Abstract:Reasoning over knowledge graphs (KGs) is a challenging task that requires a deep understanding of the complex relationships between entities and the underlying logic of their relations. Current approaches rely on learning geometries to embed entities in vector space for logical query operations, but they suffer from subpar performance on complex queries and dataset-specific representations. In this paper, we propose a novel decoupled approach, Language-guided Abstract Reasoning over Knowledge graphs (LARK), that formulates complex KG reasoning as a combination of contextual KG search and logical query reasoning, to leverage the strengths of graph extraction algorithms and large language models (LLM), respectively. Our experiments demonstrate that the proposed approach outperforms state-of-the-art KG reasoning methods on standard benchmark datasets across several logical query constructs, with significant performance gain for queries of higher complexity. Furthermore, we show that the performance of our approach improves proportionally to the increase in size of the underlying LLM, enabling the integration of the latest advancements in LLMs for logical reasoning over KGs. Our work presents a new direction for addressing the challenges of complex KG reasoning and paves the way for future research in this area.

What problem does this paper attempt to address?

This paper attempts to solve the complex logical reasoning problems on knowledge graphs (KGs). Specifically, the author points out that current methods have poor performance when dealing with complex queries, and these methods usually rely on the representation of specific datasets and are difficult to generalize to other knowledge graphs. In addition, existing methods also have scalability problems when dealing with large - scale knowledge graphs. To overcome these problems, the paper proposes a new decoupling method - Language - guided Abstract Reasoning over Knowledge graphs (LARK). LARK significantly improves the performance of complex queries on standard benchmark datasets by decomposing the complex KG reasoning task into two parts: context - KG search and logical query reasoning, combining the advantages of graph extraction algorithms and large language models (LLMs). The main contributions of LARK include: 1. Proposing the LARK model, which uses the reasoning ability of large language models to efficiently answer first - order logic (FOL) queries on knowledge graphs. 2. The model uses entities and relations in the query to find relevant sub - graph contexts in the abstract knowledge graph and performs chained reasoning through LLM prompts generated after decomposing the logical query. 3. Experimental results show that on standard KG datasets constructed by multiple logical queries, LARK improves the performance by 35% - 84% compared with the existing state - of - the - art methods. 4. Through the chained decomposition technique, when dealing with decomposed logical queries, LARK improves its performance by 20% - 33% compared with dealing with complex queries. In addition, the analysis shows that increasing the scale and design of LLMs can further improve the performance of LARK. In conclusion, this paper aims to provide a more effective method to solve the complex logical reasoning problems on knowledge graphs by combining the reasoning ability of large language models and the structural characteristics of knowledge graphs.

Complex Logical Reasoning over Knowledge Graphs using Large Language Models

Concise and Organized Perception Facilitates Large Language Models for Deductive Reasoning.

Thinking with Knowledge Graphs: Enhancing LLM Reasoning Through Structured Data

Graph-constrained Reasoning: Faithful Reasoning on Knowledge Graphs with Large Language Models

Reasoning on Graphs: Faithful and Interpretable Large Language Model Reasoning

Causal Reasoning in Large Language Models: A Knowledge Graph Approach

Retrieval and Reasoning on KGs: Integrate Knowledge Graphs into Large Language Models for Complex Question Answering

Chain-of-Knowledge: Integrating Knowledge Reasoning into Large Language Models by Learning from Knowledge Graphs

Large Language Models and Knowledge Graphs: Opportunities and Challenges

Think-on-Graph: Deep and Responsible Reasoning of Large Language Model on Knowledge Graph

KG-GPT: A General Framework for Reasoning on Knowledge Graphs Using Large Language Models

A Framework of Knowledge Graph-Enhanced Large Language Model Based on Question Decomposition and Atomic Retrieval

KARPA: A Training-free Method of Adapting Knowledge Graph as References for Large Language Model's Reasoning Path Aggregation

Multi-hop Question Answering over Knowledge Graphs using Large Language Models

KG-Agent: An Efficient Autonomous Agent Framework for Complex Reasoning over Knowledge Graph

On Exploring the Reasoning Capability of Large Language Models with Knowledge Graphs

Improving Complex Reasoning over Knowledge Graph with Logic-Aware Curriculum Tuning

Reasoning on Efficient Knowledge Paths:Knowledge Graph Guides Large Language Model for Domain Question Answering

Think-on-Graph: Deep and Responsible Reasoning of Large Language Model with Knowledge Graph

FiDeLiS: Faithful Reasoning in Large Language Model for Knowledge Graph Question Answering