Are Hypervectors Enough? Single-Call LLM Reasoning over Knowledge Graphs

2025-12-10

AI Summary

Existing LLM-based knowledge graph (KG) reasoning methods rely on multi-turn LLM invocations or heavy neural encoders, resulting in high latency, substantial GPU memory overhead, and opaque decision-making. This paper proposes PathHD—the first single-invocation LLM-based KG reasoning framework grounded in hyperdimensional computing (HDC). Its core innovations include: (i) an order-sensitive, non-commutative binding operator; (ii) block-diagonal GHRR-based hypervector encoding; (iii) block-level cosine similarity retrieval with Top-K pruning; and (iv) a one-stage prompt-driven, interpretable adjudication mechanism. On WebQSP, CWQ, and GrailQA, PathHD achieves Hits@1 scores competitive with or superior to strong neural baselines. It reduces end-to-end latency by 40–60% and cuts GPU memory usage by 3–5×. Crucially, its output paths explicitly encode reasoning justifications, ensuring both faithfulness and diagnostic interpretability.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Data Mining & Knowledge Management: Intelligent Query Processing

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Recent advances in large language models (LLMs) have enabled strong reasoning over both structured and unstructured knowledge. When grounded on knowledge graphs (KGs), however, prevailing pipelines rely on heavy neural encoders to embed and score symbolic paths or on repeated LLM calls to rank candidates, leading to high latency, GPU cost, and opaque decisions that hinder faithful, scalable deployment. We propose PathHD, a lightweight and encoder-free KG reasoning framework that replaces neural path scoring with hyperdimensional computing (HDC) and uses only a single LLM call per query. PathHD encodes relation paths into block-diagonal GHRR hypervectors, ranks candidates with blockwise cosine similarity and Top-K pruning, and then performs a one-shot LLM adjudication to produce the final answer together with cited supporting paths. Technically, PathHD is built on three ingredients: (i) an order-aware, non-commutative binding operator for path composition, (ii) a calibrated similarity for robust hypervector-based retrieval, and (iii) a one-shot adjudication step that preserves interpretability while eliminating per-path LLM scoring. On WebQSP, CWQ, and the GrailQA split, PathHD (i) attains comparable or better Hits@1 than strong neural baselines while using one LLM call per query; (ii) reduces end-to-end latency by $40-60%$ and GPU memory by $3-5 imes$ thanks to encoder-free retrieval; and (iii) delivers faithful, path-grounded rationales that improve error diagnosis and controllability. These results indicate that carefully designed HDC representations provide a practical substrate for efficient KG-LLM reasoning, offering a favorable accuracy-efficiency-interpretability trade-off.

Problem

• Reduces latency and GPU cost in KG reasoning
• Replaces neural path scoring with hyperdimensional computing
• Uses single LLM call per query for efficiency

Innovation

• Hyperdimensional computing replaces neural path scoring
• Single LLM call per query for efficiency
• Block-diagonal hypervectors enable interpretable path ranking