@inproceedings{ashkenazi-etal-2025-d2cs,
title = "{D}2{CS} - Documents Graph Clustering using {LLM} supervision",
author = "Ashkenazi, Yoel and
Harari, Etzion and
Imra, Regev Yehezkel and
Abudarham, Naphtali and
Cohen, Dekel and
Louzoun, Yoram",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-emnlp.1283/",
doi = "10.18653/v1/2025.findings-emnlp.1283",
pages = "23606--23623",
ISBN = "979-8-89176-335-7",
abstract = "Knowledge discovery from large-scale, heterogeneous textual corpora presents a significant challenge. Document clustering offers a practical solution by organizing unstructured texts into coherent groups based on content and thematic similarity. However, clustering does not inherently ensure thematic consistency. Here, we propose a novel framework that constructs a similarity graph over document embeddings and applies iterative graph-based clustering algorithms to partition the corpus into initial clusters. To overcome the limitations of conventional methods in producing semantically consistent clusters, we incorporate iterative feedback from a large language model (LLM) to guide the refinement process. The LLM is used to assess cluster quality and adjust edge weights within the graph, promoting better intra-cluster cohesion and inter-cluster separation. The LLM guidance is based on a set of success Rate metrics that we developed to measure the semantic coherence of clusters. Experimental results on multiple benchmark datasets demonstrate that the iterative process and additional user-supplied a priori edges improve the summaries' consistency and fluency, highlighting the importance of known connections among the documents. The removal of very rare or very frequent sentences has a mixed effect on the quality scores.Our full code is available here: \url{https://github.com/D2CS-sub/D2CS}"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ashkenazi-etal-2025-d2cs">
<titleInfo>
<title>D2CS - Documents Graph Clustering using LLM supervision</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yoel</namePart>
<namePart type="family">Ashkenazi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Etzion</namePart>
<namePart type="family">Harari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Regev</namePart>
<namePart type="given">Yehezkel</namePart>
<namePart type="family">Imra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naphtali</namePart>
<namePart type="family">Abudarham</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dekel</namePart>
<namePart type="family">Cohen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yoram</namePart>
<namePart type="family">Louzoun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-335-7</identifier>
</relatedItem>
<abstract>Knowledge discovery from large-scale, heterogeneous textual corpora presents a significant challenge. Document clustering offers a practical solution by organizing unstructured texts into coherent groups based on content and thematic similarity. However, clustering does not inherently ensure thematic consistency. Here, we propose a novel framework that constructs a similarity graph over document embeddings and applies iterative graph-based clustering algorithms to partition the corpus into initial clusters. To overcome the limitations of conventional methods in producing semantically consistent clusters, we incorporate iterative feedback from a large language model (LLM) to guide the refinement process. The LLM is used to assess cluster quality and adjust edge weights within the graph, promoting better intra-cluster cohesion and inter-cluster separation. The LLM guidance is based on a set of success Rate metrics that we developed to measure the semantic coherence of clusters. Experimental results on multiple benchmark datasets demonstrate that the iterative process and additional user-supplied a priori edges improve the summaries’ consistency and fluency, highlighting the importance of known connections among the documents. The removal of very rare or very frequent sentences has a mixed effect on the quality scores.Our full code is available here: https://github.com/D2CS-sub/D2CS</abstract>
<identifier type="citekey">ashkenazi-etal-2025-d2cs</identifier>
<identifier type="doi">10.18653/v1/2025.findings-emnlp.1283</identifier>
<location>
<url>https://aclanthology.org/2025.findings-emnlp.1283/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>23606</start>
<end>23623</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T D2CS - Documents Graph Clustering using LLM supervision
%A Ashkenazi, Yoel
%A Harari, Etzion
%A Imra, Regev Yehezkel
%A Abudarham, Naphtali
%A Cohen, Dekel
%A Louzoun, Yoram
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F ashkenazi-etal-2025-d2cs
%X Knowledge discovery from large-scale, heterogeneous textual corpora presents a significant challenge. Document clustering offers a practical solution by organizing unstructured texts into coherent groups based on content and thematic similarity. However, clustering does not inherently ensure thematic consistency. Here, we propose a novel framework that constructs a similarity graph over document embeddings and applies iterative graph-based clustering algorithms to partition the corpus into initial clusters. To overcome the limitations of conventional methods in producing semantically consistent clusters, we incorporate iterative feedback from a large language model (LLM) to guide the refinement process. The LLM is used to assess cluster quality and adjust edge weights within the graph, promoting better intra-cluster cohesion and inter-cluster separation. The LLM guidance is based on a set of success Rate metrics that we developed to measure the semantic coherence of clusters. Experimental results on multiple benchmark datasets demonstrate that the iterative process and additional user-supplied a priori edges improve the summaries’ consistency and fluency, highlighting the importance of known connections among the documents. The removal of very rare or very frequent sentences has a mixed effect on the quality scores.Our full code is available here: https://github.com/D2CS-sub/D2CS
%R 10.18653/v1/2025.findings-emnlp.1283
%U https://aclanthology.org/2025.findings-emnlp.1283/
%U https://doi.org/10.18653/v1/2025.findings-emnlp.1283
%P 23606-23623
Markdown (Informal)
[D2CS - Documents Graph Clustering using LLM supervision](https://aclanthology.org/2025.findings-emnlp.1283/) (Ashkenazi et al., Findings 2025)
ACL
- Yoel Ashkenazi, Etzion Harari, Regev Yehezkel Imra, Naphtali Abudarham, Dekel Cohen, and Yoram Louzoun. 2025. D2CS - Documents Graph Clustering using LLM supervision. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 23606–23623, Suzhou, China. Association for Computational Linguistics.