pith. sign in

arxiv: 2606.27347 · v1 · pith:WTOSPNCKnew · submitted 2026-06-25 · 💻 cs.CL

Mapping Political-Elite Networks in Europe with a Multilingual Joint Entity-Relation Extraction Pipeline

Kirill Solovev , Jana Lasser This is my paper

Pith reviewed 2026-06-26 03:42 UTC · model grok-4.3

classification 💻 cs.CL
keywords political elite networksentity-relation extractionknowledge graphsmultilingual NLPsigned relationsWikidata linkingcomputational social sciencenews corpora
0
0 comments X

The pith

A fully open multilingual pipeline extracts directed signed relations from news text to build temporal political-elite knowledge graphs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a modular pipeline that performs joint named-entity recognition, Wikidata linking, and ontology-guided relation extraction on large multilingual news corpora. The extracted relations are signed and temporal, allowing construction of knowledge graphs that track political coalitions, conflicts, and personnel movements at scale. Validation against a 3491-relation gold standard yields 68.2 percent strict and 93.7 percent lenient textual correctness, while case studies in Austria and Poland recover documented party fractures, court outcomes, and patronage networks. The approach replaces manual coding with replicable automated extraction for comparative politics questions about rent-seeking versus civic networks.

Core claim

We present a modular, fully open-weight pipeline for multilingual joint entity-relation extraction that builds signed, temporal knowledge graphs from massive unstructured news corpora. It combines span-based NER with a three-stage linking cascade to language-independent Wikidata identifiers and a high-throughput ontology-constrained mixture-of-experts model with guided decoding. A full-coverage spot-check against a 3491-relation gold standard shows high textual correctness. Two large-scale case studies validate the pipeline against the public record: in Austria it reconstructs a party's complete lifecycle including fractures and convictions; in Poland it maps state-enterprise patronage and t

What carries the argument

Ontology-constrained mixture-of-experts model with guided decoding that extracts directed signed relationships grounded in a domain ontology after Wikidata entity linking.

If this is right

  • The pipeline reconstructs a political party's full lifecycle, dating fractures and tracking personnel into successor factions and convictions.
  • It uncovers overlapping economic-governance networks and the structurally balanced signed conflict network between major parties in Poland.
  • The method supplies a replicable foundation for cross-national studies of elite coalitions without intensive manual coding.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same pipeline could be run on corpora from additional European countries to compare network structures across different electoral systems.
  • Signed relation graphs produced this way could be tested for structural balance properties that distinguish adversarial from cooperative elite clusters.
  • Extending the temporal window of the input corpora would allow tracking how elite networks evolve after elections or scandals.

Load-bearing premise

The signed relations extracted from news text accurately reflect real-world political ties rather than journalistic framing or selection bias.

What would settle it

Independent historical records for the Austrian party case study that show different dates for internal fractures or personnel movements than those recovered by the pipeline.

Figures

Figures reproduced from arXiv: 2606.27347 by Jana Lasser, Kirill Solovev.

Figure 1
Figure 1. Figure 1: The pipeline’s data flow. Articles pass through chunking (Section 3.1), NER (Section 3.4), entity linking (Section 3.5), and LLM-based relation extraction (Section 3.6). Offline Wikidata preprocessing (Section 3.3) produces the linking indices; the SKOS ontology (Section 3.2) constrains the output via guided decoding. lary is fixed and enforced by guided decoding, entity identity is resolved through linkin… view at source ↗
Figure 2
Figure 2. Figure 2: DSPy signature. Inputs on the left, LLM tasks in the middle, and output on the right. tion 6.6). Additionally, it canonicalises names (expanding abbreviations, resolving inflections), selects QID candidates by integer index, and records which NER mention motivated each choice. To ensure the well-formed structure and, we implement three mechanisms: Guided decoding. All type fields are Literal enums; vLLM ma… view at source ↗
Figure 3
Figure 3. Figure 3: Lifecycle of three Austrian populist projects: BZÖ, Team Stronach, and Die Reformkonservativen (REKOS); vertical dashed lines mark Nationalrat elections. election (10.7%) Haider dies 0 10 20 30 08.2008 09.2008 10.2008 11.2008 12.2008 01.2009 02.2009 Articles (co-mention) [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Haider’s death at daily resolution. Death around 01:00 Saturday 11 October 2008; the first business-day cycle peaks at 𝑛 = 36 co-mentions on Monday 13 October. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The Hypo→FPK fracture, December 2009. Daily mentions of BZÖ, FPK, the Hypo bank, and the BZÖ ∩ Hypo intersection. The two dashed lines mark the emergency nationalisation (12-14) and the FPK breakaway announcement (12-16), two days apart; the BZÖ federal reaction peaks at 𝑛 = 32 on 12-17, in agreement with the documented chronology. 5.4 Personnel Diaspora The personnel analysis is of particular interest; de… view at source ↗
Figure 6
Figure 6. Figure 6: Diaspora of the 29-person BZÖ cohort. Each row is one person; each yearly cell is split: the top half is curated party membership (Wikipedia and Wikidata P102), the bottom half the pipeline’s dominant co-mention party in the Austrian press, so a two-tone cell marks where the press frame and formal membership diverge. Vertical dashed lines mark Haider’s death (2008), the FPK split (2009), and the 2013 wipeo… view at source ↗
Figure 7
Figure 7. Figure 7: The Carinthia/Hypo bloc, a resolution-stable community on the BZÖ ego-network, exported from Neo4j. Nodes are the Carinthian BZÖ/FPK core, Jörg Haider, and the Hypo Alpe-Adria bank (cyan; the historical entity with its post-2014 Addiko rebrand merged in, as press mentions split across both). The pink nodes, Gerhard Dörfler, Uwe Scheuch, Harald Dobernig, were convicted in the Austrian courts; the blue nodes… view at source ↗
Figure 8
Figure 8. Figure 8: Haider’s co-mention concentration with the Hypo bank over time. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Scandal-edge density of the BZÖ cohort against a 1:1 press-matched, Wikidata-verified control. On hard legal-jeopardy edges the cohort runs at 3.87× the control (paired Wilcoxon 𝑝 = 0.006). 5.6 Validation and Limits Set against Wikidata’s P102 (party-membership) statements with qualifier dates, the pipeline’s year-by-year trajectory makes zero false assignments on the date-qualified cells (precision 1.000)… view at source ↗
Figure 10
Figure 10. Figure 10: Degree distribution of the Polish entity graph (log-binned). 59.9% of entities have degree 1 and 136 exceed degree 1 000. 32.5% 24.0% 14.7% 9.1% 6.8% 5.9% 5.1% Personal 1.8% Stance Enforcement Temporal Descriptive Geographic Governance Economic 0% 10% 20% 30% Share of content relations Other groups Co-dominant pair [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of relations across the eight ontology groups. Economic (32.5%) and Governance (24.0%) co-dominate. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Directional criticizes asymmetry between PiS and PO. (a) Party-to-party criticism (any endpoint). (b) Top critics (largest out-degree). (c) top targets (largest in-degree). Bars are coloured by formal party membership: orange for PiS, blue for PO, grey otherwise. Andrzej Duda (independent) and Zbigniew Ziobro (Suwerenna Polska) appear grey, not formal PiS members. 6.4 Firm-Level O-Group The structural dat… view at source ↗
Figure 13
Figure 13. Figure 13: Governance∩Economic multiplex overlap: the firm-level O-group signature. (a) Share of each layer’s entities also active in the other layer. (b) Multiplex depth: entity pairs sharing ≥ 2, ≥ 3, and ≥ 5 relationship groups. 6.5 SOE Patronage Turnover The extracted SOE–politician ties represent genuine role relations rather than mere co-mentions. Edge dates accurately track documented tenure (Daniel Obajtek’s… view at source ↗
Figure 14
Figure 14. Figure 14: Annual leadership-role transitions at major SOEs (raw count). 6.6 Signed PO–PiS Cleavage The partisan duopoly is structured by a signed cleavage, invisible to standard unsigned commu￾nity detection algorithms. Operating on the political subgraph (QID-linked persons and parties on Governance ∪ Stance edges, 6 043 nodes), unsigned detection fails to separate the opposing camps at any resolution. Leiden modu… view at source ↗
Figure 15
Figure 15. Figure 15: The signed PO–PiS cleavage among nine front-line leaders, exported from Neo4j. Node colour is camp (P102-grounded: PO vs PiS). An arrow is drawn between two leaders only where at least two relations of one sign occur between them. Cross￾camp arrows are almost entirely negative; positive arrows fall within camps, alongside intra-camp rivalry, the structural-balance signature. Applied to this signed space, … view at source ↗
Figure 16
Figure 16. Figure 16: Coverage-controlled cross-camp antagonism and duopoly share over time. Antagonism rises 2.6–3.9× to a 2023–25 peak while the PO/PiS share of party-level Stance activity falls to 36.7% and challengers rise to 15.5%. embeddedness rather than replaceability. There are two systemic caveats with this analysis. First, the member_of_political_party edge suffers from contextual contamination, occasionally returni… view at source ↗
Figure 17
Figure 17. Figure 17: Weighted PageRank of the top entities (a) and top persons (b). 26 [PITH_FULL_IMAGE:figures/full_fig_p026_17.png] view at source ↗
read the original abstract

Whether political elites organise into rent-seeking coalitions that capture public resources or civic networks that sustain governance is a central question in comparative politics. Yet observing these complex, informal, and adversarial ties at scale has historically required intensive manual coding, while automated text-as-data methods have largely been limited to simple co-occurrence. Recent large language model (LLM) approaches offer a path forward but often rely on proprietary APIs, lack cross-lingual capability, and struggle with scalable entity resolution. We present a modular, fully open-weight pipeline for multilingual joint entity-relation extraction that builds signed, temporal knowledge graphs from massive unstructured news corpora. It combines span-based named-entity recognition (NER) with a three-stage linking cascade mapping mentions to language-independent Wikidata identifiers; a high-throughput, ontology-constrained mixture-of-experts model then uses guided decoding to extract directed, signed relationships grounded in a domain ontology. A full-coverage spot-check against a 3491-relation gold standard shows high textual correctness (68.2% strict to 93.7% lenient). Two large-scale case studies validate the pipeline against the public record. In Austria, it reconstructs a political party's complete lifecycle, dating internal fractures and tracking personnel into successor factions and court convictions. In a Polish corpus, it uncovers the overlapping economic and governance networks of state-enterprise patronage, alongside the structurally balanced, signed conflict network of the polarized Civic Platform (Platforma Obywatelska, PO)--Law and Justice (Prawo i Sprawiedliwo\'s\'c, PiS) duopoly. By bridging raw multilingual text and structured relational data, our framework provides a robust, replicable foundation for cross-national empirical computational social science.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper presents a modular, fully open-weight multilingual pipeline for joint entity-relation extraction that constructs signed, temporal knowledge graphs of political elites from large news corpora. It combines span-based NER, a three-stage Wikidata linking cascade, and an ontology-constrained mixture-of-experts model with guided decoding. Validation consists of a full-coverage spot-check on a 3491-relation gold standard (68.2% strict to 93.7% lenient textual correctness) plus two case studies that reconstruct Austrian party lifecycles and Polish state-enterprise patronage and PO-PiS conflict networks, claiming these outputs provide a replicable foundation for cross-national computational social science.

Significance. If the extracted signed relations can be shown to correspond to real political ties rather than source-text framing, the pipeline would offer a scalable, replicable alternative to manual coding for studying elite coalitions and patronage at European scale. The open-weight design and cross-lingual capability are genuine strengths that could support reproducible work in comparative politics.

major comments (2)
  1. [Abstract] Abstract: the reported gold-standard evaluation measures only whether the model recovers the relation as stated in the source sentence (textual fidelity). It provides no independent test that the extracted signed edges match actual political ties rather than journalistic selection, framing, or omission. Because the Austrian and Polish case studies likewise compare output only to the same public-record sources used for training text, this untested assumption is load-bearing for the central claim that the pipeline produces usable networks for downstream social-science tasks such as coalition detection.
  2. [Abstract] Abstract: no details are supplied on model architecture, training data composition, inter-annotator agreement for the gold standard, or error analysis. Without these, it is impossible to assess whether the 68.2–93.7 % figures support the claim of a robust, replicable pipeline.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these comments, which highlight important distinctions in validation scope. We respond point-by-point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the reported gold-standard evaluation measures only whether the model recovers the relation as stated in the source sentence (textual fidelity). It provides no independent test that the extracted signed edges match actual political ties rather than journalistic selection, framing, or omission. Because the Austrian and Polish case studies likewise compare output only to the same public-record sources used for training text, this untested assumption is load-bearing for the central claim that the pipeline produces usable networks for downstream social-science tasks such as coalition detection.

    Authors: We agree the gold standard measures textual fidelity to source sentences, which is a prerequisite for any extraction pipeline. The case studies provide qualitative alignment with documented public events but do not constitute an independent test against non-media ground truth for political ties. We will revise the abstract, add a limitations paragraph on media framing/selection effects, and note implications for coalition-detection tasks. revision: partial

  2. Referee: [Abstract] Abstract: no details are supplied on model architecture, training data composition, inter-annotator agreement for the gold standard, or error analysis. Without these, it is impossible to assess whether the 68.2–93.7 % figures support the claim of a robust, replicable pipeline.

    Authors: The full manuscript details model architecture (Section 3), training data (Section 4.1), IAA during gold-standard construction, and error analysis (Section 5). To improve accessibility we will expand the abstract with concise references to these elements and key metrics. revision: yes

Circularity Check

0 steps flagged

No circularity; engineering pipeline with external validation

full rationale

The manuscript presents a modular NLP pipeline for joint entity-relation extraction, evaluated via a held-out 3491-relation gold standard (68.2 % strict / 93.7 % lenient) and two case studies that compare output to the same public-record corpora used as input text. No equations, fitted parameters, predictions, uniqueness theorems, or ansatzes appear; the central claims rest on empirical performance against external benchmarks rather than any derivation that reduces to its own inputs by construction. Self-citations, if present, are not load-bearing for any mathematical or definitional step.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the pipeline implicitly relies on the domain ontology and Wikidata as external resources whose coverage and accuracy are not audited here.

pith-pipeline@v0.9.1-grok · 5839 in / 1201 out tokens · 42775 ms · 2026-06-26T03:42:51.708143+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

69 extracted references · 37 canonical work pages

  1. [1]

    Valuing Public Goods in a Populist World: A Comparative Analysis of Network Dynamics and Societal Outcomes , date =

  2. [2]

    Networks and the Rule of Law: Uncovering Socio-Economic Outcomes , date =

  3. [3]

    2605.18337 , eprinttype =

    Lazzaroni, Ruggero Marino and Lasser, Jana and Solovev, Kirill , title =. 2605.18337 , eprinttype =

    Pith/arXiv arXiv

  4. [4]

    doi:10.1371/journal.pone.0313149 , journaltitle =

    Bro, Naim , title =. doi:10.1371/journal.pone.0313149 , journaltitle =

    work page doi:10.1371/journal.pone.0313149

  5. [5]

    , title =

    Keller, Franziska B. , title =. doi:10.1017/jea.2015.3 , journaltitle =

    work page doi:10.1017/jea.2015.3 2015

  6. [6]

    and Gleditsch, Kristian Skrede and Chiozza, Giacomo , title =

    Goemans, Henk E. and Gleditsch, Kristian Skrede and Chiozza, Giacomo , title =. doi:10.1177/0022343308100719 , journaltitle =

    work page doi:10.1177/0022343308100719

  7. [7]

    doi:10.1017/S0003055420000490 , journaltitle =

    Nyrup, Jacob and Bramwell, Stuart , title =. doi:10.1017/S0003055420000490 , journaltitle =

    work page doi:10.1017/s0003055420000490

  8. [8]

    doi:10.1017/S1049096525000046 , journaltitle =

    Lee, Kyuwon and Paci, Simone and Park, Jeongmin and You, Hye Young and Zheng, Sylvan , title =. doi:10.1017/S1049096525000046 , journaltitle =

    work page doi:10.1017/s1049096525000046

  9. [9]

    doi:10.1080/19331681.2024.2417263 , journaltitle =

    Arslan, Muhammad and Munawar, Saba and Cruz, Christophe , title =. doi:10.1080/19331681.2024.2417263 , journaltitle =

    work page doi:10.1080/19331681.2024.2417263 2024

  10. [10]

    doi:10.1111/ajps.70050 , journaltitle =

    Benoit, Kenneth and De Marchi, Scott and Laver, Conor and Laver, Michael and Ma, Jinshuai , title =. doi:10.1111/ajps.70050 , journaltitle =

    work page doi:10.1111/ajps.70050

  11. [11]

    doi:10.1177/20531680241236239 , journaltitle =

    Heseltine, Michael and Clemm von Hohenberg, Bernhard , title =. doi:10.1177/20531680241236239 , journaltitle =

    work page doi:10.1177/20531680241236239

  12. [12]

    Zaratiana, Urchade and Tomeh, Nadi and Holat, Pierre and Charnois, Thierry , booktitle =

  13. [13]

    and Moazam, Hanna and Miller, Heather and Zaharia, Matei and Potts, Christopher , booktitle =

    Khattab, Omar and Singhvi, Arnav and Maheshwari, Paridhi and Zhang, Zhiyuan and Santhanam, Keshav and Vardhamanan, Sri and Haq, Saiful and Sharma, Ashutosh and Joshi, Thomas T. and Moazam, Hanna and Miller, Heather and Zaharia, Matei and Potts, Christopher , booktitle =

  14. [14]

    and Dietrich, Nick , title =

    Karstens, Mikaela and Soules, Michael J. and Dietrich, Nick , title =. doi:10.1017/S1049096522001317 , journaltitle =

    work page doi:10.1017/s1049096522001317

  15. [15]

    Extracting Knowledge from Parliamentary Debates for Studying Political Culture and Language , pages =

    Tamper, Minna and Leal, Rafael and Sinikallio, Laura and Leskinen, Petri and Tuominen, Jouni and Hyvönen, Eero , booktitle =. Extracting Knowledge from Parliamentary Debates for Studying Political Culture and Language , pages =

  16. [16]

    Fenwick, Clare and Shirali, Ramin and Fazekas, Mihaly and Kantorowicz, Jaroslaw , title =

  17. [17]

    The Rise and Decline of Nations: Economic Growth, Stagflation, and Social Rigidities , date =

    Olson, Mancur , publisher =. The Rise and Decline of Nations: Economic Growth, Stagflation, and Social Rigidities , date =

  18. [18]

    and Leonardi, Robert and Nanetti, Raffaella Y

    Putnam, Robert D. and Leonardi, Robert and Nanetti, Raffaella Y. , publisher =. Making Democracy Work: Civic Traditions in Modern Italy , date =

  19. [19]

    , publisher =

    Acemoglu, Daron and Robinson, James A. , publisher =. Why Nations Fail: The Origins of Power, Prosperity, and Poverty , date =

  20. [20]

    Political Networks: The Structural Perspective , series =

    Knoke, David , publisher =. Political Networks: The Structural Perspective , series =

  21. [21]

    , title =

    Grimmer, Justin and Stewart, Brandon M. , title =. doi:10.1093/pan/mps028 , journaltitle =

    work page doi:10.1093/pan/mps028

  22. [22]

    and Stewart, Brandon M

    Grimmer, Justin and Roberts, Margaret E. and Stewart, Brandon M. , publisher =. Text as Data: A New Framework for Machine Learning and the Social Sciences , date =

  23. [23]

    doi:10.1111/1468-4446.13203 , journaltitle =

    Bühlmann, Felix and Christesen, Caroline Ahler and Cousin, Bruno and Denord, François and Ellersgaard, Christoph Houman and Lagneau-Ymonet, Paul and Larsen, Anton Grau and Savage, Mike and Thine, Sylvain and Young, Kevin and others , title =. doi:10.1111/1468-4446.13203 , journaltitle =

    work page doi:10.1111/1468-4446.13203

  24. [24]

    doi:10.1017/psrm.2017.28 , journaltitle =

    Mahdavi, Paasha , title =. doi:10.1017/psrm.2017.28 , journaltitle =

    work page doi:10.1017/psrm.2017.28 2017

  25. [25]

    doi:10.1093/pan/mpn006 , journaltitle =

    van Atteveldt, Wouter and Kleinnijenhuis, Jan and Ruigrok, Nel , title =. doi:10.1093/pan/mpn006 , journaltitle =

    work page doi:10.1093/pan/mpn006

  26. [26]

    doi:10.1017/nws.2025.4 , journaltitle =

    Angst, Mario and Müller, Neitah Noemi and Walker, Viviane , title =. doi:10.1017/nws.2025.4 , journaltitle =

    work page doi:10.1017/nws.2025.4 2025

  27. [27]

    2603.18010 , eprinttype =

    Zhu, Yifei and Yang, Songpo and Zhu, Jiangnan and Jiang, Junyan , title =. 2603.18010 , eprinttype =

    arXiv

  28. [28]

    doi:10.3233/SW-222986 , journaltitle =

    Sevgili, Özge and Shelmanov, Artem and Arkhipov, Mikhail and Panchenko, Alexander and Biemann, Chris , title =. doi:10.3233/SW-222986 , journaltitle =

    work page doi:10.3233/sw-222986

  29. [29]

    and Hussain, Shakeel and Smith, Geoffrey , title =

    Guellil, Imane and Garcia-Dominguez, Antonio and Lewis, Peter R. and Hussain, Shakeel and Smith, Geoffrey , title =. doi:10.1007/s10115-023-02059-2 , journaltitle =

    work page doi:10.1007/s10115-023-02059-2

  30. [30]

    Multilingual Autoregressive Entity Linking

    De Cao, Nicola and Wu, Ledell and Popat, Kashyap and Artetxe, Mikel and Goyal, Naman and Plekhanov, Mikhail and Zettlemoyer, Luke and Cancedda, Nicola and Riedel, Sebastian and Petroni, Fabio , title =. doi:10.1162/tacl_a_00460 , journaltitle =

    work page doi:10.1162/tacl_a_00460

  31. [31]

    WikiProject Every Politician , url =

  32. [32]

    EveryPolitician: Data About Every National Legislature in the World , url =

  33. [33]

    Miles, Alistair and Bechhofer, Sean , institution =

  34. [34]

    Grammar-Constrained Decoding for Structured

    Geng, Saibo and Josifoski, Martin and Peyrard, Maxime and West, Robert , booktitle =. Grammar-Constrained Decoding for Structured

  35. [35]

    doi:10.31286/JP.97.1.7 , journaltitle =

    Kieraś, Witold and Woliński, Marcin , title =. doi:10.31286/JP.97.1.7 , journaltitle =

    work page doi:10.31286/jp.97.1.7

  36. [36]

    , booktitle =

    Qi, Peng and Zhang, Yuhao and Zhang, Yuhui and Bolton, Jason and Manning, Christopher D. , booktitle =. Stanza: A Python Natural Language Processing Toolkit for Many Human Languages , pages =

  37. [37]

    Garbe, Wolf , title =

  38. [38]

    Qdrant: High-Performance Vector Search Engine , date =

  39. [39]

    and Zhang, Hao and Stoica, Ion , booktitle =

    Kwon, Woosuk and Li, Zhuohan and Zhuang, Siyuan and Sheng, Ying and Zheng, Lianmin and Yu, Cody Hao and Gonzalez, Joseph E. and Zhang, Hao and Stoica, Ion , booktitle =. Efficient Memory Management for Large Language Model Serving with PagedAttention , pages =

  40. [40]

    doi:10.1177/0888325420953485 , journaltitle =

    Szarzec, Katarzyna and Totleben, Bartosz and Piątek, Dawid , title =. doi:10.1177/0888325420953485 , journaltitle =

    work page doi:10.1177/0888325420953485

  41. [41]

    doi:10.1111/j.1475-6765.2004.00169.x , journaltitle =

    McMenamin, Iain , title =. doi:10.1111/j.1475-6765.2004.00169.x , journaltitle =

    work page doi:10.1111/j.1475-6765.2004.00169.x 2004

  42. [42]

    doi:10.1177/0888325420950800 , journaltitle =

    Bill, Stanley , title =. doi:10.1177/0888325420950800 , journaltitle =

    work page doi:10.1177/0888325420950800

  43. [43]

    The Hidden Cost of Structure: How Constrained Decoding Affects Language Model Performance , pages =

    Schall, Maximilian and de Melo, Gerard , booktitle =. The Hidden Cost of Structure: How Constrained Decoding Affects Language Model Performance , pages =. doi:10.26615/978-954-452-098-4-124 , url =

    work page doi:10.26615/978-954-452-098-4-124

  44. [44]

    A Glimpse into Babel: An Analysis of Multilinguality in Wikidata , date =

    Kaffee, Lucie-Aimée and Piscopo, Alessandro and Vougiouklis, Pavlos and Simperl, Elena and Carr, Leslie and Pintscher, Lydia , booktitle =. A Glimpse into Babel: An Analysis of Multilinguality in Wikidata , date =

  45. [45]

    doi:10.1140/epjds/s13688-024-00514-w , journaltitle =

    Schoch, David and Chan, Chung-hong and Wagner, Claudia and Bleier, Arnim , title =. doi:10.1140/epjds/s13688-024-00514-w , journaltitle =

    work page doi:10.1140/epjds/s13688-024-00514-w

  46. [46]

    and Desposato, Scott and Dreber, Anna and Genovese, Federica and Green, Donald P

    Brodeur, Abel and Esterling, Kevin and Ankel-Peters, Jörg and Bueno, Natália S. and Desposato, Scott and Dreber, Anna and Genovese, Federica and Green, Donald P. and Hepplewhite, Matthew and Hoces de la Guardia, Fernando and Johannesson, Magnus and Kotsadam, Andreas and Miguel, Edward and Velez, Yamil R. and Young, Lauren , title =. doi:10.1177/2053168024...

    work page doi:10.1177/20531680241233439

  47. [47]

    , publisher =

    Burt, Ronald S. , publisher =. Structural Holes: The Social Structure of Competition , date =

  48. [48]

    and Ansell, Christopher K

    Padgett, John F. and Ansell, Christopher K. , title =. doi:10.1086/230190 , journaltitle =

    work page doi:10.1086/230190

  49. [49]

    and King, Gary , title =

    Hopkins, Daniel J. and King, Gary , title =. doi:10.1111/j.1540-5907.2009.00428.x , journaltitle =

    work page doi:10.1111/j.1540-5907.2009.00428.x 2009

  50. [50]

    doi:10.1162/99608f92.5317da47 , journaltitle =

    Chen, Lingjiao and Zaharia, Matei and Zou, James , title =. doi:10.1162/99608f92.5317da47 , journaltitle =

    work page doi:10.1162/99608f92.5317da47

  51. [51]

    doi:10.1080/01402380600842452 , journaltitle =

    Markowski, Radosław , title =. doi:10.1080/01402380600842452 , journaltitle =

    work page doi:10.1080/01402380600842452

  52. [52]

    doi:10.1177/0002716218809322 , journaltitle =

    Tworzecki, Hubert , title =. doi:10.1177/0002716218809322 , journaltitle =

    work page doi:10.1177/0002716218809322

  53. [53]

    doi:10.1177/002234336500200104 , journaltitle =

    Galtung, Johan and Ruge, Mari Holmboe , title =. doi:10.1177/002234336500200104 , journaltitle =

    work page doi:10.1177/002234336500200104

  54. [54]

    doi:10.1080/1461670X.2016.1150193 , journaltitle =

    Harcup, Tony and O'Neill, Deirdre , title =. doi:10.1080/1461670X.2016.1150193 , journaltitle =

    work page doi:10.1080/1461670x.2016.1150193 2016

  55. [55]

    doi:10.2753/PPC1075-8216550304 , journaltitle =

    Heinisch, Reinhard , title =. doi:10.2753/PPC1075-8216550304 , journaltitle =

    work page doi:10.2753/ppc1075-8216550304

  56. [56]

    doi:10.1177/1354068811400522 , journaltitle =

    Luther, Kurt Richard , title =. doi:10.1177/1354068811400522 , journaltitle =

    work page doi:10.1177/1354068811400522

  57. [57]

    The Haider Phenomenon in Austria , location =

  58. [58]

    and Plasser, Fritz , editor =

    Müller, Wolfgang C. and Plasser, Fritz , editor =. Austria: The 1990 Campaign , location =. Electoral Strategies and Political Marketing , publisher =

    1990

  59. [59]

    227, National Bank of Poland) , date =

    The Constitution of the Republic of Poland (Art. 227, National Bank of Poland) , date =

  60. [60]

    Cartwright, Dorwin and Harary, Frank , title =

  61. [61]

    Schuldspruch gegen Ex-

  62. [62]

    Dobernig-Urteil ist rechtskräftig , date =

  63. [63]

    Revisiting Large Language Models as Zero-Shot Relation Extractors , pages =

    Li, Guozheng and Wang, Peng and Ke, Wenjun , booktitle =. Revisiting Large Language Models as Zero-Shot Relation Extractors , pages =

  64. [64]

    Wikidata: a free collaborative knowledgebase , url =

    Vrandečić, Denny and Krötzsch, Markus , title =. doi:10.1145/2629489 , journaltitle =

    work page doi:10.1145/2629489

  65. [65]

    From Louvain to Leiden: Guaranteeing Well-Connected Communities

    Traag, Vincent A. and Waltman, Ludo and van Eck, Nees Jan , title =. doi:10.1038/s41598-019-41695-z , journaltitle =

    work page doi:10.1038/s41598-019-41695-z

  66. [66]

    and Guillaume, Jean-Loup and Lambiotte, Renaud and Lefebvre, Etienne , title =

    Blondel, Vincent D. and Guillaume, Jean-Loup and Lambiotte, Renaud and Lefebvre, Etienne , title =. doi:10.1088/1742-5468/2008/10/P10008 , journaltitle =

    work page doi:10.1088/1742-5468/2008/10/p10008 2008

  67. [67]

    doi:10.1080/0022250X.2001.9990249 , journaltitle =

    Brandes, Ulrik , title =. doi:10.1080/0022250X.2001.9990249 , journaltitle =

    work page doi:10.1080/0022250x.2001.9990249 2001

  68. [68]

    Page, Lawrence and Brin, Sergey and Motwani, Rajeev and Winograd, Terry , institution =. The

  69. [69]

    Content Analysis: An Introduction to Its Methodology , edition =

    Krippendorff, Klaus , publisher =. Content Analysis: An Introduction to Its Methodology , edition =