Abstract:Heterogeneous information networks (HIN) have gained increasing popularity in recent years for capturing complex relations between diverse types of nodes. Meta-structures are proposed as a useful tool to identify the important patterns in HINs, but hand-crafted meta-structures pose significant challenges for scaling up, drawing wide research attention towards developing automatic search algorithms. Previous efforts primarily focused on searching for meta-structures with good empirical performance, overlooking the importance of human comprehensibility and generalizability. To address this challenge, we draw inspiration from the emergent reasoning abilities of large language models (LLMs). We propose ReStruct, a meta-structure search framework that integrates LLM reasoning into the evolutionary procedure. ReStruct uses a grammar translator to encode the meta-structures into natural language sentences, and leverages the reasoning power of LLMs to evaluate their semantic feasibility. Besides, ReStruct also employs performance-oriented evolutionary operations. These two competing forces allow ReStruct to jointly optimize the semantic explainability and empirical performance of meta-structures. Furthermore, ReStruct contains a differential LLM explainer to generate and refine natural language explanations for the discovered meta-structures by reasoning through the search history. Experiments on eight representative HIN datasets demonstrate that ReStruct achieves state-of-the-art performance in both recommendation and node classification tasks. Moreover, a survey study involving 73 graduate students shows that the discovered meta-structures and generated explanations by ReStruct are substantially more comprehensible. Our code and questionnaire are available at <a class="link-external link-https" href="https://github.com/LinChen-65/ReStruct" rel="external noopener nofollow">this https URL</a>.

LLM4HIN: Discovering Meta-path with Large Language Model for Reasoning on Complex Heterogeneous Information Networks

EvoPath: Evolutionary meta-path discovery with large language models for complex heterogeneous information networks

Inductive Meta-path Learning for Schema-complex Heterogeneous Information Networks

Large Language Model-driven Meta-structure Discovery in Heterogeneous Information Network

Meta-Path Constrained Random Walk Inference for Large-Scale Heterogeneous Information Networks

ABLE: Meta-Path Prediction in Heterogeneous Information Networks.

M-HIN: Complex Embeddings for Heterogeneous Information Networks via Metagraphs

Classification with Active Learning and Meta-Paths in Heterogeneous Information Networks

Long-range Meta-path Search on Large-scale Heterogeneous Graphs

A Hierarchical Language Model For Interpretable Graph Reasoning

MetaFill: Text Infilling for Meta-Path Generation on Heterogeneous Information Networks

Leveraging LLMs for Hypothetical Deduction in Logical Inference: A Neuro-Symbolic Approach

LLM-A*: Large Language Model Enhanced Incremental Heuristic Search on Path Planning

Few-shot Learning for Heterogeneous Information Networks

Meta-Path and Hypergraph Fused Distillation Framework for Heterogeneous Information Networks Embedding

Path-Based Reasoning over Heterogeneous Networks for Recommendation via Bidirectional Modeling

RL4HIN: Representation Learning for Heterogeneous Information Networks.

DiffMG: Differentiable Meta Graph Search for Heterogeneous Graph Neural Networks

HAHE: Hierarchical Attentive Heterogeneous Information Network Embedding.

Path-LLM: A Shortest-Path-based LLM Learning for Unified Graph Representation

Differentiable Meta Multigraph Search with Partial Message Propagation on Heterogeneous Information Networks