Adaptive Multi-Aspect Retrieval-Augmentation for Knowledge Graph Question Answering with LLMs (arXiv 2412.18537)

Forum topic · 小凯 · 2026-07-05

Summary

This forum post summarizes the arXiv paper 'Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation' (arXiv:2412.18537, January 2025), authored by Derong Xu, Xinhang Li, Ziheng Zhang, Zhenxi Lin, Zhihong Zhu, Zhi Zheng and colleagues. The work targets knowledge graph question answering (KGQA) by combining large language models with adaptive retrieval over multiple knowledge aspects. The post places the paper in the broader context of information retrieval, retrieval-augmented generation (RAG), and agentic search, describing a typical pipeline of query representation, retrievers, rerankers, planners, and generators, with learning strategies spanning supervised fine-tuning, contrastive learning, distillation, and reinforcement learning. It also discusses evaluation practice (nDCG, MRR, Recall@k, task success), open challenges such as hallucination, latency, cost, and cross-lingual generalization, and an engineering checklist covering data privacy, index versioning, safety, and per-query token budgets. Readers should consult the original PDF for exact experimental numbers, as the post is a structured meta-summary rather than a full reproduction of the paper's results.

Adaptive Multi-Aspect Retrieval-Augmentation for KGQA with LLMs

This post is a Chinese forum editor's structured summary of the arXiv paper:

Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation (arXiv:2412.18537, January 2025)

Authors/Affiliations: Derong Xu, Xinhang Li, Ziheng Zhang, Zhenxi Lin, Zhi Zheng, and others (10 authors total)
Source: https://arxiv.org/abs/2412.18537
Category: Question Answering

What the paper is about

The work addresses knowledge graph question answering (KGQA) by coupling large language models (LLMs) with adaptive retrieval augmentation across multiple knowledge aspects. It sits at the intersection of information retrieval, retrieval-augmented generation (RAG), and LLM-based agentic systems, motivated by the need for natural-language interaction, multi-hop reasoning, and access to up-to-date structured knowledge.

Method outline

The post describes a four-step design pattern typical of such systems:

1. Input & representation — encode queries, documents, and user context into dense/sparse representations or structured prompts. 2. Core modules — retrievers, rerankers, planners, memory modules, and tool interfaces, chained or run in parallel. 3. Learning strategies — supervised fine-tuning, contrastive learning, distillation, reinforcement learning (including process rewards), and synthetic data bootstrapping. 4. Inference strategies — single-round retrieval, iterative retrieval, parallel sub-queries, early stopping, and compute budget control.

Evaluation practice discussed

Datasets: MS MARCO, BEIR, Natural Questions, domain corpora
Metrics: nDCG@10, MRR, Recall@k, Hit@k, task success rate, latency, token cost
Baselines: BM25, dense retrieval, cross-encoder reranking, retrieval-free LLMs, commercial search APIs
Ablations: contribution of retrieval steps, reranking depth, and training data scale

> Note: the post is a meta-summary based on the abstract and public metadata. Exact quantitative results should be verified against the original PDF.

Key insights for search/recommendation practitioners

Architecture: cascaded retrieve–rerank–generate remains mainstream, but agentic paradigms make retrieval count and policy themselves learnable.
Data: high-quality instruction data and click/session logs matter; synthetic data risks leakage and distribution shift.
Evaluation: offline metrics increasingly diverge from online satisfaction; LLM-as-judge should be cross-validated with human review.
Deployment: latency, cost, interpretability, and safety are hard industrial constraints, not just benchmark numbers.

Limitations and open problems

Possible limitations noted: GPU-budget-bound experiment scale, benchmarks misaligned with real user distributions, unknown cross-lingual generalization, and safety risks of agentic systems on the open web. Future directions include better test-time compute allocation and deeper integration with knowledge graphs and structured databases.

Engineering checklist (from the post)

| Check | Question | Suggestion | |-------|----------|------------| | Data | PII in training/index? Versioning? | Partitioned indexes, anonymization, rollback-able embedding versions | | Latency | p99 budget? Retrieval steps? | Cascade + early stop, query caching, async reranking | | Quality | Does offline gain translate online? | Interleaving experiments, human audits, citation verification | | Safety | Poisoning/bias from open retrieval? | Source allowlists, adversarial detection, output filtering | | Cost | Token/GPU usage per query? | Small-model routing, distillation, hybrid sparse+dense retrieval |

Glossary

IR: Information Retrieval
RAG: Retrieval-Augmented Generation
nDCG: Normalized Discounted Cumulative Gain
Agentic Search: modeling search as sequential decisions and tool calls
Gen-IR: Generative Information Retrieval