EA-Agent: A Structured Multi-Step Reasoning Agent for Entity Alignment¶

Conference: ACL 2026
arXiv: 2604.11686
Code: GitHub
Area: LLM Agent
Keywords: Entity Alignment, Knowledge Graph, Multi-step Reasoning, Tool Planning, Reward-guided Optimization

TL;DR¶

EA-Agent is proposed to decompose Entity Alignment (EA) into a structured multi-step reasoning process. By planning and executing a tool pool (triplet selector + alignment tool + reflector), it achieves interpretable alignment decisions. Combined with reward-guided offline policy optimization for continuous improvement of planning capabilities, it improves Hits@1 on DBP15K by up to 3.17% while mitigating efficiency issues caused by redundant triplets.

Background & Motivation¶

Background: Entity alignment is a fundamental technology for knowledge fusion, aiming to identify nodes in different knowledge graphs that refer to the same entity. Traditional methods based on knowledge representation learning (e.g., TransE, GCN-Align) show limited performance in noisy or sparse supervised scenarios. Recent LLM-based methods (e.g., ChatEA, LLMEA) utilize semantic understanding to improve performance.

Limitations of Prior Work: (1) Existing LLM-based EA methods treat LLMs as black-box decision-makers, lacking interpretability—making it difficult to determine which information is critical for alignment decisions; (2) Directly inputting large numbers of attribute and relation triplets results in excessively long prompts and high inference costs; (3) Many triplets are redundant or even noisy, which can interfere with judgment.

Key Challenge: The need to utilize the powerful semantic understanding of LLMs while addressing the issues of black-box non-interpretability and the efficiency of large-scale triplet processing.

Goal: Design a reasoning-driven Agent framework that achieves interpretable, controllable, and efficient entity alignment through multi-step tool planning and execution.

Key Insight: Treat EA as a multi-step decision-making problem—selecting the most informative triplets first, then making alignment decisions, and finally performing reflection for verification when uncertainty exists.

Core Idea: Tool pool (Attribute/Relation Triplet Selector + Alignment Tool + Reflector) + Path Planning + Reward-guided Offline Policy Optimization.

Method¶

Overall Architecture¶

Three stages: (1) Path Planning: The Agent autonomously plans tool invocation paths based on the structural features of the source entity and candidate similarity scores; (2) Tool Invocation: Triplet selection, alignment decision, and reflection verification are executed in the planned order; (3) Agent Optimization: A reward function evaluates path quality → offline policy update → iterative improvement.

Key Designs¶

Attribute/Relation Triplet Selector:
- Function: Filters redundant triplets before LLM reasoning to retain the most discriminative information.
- Mechanism: The attribute selector uses an entropy-based criterion \(H(a) = -\sum p(v)\log p(v)\)—a more uniform distribution of attribute values among candidate entities (lower entropy) indicates stronger discriminative power. The relation selector uses inverse frequency weighting \(I(r) = \log(N/(\text{freq}(r)+1))\)—rare relations are more discriminative. Predefined important attributes are also retained.
- Design Motivation: Inputting all triplets without selection wastes tokens and introduces noise. The selector acts as an information bottleneck, preserving only critical signals.
Reward-guided Path Optimization:
- Function: Continuously improves the Agent's tool planning strategy.
- Mechanism: The reward function \(\gamma = \gamma_\mu + c \cdot \gamma_{\text{ref}} + \gamma_e\) comprises three components: (1) Alignment correctness \(\gamma_\mu\) (core); (2) Reflection rationality \(\gamma_{\text{ref}}\) (rewarding successful corrections, penalizing incorrect modifications, and slightly penalizing redundant reflections); (3) Path efficiency \(\gamma_e = e^{-\beta \cdot l}\) (penalizing excessively long paths). Strategies are optimized by rewriting paths via offline SFT under reward guidance.
- Design Motivation: Single-turn planning may generate redundant or inefficient paths. The closed-loop "planning → execution → evaluation → update" iteration ensures continuous strategy improvement.
Reflector (Conditional):
- Function: Verifies and corrects alignment results for uncertain cases.
- Mechanism: An LLM-based module activated only when candidate similarity scores indicate ambiguity. It re-evaluates candidates based on previous context and provides revised predictions.
- Design Motivation: Not all alignments requires reflection—direct decisions are more efficient for simple cases, and extra verification is only triggered when uncertainty exists.

Key Experimental Results¶

Main Results (DBP15K)¶

Method	FR-EN Hits@1	JA-EN Hits@1	ZH-EN Hits@1
GCN-Align	~40	~40	~40
TEA	~90	~90	~85
ChatEA	~92	~91	~88
Ours (EA-Agent)	~95	~94	~91

Ablation Study¶

Configuration	Description
w/o Triplet Selection	Token consumption increases significantly; performance drops slightly
w/o Reflector	Error rate for uncertain cases increases
w/o Path Optimization	Planning strategy becomes unstable; redundant tool calls increase
Full EA-Agent	Optimal performance + maximum efficiency

Key Findings¶

EA-Agent achieves SOTA on all datasets, with Hits@1 improvements up to 3.17% and consistent MRR gains.
The triplet selector significantly reduces token consumption while maintaining or even improving performance—proving that many triplets are indeed redundant.
Path optimization significantly enhances planning quality: Both path efficiency and alignment accuracy improve steadily after 3 iterations.
Conditional activation of the reflector is the optimal strategy: Always enabling it is less effective than demand-based activation.
Interpretability: Every alignment decision can be traced back to a specific tool invocation path and key triplets.

Highlights & Insights¶

Modeling EA as a multi-step tool planning problem opens up the application space for the Agent paradigm in knowledge graph tasks.
The three-component reward function design is highly practical: it balances correctness, reflection rationality, and efficiency, avoiding optimization bias toward a single goal.
The triplet selector utilizing information theory criteria (entropy and inverse frequency) is a simple yet effective solution that can be directly migrated to other KG tasks.

Limitations & Future Work¶

Dependent on TEA for generating the initial candidate list; candidate quality limits the upper bound.
Path optimization requires multiple iterations, leading to high training costs.
Validated only on cross-lingual EA; same-language or cross-domain EA remains to be explored.
The tool pool is manually designed; can new tools be discovered automatically?
Reflector judgments may introduce new hallucinations.

vs ChatEA: ChatEA uses code to format KG structures but remains a black-box decision-maker. EA-Agent achieves interpretable decisions through tool planning.
vs LLMEA: LLMEA inputs all triplets directly; EA-Agent is more efficient by selecting before aligning.
vs General Agent Frameworks: EA-Agent specializes the Agent paradigm for KG tasks; tool design and reward functions are task-specific.

Rating¶

Novelty: ⭐⭐⭐⭐ Introducing the Agent paradigm to EA is new, though individual components (tool planning, LoRA fine-tuning) are established.
Experimental Thoroughness: ⭐⭐⭐⭐⭐ 3 datasets + 10 baselines + ablation + efficiency analysis + interpretability cases.
Writing Quality: ⭐⭐⭐⭐ RQ-driven with clear formalization.
Value: ⭐⭐⭐⭐ Provides methodological inspiration for LLM applications in the KG field.