial one as training samples. Finally, the prompt-driven fine-tuning component leverages these samples as prompts to guide the final community prediction. Experimental results on five real-world datasets demonstrate that PPCS performs better than baseline algorithms. It also achieves higher community search efficiency than semi-supervised community search baseline methods, with ablation studies verifying the effectiveness of each component of PPCS.

Problem

• Enhances community search accuracy and efficiency
• Addresses limitations of semi-supervised community detection
• Proposes prompt-driven fine-tuning for community prediction

Innovation

• Pre-trained prompt-driven community search model
• Graph neural networks for node encoding
• Prompt-driven fine-tuning for community prediction