PLACE: Prompt Learning Framework for Attributed Community Search
A new framework called PLACE (Prompt Learning for Attributed Community Search) has been introduced by researchers for attributed community search (ACS). Drawing inspiration from prompt-tuning techniques in natural language processing (NLP), PLACE incorporates structural and learnable prompt tokens into the graph, serving as a query-dependent refinement tool and creating a prompt-augmented graph. These learned tokens enhance the relationships between graph nodes relevant to the query, allowing a graph neural network (GNN) to more effectively detect patterns of structural cohesiveness and attribute similarity. An alternating training approach optimizes both the prompt parameters and the GNN simultaneously, while a divide-and-conquer method improves scalability for graphs with millions of nodes. The research is available on arXiv with the ID 2507.05311.
Key facts
- PLACE stands for Prompt Learning for Attributed Community Search.
- It is a graph prompt learning framework for ACS.
- The framework is inspired by prompt-tuning in NLP.
- Learnable prompt tokens are inserted into the graph.
- The prompt-augmented graph refines query-dependent connections.
- An alternating training paradigm optimizes prompt parameters and GNN jointly.
- A divide-and-conquer strategy improves scalability to million-scale graphs.
- The paper is available on arXiv with ID 2507.05311.
Entities
Institutions
- arXiv