---
title: 'KDR-Agent: Multi-Agent NER Framework'
url: https://www.emergentmind.com/topics/kdr-agent
type: topic
---

# KDR-Agent: Multi-Agent NER Framework

Searching arXiv for the specified paper and closely related NER/ICL context.
KDR-Agent is a multi-agent framework for multi-domain low-resource named entity recognition (NER) via in-context learning (ICL) with large language models (LLMs). It integrates **Knowledge retrieval**, **Disambiguation**, and **Reflective analysis** under a central planner, with the stated goal of addressing three key challenges in low-resource ICL-based NER: reliance on large annotated support sets, insufficient domain knowledge of the LLM, and entity ambiguity [2511.19083]. The framework operates in two stages—**Knowledge In-context Construction** and **Reflection & Correction**—and combines natural-language type definitions, a static set of entity-level contrastive demonstrations, Wikipedia-derived factual knowledge, contextual disambiguation, and structured self-assessment to improve NER across multiple domains [2511.19083].

## 1. Problem Setting and Design Objective

KDR-Agent is designed for NER in **low-resource, multi-domain settings** using ICL with LLMs [2511.19083]. The paper identifies three explicit issues motivating the framework: **reliance on large annotated support sets (Issue 1)**, **insufficient domain knowledge of the LLM (Issue 2)**, and **entity ambiguity (Issue 3)** [2511.19083]. In this formulation, the framework is intended not merely to predict entity spans and labels from a sentence, but to augment the inference process with external factual evidence and post hoc diagnostic correction.

The central methodological premise is that low-resource NER can benefit from decomposing inference into specialized subtasks handled by distinct agents. This suggests a modular alternative to single-pass prompting: instead of relying exclusively on retrieved labeled exemplars or the latent knowledge of the backbone LLM, KDR-Agent injects external knowledge, explicit disambiguation rationale, and reflective error analysis into the prompt construction and revision process [2511.19083].

## 2. Multi-Agent Architecture and Coordination

KDR-Agent divides inference into two stages: **A. Knowledge In-context Construction** and **B. Reflection & Correction** [2511.19083]. The architecture comprises four coordinated roles: a **Central LLM Planner**, a **Knowledge Retrieval Agent**, a **Disambiguation Agent**, and a **Reflective Analysis Agent** [2511.19083].

The **Central LLM Planner** scans the input sentence $x$ to identify both concepts requiring external facts and ambiguous mentions. It emits a set of Wikipedia search queries,
$$
Q = \{\,q_i = \mathrm{GenerateQuery}(m_i, x)\,\},
$$
and a prompt segment $\mathcal{P}_{\mathrm{ambig}}$ listing mentions to disambiguate [2511.19083]. The **Knowledge Retrieval Agent** issues Wikipedia lookups for each query,
$$
k_i \;=\;\mathrm{RetrieveFromWiki}(q_i),
$$
and returns lead-paragraph snippets $\{k_i\}$ formatted as $\mathcal{P}_{\mathrm{know}}$ [2511.19083]. The **Disambiguation Agent** consumes the input $x$ together with $\mathcal{P}_{\mathrm{ambig}}$ and produces natural-language reasoning statements,
$$
\mathcal{P}_{\mathrm{disamb}}
= \mathrm{Disambiguate}(x,\;\mathcal{P}_{\mathrm{ambig}}),
$$
which are intended to guide the prediction toward the correct sense or entity type [2511.19083].

After the first-pass prediction $\hat y^{(0)}$, the **Reflective Analysis Agent** performs structured error analysis under a reflection guideline $\mathcal{P}_{\mathrm{Reflection}}$,
$$
\mathcal{R}
= \mathrm{Reflect}(x,\;\hat y^{(0)},\;\mathcal{P}_{\mathrm{Reflection}}),
$$
and generates a diagnostic report used in a second correction pass [2511.19083]. The paper specifies the coordination flow as follows:  
1. Central planner generates $Q,\;\mathcal{P}_{\mathrm{ambig}}$  
2. Retrieval agent builds $\mathcal{P}_{\mathrm{know}}$  
3. Disambiguation agent builds $\mathcal{P}_{\mathrm{disamb}}$  
4. Prompt assembly in Stage 1 yields initial inference $\hat y^{(0)}$  
5. Reflective agent produces $\mathcal{R}$  
6. Final correction in Stage 2 yields $\hat y^{(1)}$ [2511.19083]

A plausible implication is that the framework separates epistemic augmentation from corrective reasoning: factual completion and ambiguity resolution occur before initial decoding, whereas structured error repair is deferred to a subsequent pass.

## 3. Knowledge Retrieval and Disambiguation

The knowledge retrieval component is explicitly based on Wikipedia. For each flagged concept $m_i$, the planner formulates
$$
q_i = \mathrm{GenerateQuery}(m_i,x),
$$
and the retrieval call returns
$$
k_i = \mathrm{RetrieveFromWiki}(q_i),
$$
where $k_i$ is the lead paragraph of the top-ranked Wikipedia entry [2511.19083]. The retrieved set $\{k_i\}$ is concatenated and formatted into the prompt segment
$$
\mathcal{P}_{\mathrm{know}} = \mathrm{Format}(\{k_i\}).
$$
The paper states that **no additional similarity ranking or embedding-based retrieval is used**; the system relies on the **MediaWiki Action API**, restricted to **pre-May 1 2025 content** [2511.19083].

This design is narrower than retrieval pipelines that employ dense retrieval or re-ranking. Here, the retrieval component is deliberately lightweight and factual rather than learned. For low-resource NER, the main function of the knowledge snippets is to compensate for insufficient domain knowledge in the backbone LLM, especially when domain-specific mentions are poorly represented in the model’s internal knowledge [2511.19083].

The disambiguation process is also formulated in natural language rather than through a learned scoring function. Given the input $x$ and the planner’s $\mathcal{P}_{\mathrm{ambig}}$, the Disambiguation Agent performs a “self-check” dialogue for each mention $m_i$ to explain its sense in context:
$$
\mathcal{P}_{\mathrm{disamb}}
= \mathrm{Disambiguate}(x,\;\mathcal{P}_{\mathrm{ambig}}).
$$
The output is a set of natural-language interpretation statements, and the paper explicitly notes that it does **not** introduce an explicit scoring or ranking function for this stage [2511.19083]. This suggests that KDR-Agent treats ambiguity resolution as a prompt-mediated reasoning problem rather than a separate classification submodule.

## 4. Reflective Analysis and Correction

The reflection mechanism is designed to detect and correct four common error types: **span errors**, **type errors**, **spurious detections**, and **omissions** [2511.19083]. After the initial prediction $\hat y^{(0)}$, the Reflective Analysis Agent generates a structured report
$$
\mathcal{R} = \mathrm{Reflect}(x,\;\hat y^{(0)},\;\mathcal{P}_{\mathrm{Reflection}}),
$$
which is then serialized into $\mathcal{P}_{\mathrm{reflect}}$, and a final correction prompt $\mathcal{P}_{\mathrm{Correction}}$ is constructed [2511.19083].

The paper provides a pseudocode sketch in which the agent iterates over predicted mentions in $y_0$, records boundary mismatches and type mismatches with justifications, then identifies missing gold mentions and spurious predicted mentions, aggregates the findings into $R$, and outputs the structured report [2511.19083]. Although the pseudocode references gold mentions and gold boundaries, the operational description in the framework is that the agent performs structured self-assessment by comparing $\hat y^{(0)}$ with the input sentence $x$ under the reflection guideline [2511.19083].

Within the framework, reflection is not merely explanatory. It is part of a two-pass inference protocol in which the first-pass output is treated as revisable. The ablation results indicate that this component has the largest single impact among the major modules, which supports the interpretation that explicit post hoc diagnosis is central to KDR-Agent’s effectiveness [2511.19083].

## 5. Prompt Construction: Type Definitions and Contrastive Demonstrations

KDR-Agent supplements retrieved knowledge and disambiguation rationale with two additional prompt components: **natural-language type definitions** and **entity-level contrastive demonstrations** [2511.19083].

For a target type set $\mathcal{T}=\{t_1,\dots,t_m\}$, each type $t_i$ is associated with a textual description $\mathcal{D}(t_i)$ specifying scope and inclusions or exclusions. These are concatenated into
$$
\mathcal{P}_{\mathrm{type}}
= \mathrm{Concat}\bigl(\mathcal{D}(t_1),\dots,\mathcal{D}(t_m)\bigr)
$$
[2511.19083]. This provides explicit label semantics in the prompt, reducing dependence on latent label understanding.

The framework also uses a **static support set** $\mathcal{E}=\{(x_j,y_j)\}_{j=1}^k$, where each $y_j$ contains both positive and negative mention–type pairs:
$$
y_j \;=\;\{(m_p,t_p)^+\}\;\cup\;\{(m_n,t_n)^-\}.
$$
The negative examples simulate four error types: **boundary shifts**, **wrong labels**, **spurious mentions**, and **omissions** [2511.19083]. These demonstrations are formatted into
$$
\mathcal{P}_{\mathrm{demo}} = \mathrm{Format}(\mathcal{E}).
$$
The paper states that, although no explicit loss function is trained, these contrastive pairs encourage the LLM to distinguish correct from incorrect spans and types in context [2511.19083].

A notable feature of this design is the use of **static** rather than dynamically retrieved demonstrations. This directly addresses the paper’s first stated issue—reliance on large annotated support sets—and shifts the burden from retrieval over many labeled exemplars to richer prompt structure and auxiliary reasoning [2511.19083].

## 6. Experimental Setup and Empirical Results

The experimental evaluation covers **10 datasets** spanning **5 domains**: **Biomedical** (BC5CDR, NCBI), **Task-oriented Dialogue** (MIT Movie, MIT Restaurant), **News** (CoNLL-2003, OntoNotes 5.0), **Social Media** (Twitter Broad, Twitter NER-7), and **Open-domain** (WikiANN English subset, sampled 5 K; WNUT-17) [2511.19083]. In the low-resource setting, the shared static support set uses **$k=10$** for datasets with many types—MIT Movie, MIT Restaurant, and OntoNotes 5.0—and **$k=5$** for the remaining seven datasets [2511.19083]. The LLM backbones evaluated are **GPT-4o**, **Qwen-2.5-72B**, and **DeepSeek-V3** [2511.19083].

Selected GPT-4o results reported in the paper are summarized below.

| Model | BC5CDR | NCBI | MIT Movie | MIT Rest. | CoNLL | OntoNotes |
|---|---:|---:|---:|---:|---:|---:|
| Zero-Shot (CMAS) | 73.21 | 69.91 | 67.61 | 51.64 | 78.31 | 60.79 |
| Few-Shot (CodeIE) | 77.61 | 71.97 | 70.67 | 56.93 | 83.01 | 65.67 |
| KDR-Agent | 82.47 | 79.41 | 76.16 | 69.98 | 83.34 | 71.85 |

The paper reports **absolute gains of 4–8 F1 pts over the strongest few-shot baselines (CodeIE / GPT-NER)** and states that improvements are **consistent across all ten datasets and all three backbones** [2511.19083]. Examples given include **Qwen-2.5-72B: BC5CDR 81.45 vs 76.10 (CodeIE)** and **DeepSeek-V3: MIT Restaurant 65.99 vs 58.11** [2511.19083]. It also notes that **formal p-values are not reported**, although the improvements are described as large and consistent [2511.19083].

These results position KDR-Agent as a strong low-resource ICL baseline for cross-domain NER. The empirical pattern suggests that the combination of retrieved factual context, explicit ambiguity handling, and reflective correction is beneficial even when the backbone model is already a strong general-purpose LLM.

## 7. Ablations, Component Contributions, and Scope

The ablation study, reported on **NCBI**, **OntoNotes 5.0**, and **Twitter NER-7** with the GPT-4o backbone, evaluates the contributions of the major components [2511.19083].

| Ablation | NCBI | OntoNotes | Twitter-7 |
|---|---:|---:|---:|
| Full KDR-Agent | 79.41 | 71.85 | 60.87 |
| – Reflection Stage | 75.91 | 70.17 | 57.81 |
| – Knowledge Retrieval (KRA) | 76.21 | 71.70 | 59.34 |
| – Disambiguation Agent (DA) | 75.49 | 70.73 | 55.81 |
| – KRA + DA removed | 74.16 | 69.94 | 55.07 |
| – No Contrastive Negatives | 78.36 | 70.69 | 58.99 |

The paper’s stated insights are that **Reflective Analysis has the largest single impact (−3–4 F1 when removed)**, that **Biomedical and Social Media suffer most when KRA or DA are ablated**, and that **contrastive negative examples yield a modest but consistent +1 F1 uplift** [2511.19083]. These findings indicate that the framework’s gains are not attributable to a single prompt engineering trick; rather, they arise from the interaction of multiple targeted components.

The paper does **not explicitly list limitations or future directions** [2511.19083]. It does, however, include reader-noted possible extensions: **broader external knowledge sources beyond Wikipedia**, **more fine-grained disambiguation criteria or learned scoring functions**, and **automatic expansion of contrastive demonstration pools without manual annotation** [2511.19083]. Because these are explicitly marked as not being paper-authored limitations, they should be understood as plausible implications rather than claims of the work itself.

A potential misconception is that KDR-Agent is a generic “knowledgeable deep research” system. In fact, despite the acronym overlap with later work on **Knowledgeable Deep Research**, KDR-Agent in this context denotes a specialized **NER** framework whose components are tailored to low-resource, multi-domain in-context entity extraction rather than to long-form report generation [2511.19083].

Source: https://www.emergentmind.com/topics/kdr-agent