His research focuses on graph agent performance tuning based on large-scale graph data management and mining and graph database query optimization techniques, which can be divided into the following three directions:

Leveraging ultra-large-scale network motif identification and statistical significance testing to assist large models in retrieving target nodes in knowledge graphs and denoising the information aggregation process, thereby achieving interpretable extraction of multi-dimensional heterogeneous data:

1. Network motif identification based on large-scale subgraph discovery: BEACON@ICDE26, MuSha@ICDE25, MOSER@VLDB24, CSCE@ICDE24, DDS@VLDBJ23, LINC@VLDB20, SAC@TKDE18, Spatial CS@TKDE17

2. LLM + knowledge graph subgraph denoising: ZeroEA@VLDB24

3. Interpretable subgraph data extraction: Graph Rewiring@ICDE26, CD@ICDE24, Motif-Path@VLDB 2021

Optimizing graph retrieval strategy design and evaluating and optimizing graph reasoning processes for graph retrieval-augmented generation and graph agents:

1. Knowledge conflict resolution in retrieval re-ranking and tabular data analysis: CoTA@ICML25, Micro-Act@ACL25

2. Evaluation of intermediate reasoning processes for multi-hop question answering: PER@ACL25

Query optimization techniques on different graph database models and their applications:

1. Traffic graph models: Hypergraph Weight Completion@TKDD25, FDM@IS24, Weight Completion@SDM22, Incident Detection@ICDE20

2. Power graph models: Subgraph RL@IJCAI26, EERL@IJCAI25, SOC Estimation@A.Anergy25, GMRL@PR25

3. E-commerce graph models: HDRec@WSDM26, GenPlugin@DASFAA26, SC-DAG@CIKM25

4. Biological graph models: Drug Repurposing@A.Therapeutics21, MCypher@CIKM20

5. Probabilistic graph models: DURS@MDM19, UKNN@EDBT18