<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="grover-etal-2026-one">
    <titleInfo>
        <title>Is One Token All It Takes? Graph Pooling Tokens for LLM-based GraphQA</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Ankit</namePart>
        <namePart type="family">Grover</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Lodovico</namePart>
        <namePart type="family">Giaretta</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Remi</namePart>
        <namePart type="family">Bourgerie</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sarunas</namePart>
        <namePart type="family">Girdzijauskas</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-05</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the Knowledge Graphs and Large Language Models Workshop (KG-LLM) @ LREC26</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Gilles</namePart>
            <namePart type="family">Sérasset</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Katerina</namePart>
            <namePart type="family">Gkirtzou</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Michael</namePart>
            <namePart type="family">Cochez</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jan-Christoph</namePart>
            <namePart type="family">Kalo</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>ELRA Language Resources Association (ELRA)</publisher>
            <place>
                <placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>The integration of Graph Neural Networks (GNNs) with Large Language Models (LLMs) has emerged as a promising paradigm for Graph Question Answering (GraphQA). However, effective methods for encoding complex structural information into the LLM’s latent space remain an open challenge. Current state-of-the-art architectures, such as G-Retriever, typically rely on standard GNNs and aggressive mean pooling to compress entire graph substructures into a single token, creating a severe information bottleneck. This work mitigates this bottleneck by investigating two orthogonal strategies: (1) increasing the bandwidth of the graph-to-LLM interface via multi-token pooling, and (2) enhancing the semantic quality of the graph encoder via global attention mechanisms. We evaluate a suite of hierarchical pruning and clustering-based pooling operators—including Top-k, SAGPool, DiffPool, MinCutPool, and Virtual Node Pooling (VNPool) to project graph data into multiple learnable tokens. Empirically, we demonstrate that while pooling introduces significant instability during soft prompt tuning, the application of Low-Rank Adaptation (LoRA) effectively stabilizes these projections, allowing compressed representations to rival full-graph baselines (achieving \ensuremath\sim73% Hit@1 on WebQSP). Conceptually, we demonstrate that a Graph Transformer with VNPool implementation functions structurally as a single-layer Perceiver IO encoder. Finally, we adapt the FandE (Features and Edges) Score to the generative GraphQA domain. Our analysis reveals that current the GraphQA benchmark suffer from representational saturation, where the target answers are often highly correlated with isolated node features. The implementation of our experiments is available at https://anonymous.4open.science/r/Pool-A85D/README.md.</abstract>
    <identifier type="citekey">grover-etal-2026-one</identifier>
    <identifier type="doi">10.63317/5fjja5d5fw33</identifier>
    <location>
        <url>https://aclanthology.org/2026.kallm-1.4/</url>
    </location>
    <part>
        <date>2026-05</date>
        <extent unit="page">
            <start>36</start>
            <end>44</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Is One Token All It Takes? Graph Pooling Tokens for LLM-based GraphQA
%A Grover, Ankit
%A Giaretta, Lodovico
%A Bourgerie, Remi
%A Girdzijauskas, Sarunas
%Y Sérasset, Gilles
%Y Gkirtzou, Katerina
%Y Cochez, Michael
%Y Kalo, Jan-Christoph
%S Proceedings of the Knowledge Graphs and Large Language Models Workshop (KG-LLM) @ LREC26
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F grover-etal-2026-one
%X The integration of Graph Neural Networks (GNNs) with Large Language Models (LLMs) has emerged as a promising paradigm for Graph Question Answering (GraphQA). However, effective methods for encoding complex structural information into the LLM’s latent space remain an open challenge. Current state-of-the-art architectures, such as G-Retriever, typically rely on standard GNNs and aggressive mean pooling to compress entire graph substructures into a single token, creating a severe information bottleneck. This work mitigates this bottleneck by investigating two orthogonal strategies: (1) increasing the bandwidth of the graph-to-LLM interface via multi-token pooling, and (2) enhancing the semantic quality of the graph encoder via global attention mechanisms. We evaluate a suite of hierarchical pruning and clustering-based pooling operators—including Top-k, SAGPool, DiffPool, MinCutPool, and Virtual Node Pooling (VNPool) to project graph data into multiple learnable tokens. Empirically, we demonstrate that while pooling introduces significant instability during soft prompt tuning, the application of Low-Rank Adaptation (LoRA) effectively stabilizes these projections, allowing compressed representations to rival full-graph baselines (achieving \ensuremath\sim73% Hit@1 on WebQSP). Conceptually, we demonstrate that a Graph Transformer with VNPool implementation functions structurally as a single-layer Perceiver IO encoder. Finally, we adapt the FandE (Features and Edges) Score to the generative GraphQA domain. Our analysis reveals that current the GraphQA benchmark suffer from representational saturation, where the target answers are often highly correlated with isolated node features. The implementation of our experiments is available at https://anonymous.4open.science/r/Pool-A85D/README.md.
%R 10.63317/5fjja5d5fw33
%U https://aclanthology.org/2026.kallm-1.4/
%U https://doi.org/10.63317/5fjja5d5fw33
%P 36-44

Markdown (Informal)

[Is One Token All It Takes? Graph Pooling Tokens for LLM-based GraphQA](https://aclanthology.org/2026.kallm-1.4/) (Grover et al., KaLLM 2026)

• Is One Token All It Takes? Graph Pooling Tokens for LLM-based GraphQA (Grover et al., KaLLM 2026)

ACL

• Ankit Grover, Lodovico Giaretta, Remi Bourgerie, and Sarunas Girdzijauskas. 2026. Is One Token All It Takes? Graph Pooling Tokens for LLM-based GraphQA. In Proceedings of the Knowledge Graphs and Large Language Models Workshop (KG-LLM) @ LREC26, pages 36–44, Palma, Mallorca (Spain). ELRA Language Resources Association (ELRA).