# Beyond Prefixes: Graph-as-Memory Cross-Attention for Knowledge Graph Completion with Large Language Models

arXiv: https://arxiv.org/abs/2510.08966 · HTML: https://mlsanity.com/paper/2510.08966/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 39 references that are in the map.
- First posted on arXiv 10 Oct 2025.

## Builds on (most cited first, 23 of 23)

- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Embedding Entities and Relations for Learning and Inference in Knowledge Bases](https://mlsanity.com/paper/1412.6575/) (2014)
- [Complex Embeddings for Simple Link Prediction](https://mlsanity.com/paper/1606.06357/) (2016)
- [RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space](https://mlsanity.com/paper/1902.10197/) (2019)
- [Convolutional 2D Knowledge Graph Embeddings](https://mlsanity.com/paper/1707.01476/) (2017)
- [Unifying Large Language Models and Knowledge Graphs: A Roadmap](https://mlsanity.com/paper/2306.08302/) (2023)
- [Composition-based Multi-Relational Graph Convolutional Networks](https://mlsanity.com/paper/1911.03082/) (2019)
- [KG-BERT: BERT for Knowledge Graph Completion](https://mlsanity.com/paper/1909.03193/) (2019)
- [Neural Bellman-Ford Networks: A General Graph Neural Network Framework for Link Prediction](https://mlsanity.com/paper/2106.06935/) (2021)
- [Learning Hierarchy-Aware Knowledge Graph Embeddings for Link Prediction](https://mlsanity.com/paper/1911.09419/) (2019)
- [BoxE: A Box Embedding Model for Knowledge Base Completion](https://mlsanity.com/paper/2007.06267/) (2020)
- [SimKGC: Simple Contrastive Knowledge Graph Completion with Pre-trained Language Models](https://mlsanity.com/paper/2203.02167/) (2022)
- [Making Large Language Models Perform Better in Knowledge Graph Completion](https://mlsanity.com/paper/2310.06671/) (2023)
- [KICGPT: Large Language Model with Knowledge in Context for Knowledge Graph Completion](https://mlsanity.com/paper/2402.02389/) (2024)
- [A Universal Question-Answering Platform for Knowledge Graphs](https://mlsanity.com/paper/2303.00595/) (2023)
- [Data Scarcity in Recommendation Systems: A Survey](https://mlsanity.com/paper/2312.10073/) (2023)
- [MKGL: Mastery of a Three-Word Language](https://mlsanity.com/paper/2410.07526/) (2024)
- [KG-FIT: Knowledge Graph Fine-Tuning Upon Open-World Knowledge](https://mlsanity.com/paper/2405.16412/) (2024)
- [Path-based Explanation for Knowledge Graph Completion](https://mlsanity.com/paper/2401.02290/) (2024)
- [Self-supervised Quantized Representation for Seamlessly Integrating Knowledge Graphs with Large Language Models](https://mlsanity.com/paper/2501.18119/) (2025)
- [Comprehending Knowledge Graphs with Large Language Models for Recommender Systems](https://mlsanity.com/paper/2410.12229/) (2024)
- [GLTW: Joint Improved Graph Transformer and LLM via Three-Word Language for Knowledge Graph Completion](https://mlsanity.com/paper/2502.11471/) (2025)
- [Flow-Modulated Scoring for Semantic-Aware Knowledge Graph Completion](https://mlsanity.com/paper/2506.23137/) (2025)

## Abstract (arXiv)

Fusing Knowledge Graphs with Large Language Models (LLMs) is crucial for knowledge-intensive tasks like knowledge graph completion. Existing LLM-based approaches typically inject graph information via prefix concatenation, resulting in shallow interactions that fail to support fine-grained evidence retrieval during generation. Beyond prefixes, we propose Graph-as-Memory Tuning (GMT), a new paradigm that represents local graph structure as explicit graph memory and injects it into LLMs via deep, token-wise cross-attention. Specifically, GMT first employs a Semantic Graph Module to encode context-aware semantics from local neighborhoods guided by knowledge-enhanced relations, and compresses them into a fixed number of graph memory tokens. A Graph-as-Memory Cross-Attention Fusion Module then integrates these tokens into multiple Transformer layers, allowing LLM hidden state to dynamically retrieve relevant graph evidence. To enable efficient adaptation, GMT applies LoRA only to the memory cross-attention while keeping the base LLM frozen. Extensive experiments show that GMT significantly outperforms prefix-tuning and other strong baselines, providing more potent signals for robust reasoning. The code is published at https://github.com/tongruiliu/GMT.