Overview of the TalentCLEF 2025: Skill and Job Title Intelligence for Human Capital Management

Alvaro Rodrigo; Daniel Deniz; Hermenegildo Fabregat; Laura Garc\'ia-Sardi\~na; Luis Gasco; Paula Estrella; Rabih Zbib

REVIEW 3 major objections 6 minor 50 references

Overview of the TalentCLEF 2025: Skill and Job Title Intelligence for Human Capital Management

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read TalentCLEF 2025 establishes the first public benchmark for skill and job title intelligence, and its results indicate that training strategy matters more than model size.

desk verdict A genuinely useful public benchmark for HR-NLP, but the gold-standard reliability is under-documented and at least one table error needs fixing before the results can be fully trusted. read the letter →

arxiv 2507.13275 v1 pith:OVVEUZV2 submitted 2025-07-17 cs.CL cs.AIcs.IR

Luis Gasco , Hermenegildo Fabregat , Laura García-Sardiña , Paula Estrella , Daniel Deniz , Alvaro Rodrigo , Rabih Zbib This is my paper

classification cs.CLcs.AIcs.IR

keywords TalentCLEF2025jobtitlematchingskillpredictionhumancapitalmanagementevaluationbenchmarkmultilingualNLPgenderbiascontrastivelearning

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

TalentCLEF 2025 is the first public evaluation benchmark for skill and job title intelligence in human capital management, built from real, anonymized job applications and manual annotations in English, Spanish, German, and Chinese. The paper reports two tasks, multilingual job title matching and job-title-based skill prediction, and 76 registered teams with more than 280 submissions. Its central finding is that the way a model is trained matters more than its size: fine-tuned encoder models around 500M parameters outperformed decoder-based embedding models up to 7B parameters, and LLM-based data augmentation added up to 8 MAP points. If the finding holds, organizations can build better talent-matching systems by focusing on training strategy and data rather than on scaling models.

What carries the argument

The argument is carried by the benchmark itself: positive relevance pairs derived from real application behavior, a training set generated from the ESCO taxonomy, manually annotated development and test sets that preserve real-world noise and gender-marked variation, and evaluation with Mean Average Precision plus Rank Biased Overlap for gender bias. Participant systems, described via a common scheme, supply the evidence that contrastive fine-tuning with losses such as InfoNCE and GIST, together with LLM-based data augmentation and reranking, drives performance more than parameter count.

What would settle it

Re-annotate a random sample of Task A and Task B test queries with a second, independent annotation team and compute an agreement statistic; if agreement is low, the gold labels are unreliable and the rankings, along with the conclusion that training strategy matters more than model size, could shift. A simpler check is to drop queries whose labels were edited during the review phase and see whether the top systems' ordering and the size-versus-strategy pattern survive.

Watch

Extended reading notes

Core claim

The paper's central claim is that TalentCLEF 2025 provides the first shared, openly available evaluation resource for two human capital management NLP tasks, and that the campaign's results show training strategy outweighs model size. The corpora were built from real job applications: queries are job-offer titles, Task A corpus elements are candidates' most recent job titles, and Task B corpus elements are skills extracted from CVs, with training data generated from the ESCO taxonomy and development and test sets manually annotated. The best Task A system reached an average MAP of 0.534 across English, Spanish, and German, and the best cross-lingual systems reached 0.514; the best Task B system reached 0.360. The paper argues that encoder-based models around 500M parameters, fine-tuned with contrastive losses and sometimes enhanced with LLM-based augmentation or reranking, outperformed larger decoder-based embeddings, and that gender-bias evaluation via rank consistency showed top systems with RBO values above 0.97, with two systems producing identical rankings for gendered forms.

Load-bearing premise

The load-bearing premise is that the manually produced relevance judgments in the development and test sets are correct and consistent enough to serve as gold standard, even though the paper reports that 28.4% of Task A matches and about a third of Task B test annotations were edited during review and gives no inter-annotator agreement statistic.

Editorial extensions

If this is right

Future research on job title matching and skill prediction can compare methods on a common public resource instead of proprietary datasets.
Fine-tuned encoder models of about 500M parameters are a cost-effective default for job-title and skill ranking, avoiding the compute of 7B-parameter decoders.
LLM-based data augmentation and reranking provide measurable gains, up to 8 MAP points, and should be part of the standard toolkit.
Gender bias in ranking systems can be measured with RBO and reported alongside MAP, giving a concrete fairness signal.
Cross-lingual job title matching is feasible but weaker than monolingual matching, so multilingual transfer remains an open target.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The size-versus-strategy conclusion comes from a single campaign, so its generality across other languages, taxonomies, and labor-market settings is untested.
Because the gold labels come from one annotation process with no reported inter-annotator agreement, future users should treat the rankings as provisional until independent re-annotation confirms stability.
Organizations could adapt the benchmark's gender-bias protocol as an audit for their own production matching systems before deployment.
The Chinese track, which had no training data, suggests that LLM translation and augmentation can provide a practical recipe for entering new languages quickly.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This is a useful shared-task overview and a genuine resource: TalentCLEF 2025 gives the HR-NLP community public data for job title matching and skill prediction, with real job-application text, manual annotation, and an open evaluation platform. The dataset release on Zenodo plus the Codabench setup is the kind of contribution the field needed, and the paper describes the corpus construction in enough detail that someone could reproduce or critique the curation pipeline.

The paper does several things well. It is transparent about the messiness of the data: cleaning rules, annotator flags, the review-stage edit rates, and the use of ESCO for training and mapping are all stated. The gender-bias evaluation via RBO is a nice addition, and the aggregate observation that fine-tuned 500M encoders beat much larger decoder models in Task A is worth reporting, even if it is not a controlled comparison. The cross-lingual and Chinese tracks also stretch the benchmark in useful directions.

The soft spots are real but mostly fixable. The biggest is the gold standard. Section 3.2 reports that 28.4% of Task A initial matches were edited, most from one annotator, and about a third of Task B test annotations were edited, but no inter-annotator agreement or label-stability analysis is given. The MAP gaps between the top two teams in each task are small (0.534 vs 0.517; 0.360 vs 0.345), so without a noise floor those differences may not mean much. The paper should either compute IAA on a held-out sample or acknowledge that the benchmark currently certifies relative ordering only for larger gaps.

Second, Tables 5 and 8 list identical scores for VerbaNexAI and the TalentCLEF Baseline; Table 8 even gives the baseline a different system ID. That looks like a copy/paste error and needs a correction before the results can be trusted. Third, the ESCO circularity for Task B is real: training data and gold standard both derive from ESCO, so teams like NLPnorth and moali that used ESCO metadata get an unmeasured advantage. That does not sink the benchmark, but the organizers should quantify or at least discuss it. Finally, the 'first benchmark' claim should be worded precisely, since the group's own MELO benchmark exists; the distinction between a single benchmark and an evaluation campaign is fine but should be explicit.

Who should read this? Anyone working on job title matching, skill extraction, or fair HR NLP. It deserves a serious referee: the resource is valuable, the writing is clear, and the issues are correctable. I would accept it with revisions requiring the table fix and an explicit discussion of label reliability.

Referee Report

3 major / 6 minor

Summary. This paper presents the overview of TalentCLEF 2025, a shared evaluation campaign organized at CLEF comprising Task A (multilingual job title matching in English, Spanish, German, and Chinese) and Task B (job title-based skill prediction in English). The datasets are built from real anonymized job applications, with training data automatically generated from the ESCO taxonomy and development/test sets manually annotated. The paper describes the corpus creation workflow, the participant systems, and the official results, reporting MAP for the main rankings and RBO for gender bias evaluation. The two central claims are that TalentCLEF is the first public benchmark for skill and job title intelligence and that the results show training strategy has a larger effect than model size alone.

Significance. If the benchmark is sound, the paper makes a valuable contribution: it provides a public, reusable evaluation resource for an underserved area of NLP, documents a detailed data-construction pipeline, opens the evaluation on Codabench for continuous use, and attracts substantial community participation (76 teams, 280+ submissions). The inclusion of gender-bias evaluation through RBO is a constructive addition. The observational comparison of encoder- versus decoder-based systems across two tasks gives the field a useful starting point. However, the scientific conclusions rest on the reliability of the manual relevance judgments and on the statistical robustness of small MAP differences, and these are not currently demonstrated; the paper also does not fully address the circularity introduced by using ESCO both for training generation and for gold-standard annotation in Task B.

major comments (3)

[Section 3.2] The reliability of the gold standard is not established. The review phase edited 28.4% of the initial Task A relevance matches, most originating from a single annotator, and about one-third of the Task B test annotations, yet no inter-annotator agreement measure (e.g., Cohen's kappa) and no label-stability analysis are reported. Since all team rankings and the abstract's central claim depend on these manual judgments, the paper should report IAA on a sample and ideally show that the final rankings are robust to annotation noise, for example by perturbing a fraction of the judgments and recomputing MAP.
[Section 5.1 (Table 5) and Section 5.2 (Table 9)] The reported differences between top-ranked systems are small (0.534 vs 0.517 vs 0.515 in Task A; 0.360 vs 0.345 in Task B), but no confidence intervals, error bars, or significance tests are provided. The claim in Section 5.3 that 'training strategies have a larger effect than the size of the model alone' is an observational comparison of non-controlled system configurations; without query-level variance or paired significance tests (e.g., bootstrap or Wilcoxon signed-rank), the conclusion is supported only by point estimates. Please add query-level statistical analysis or temper the claim.
[Section 2.2 and Section 3.2] There is a potential circularity in Task B: both the training data are generated from the ESCO taxonomy and the gold-standard relevance judgments are produced using the ESCO taxonomy as a reference, while several participating teams (e.g., NLPnorth, moali) explicitly use ESCO descriptions as external data. This design advantages ESCO-aware systems and makes it difficult to interpret Task B rankings as evidence of generalizable job-skill prediction. The paper should discuss this limitation and, if possible, quantify the ESCO advantage, for example by comparing systems with and without ESCO-derived features or by ablating ESCO-based elements from the gold standard.

minor comments (6)

[Tables 5 and 8] The rows for VerbaNexAI and TalentCLEF Baseline are identical in both tables (0.360/0.408/0.348/0.324 in Table 5 and 0.336/0.915/0.893/0.937 in Table 8); this is almost certainly a copy/paste error and the actual values should be verified.
[Section 1] The claim that TalentCLEF is 'the first evaluation campaign' in this area should be qualified in light of the authors' own MELO benchmark [29], which is an evaluation benchmark for occupation entity linking; please clarify the difference in scope.
[Section 5.3 and Figures 2-3] The figures are informative, but the paper should state how many submissions fall into each parameter-size bin and clarify whether 'training strategy' is a controlled variable or a post-hoc grouping; without this, the reader cannot separate the effect of model size from the effect of other system choices.
[Section 3.2] The paper reports a 'TalentCLEF Baseline' in several tables but never describes what this baseline is (model architecture, training data, or parameters); a short description is needed to interpret the comparison.
[Section 3.2] For Task B, the selection threshold ('7,493 skills that appeared at least 90 times') should specify whether the frequency was computed across the whole corpus or per job offer, and the choice of 90 should be justified as it is a free parameter in the corpus construction.
[Introduction] The phrase 'real-word complexities' in the first section should be 'real-world complexities'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: this is a benchmark overview, not a derivation chain; its conclusions rest on external participant submissions and manually annotated gold labels.

full rationale

TalentCLEF 2025 is a shared-task overview and contains no derivation chain in which an output is constructed from its own input. The central empirical claim (Section 5.3) that training strategy matters more than model size is an observational comparison of participant submissions, not a quantity fitted to the labels and then re-reported as a prediction. The Task B setup does have train/gold overlap with the ESCO taxonomy: Section 2.2 states the training set was 'automatically generated leveraging the job title to skill relevance information from ESCO,' and Section 3.2 states annotators used 'the ESCO taxonomy as a reference.' This is a potential shortcut or confound for systems that exploit ESCO metadata, but it is benchmark design rather than a circular reduction, because the task is explicitly to rank ESCO skills and the gold labels were manually reviewed. The annotation reliability concerns flagged in Section 3.2 (28.4% of Task A initial matches edited, about a third of Task B test annotations edited, with no inter-annotator agreement reported) and the identical Baseline/VerbaNexAI rows in Tables 5 and 8 are correctness and data-quality risks, not self-referential derivations. The self-citations ([10], [14], [29], [44]) appear in the literature review and are not load-bearing for the benchmark's results or conclusions.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The benchmark construction relies on a few hand-chosen parameters (xN, skill frequency threshold, cluster count) and domain assumptions about the validity of application-based relevance, the reliability of manual annotation, and the authority of ESCO. No new theoretical entities are introduced.

free parameters (3)

Number of candidate profiles per job offer (xN)
Section 3.1: a fixed number xN of relevant candidate profiles was retrieved for each selected offer. The value is not specified, but it determines the size and composition of the corpus.
Minimum skill frequency threshold = 90
Section 3.2: Task B selected 7,493 skills that appeared at least 90 times in the dataset. This hand-chosen threshold determines the skill vocabulary.
K-means cluster count
Section 3.1: job offers were clustered using K-means to sample diverse offers. The number of clusters is not specified.

assumptions (3)

domain assumption Applicants who applied to a job are a valid source of positive relevance pairs
Section 3: 'applicants were considered relevant as they had applied to the corresponding position.' This is the basis of the automatically extracted positive pairs before manual annotation.
domain assumption Manual annotation of relevance is reliable ground truth
Section 3.2: dev/test sets are manually annotated, yet 28.4% of Task A initial matches and about a third of Task B test annotations were edited during review, with no inter-annotator agreement reported.
domain assumption ESCO taxonomy is the correct reference for skill relevance
Section 3.2: Task B skills were manually mapped to ESCO v1.2.0 and annotators used ESCO as a reference, tying the gold standard to ESCO definitions.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Overview of the TalentCLEF 2025: Skill and Job Title Intelligence for Human Capital Management." pith.science (2026). https://pith.science/paper/OVVEUZV2

@misc{pith2026250713275,
  author       = {Pith},
  title        = {Pith review of: Overview of the TalentCLEF 2025: Skill and Job Title Intelligence for Human Capital Management},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OVVEUZV2}},
  note         = {Machine review of arXiv:2507.13275}
}

read the original abstract

Advances in natural language processing and large language models are driving a major transformation in Human Capital Management, with a growing interest in building smart systems based on language technologies for talent acquisition, upskilling strategies, and workforce planning. However, the adoption and progress of these technologies critically depend on the development of reliable and fair models, properly evaluated on public data and open benchmarks, which have so far been unavailable in this domain. To address this gap, we present TalentCLEF 2025, the first evaluation campaign focused on skill and job title intelligence. The lab consists of two tasks: Task A - Multilingual Job Title Matching, covering English, Spanish, German, and Chinese; and Task B - Job Title-Based Skill Prediction, in English. Both corpora were built from real job applications, carefully anonymized, and manually annotated to reflect the complexity and diversity of real-world labor market data, including linguistic variability and gender-marked expressions. The evaluations included monolingual and cross-lingual scenarios and covered the evaluation of gender bias. TalentCLEF attracted 76 registered teams with more than 280 submissions. Most systems relied on information retrieval techniques built with multilingual encoder-based models fine-tuned with contrastive learning, and several of them incorporated large language models for data augmentation or re-ranking. The results show that the training strategies have a larger effect than the size of the model alone. TalentCLEF provides the first public benchmark in this field and encourages the development of robust, fair, and transferable language technologies for the labor market.

Figures

Figures reproduced from arXiv: 2507.13275 by the authors.

**Figure 2.** Relationship between model size and average Mean Average Preci [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗

**Figure 3.** Relationship between model size (number of parameters) and average [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 36 canonical work pages

[1]

In: CLEF (Working Notes) (2025)

Ali, M.: Enhancing Job-Skill Matching with LLM-Driven Data Augmentation and Fine-Tuned Bi-Encoders. In: CLEF (Working Notes) (2025)

work page 2025
[2]

In: CLEF (Working Notes) (2025)

Barakat, R., Mokhtar, O., Torki, M., Elmakky, N.: AlexU-NLP at TalentCLEF 2025: Curriculum-Driven Hybrid Retrieval for Multilingual Job Title Matching. In: CLEF (Working Notes) (2025)

work page 2025
[3]

In: CLEF (Working Notes) (2025)

Bhangale, C., Gabhane, P ., Kumar M, A.: Fine-Tuned Sentence Transformer for Multilingual Job Title Matching. In: CLEF (Working Notes) (2025)

work page 2025
[4]

In: Proceedings of the 18th ACM Conference on Recommender Systems

Bogers, T., Graus, D., Kaya, M., Johnson, C., Decorte, J.J., De Bie, T.: Fourth workshop on recommender systems for human resources (recsys in hr 2024). In: Proceedings of the 18th ACM Conference on Recommender Systems. p. 1222–1226. RecSys ’24, Association for Computing Machin- ery, New York, NY, USA (2024). https://doi.org/10.1145/3640457.3687109, https...

work page arXiv 2024
[5]

In: CLEF (Working Notes) (2025)

Brikman, A., Sana, M., Ruegger, H.: Multilingual Job Title Matching with MPNet- Based Sentence Transformers. In: CLEF (Working Notes) (2025)

work page 2025
[6]

In: Ku, L., Martins, A., Srikumar, V

Chen, J., Xiao, S., Zhang, P ., Luo, K., Lian, D., Liu, Z.: M3-embedding: Multi- linguality, multi-functionality, multi-granularity text embeddings through self- knowledge distillation. In: Ku, L., Martins, A., Srikumar, V . (eds.) Findings of the Association for Computational Linguistics, ACL 2024, Bangkok, Thailand and virtual meeting, August 11-16, 202...

doi:10.18653/v1/2024.findings-acl.137 2024
[7]

CoRR abs/2412.03223 (2024)

Choi, C., Kim, J., Lee, S., Kwon, J., Gu, S., Kim, Y., Cho, M., Sohn, J.: Linq-embed-mistral technical report. CoRR abs/2412.03223 (2024). https://doi.org/10.48550/ARXIV .2412.03223, https://doi.org/10.48550/arXiv.2412.03223

arXiv doi:10.48550/arxiv.2412.03223 2024
[8]

In: CLEF (Working Notes) (2025)

Decorte, J.J., De Lange, M., Van Hautte, J.: TechWolf at TalentCLEF 2025: Mul- tilingual JobBERT-V2 for Cross-Lingual Job Title Matching. In: CLEF (Working Notes) (2025)

work page 2025

Show all 50 references

[9]

CoRR abs/2109.09605 (2021), https://arxiv.org/abs/2109.09605

Decorte, J., Hautte, J.V ., Demeester, T., Develder, C.: Jobbert: Un- derstanding job titles through skills. CoRR abs/2109.09605 (2021), https://arxiv.org/abs/2109.09605

2021 arXiv
[10]

In: Kaya, M., Bogers, T., Graus, D., Johnson, C., Decorte, J., Bie, T.D

Deniz, D., Retyk, F ., García-Sardiña, L., Fabregat, H., Gascó, L., Zbib, R.: Com- bined unsupervised and contrastive learning for multilingual job recommenda- tion. In: Kaya, M., Bogers, T., Graus, D., Johnson, C., Decorte, J., Bie, T.D. (eds.) Proceedings of the 4th Workshop...

2024
[11]

Di Battista, Attilio and Grayling, Sam and Hasselaar, Elselot and Leopold, Till and Li, Ricky and Rayner, Mark and Zahidi, Saadia: Future of jobs report
[12]

CoRR abs/2407.21783 (2024)

Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Yang, A., Fan, A., Goyal, A., Hartshorn, A., Yang, A., Mitra, A., Sravankumar, A., Korenev, A., Hinsvark, A., Rao, A., Zhang, A., Rodriguez, A., Gregerson, A., Spataru, A., Roziè...

2024 doi arXiv
[13]

Procesamiento del Lenguaje Natural 73, 83–94 (2024), http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/view/6602

Fabregat, H., Poves, R., Alvarez, L.L., Retyk, F ., García-Sardiña, L., Zbib, R.: Inductive graph neural network for job-skill frame- work analysis. Procesamiento del Lenguaje Natural 73, 83–94 (2024), http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/view/6602

2024
[14]

In: European Conference on Information Retrieval

Gasco, L., Fabregat, H., García-Sardiña, L., Deniz, D., Rodrigo, A., Estrella, P ., Zbib, R.: TalentCLEF at CLEF2025: Skill and Job Title Intelligence for Human Capital Management. In: European Conference on Information Retrieval. pp. 479–486. Springer (2025)

2025
[15]

Applied Soft Computing 101, 107049 (2021)

Giabelli, A., Malandri, L., Mercorio, F ., Mezzanzanica, M., Seveso, A.: Skills2job: A recommender system that encodes job of- fer embeddings on graph databases. Applied Soft Computing 101, 107049 (2021). https://doi.org/10.1016/J.ASOC.2020.107049, https://doi.org/10.1016/j.as...

2021
[16]

In: CLEF (Working Notes) (2025)

Ho, T.N., Ho, T.T.T., Dang, V .T.: NT Team at Multilingual Job Title Matching Task A: Job Matching via Large Language Model-Based Description Generation and Retrieval. In: CLEF (Working Notes) (2025)

2025
[17]

Association for Computational Linguistics

Hruschka, E., Lake, T., Otani, N., Mitchell, T.: Proceedings of the first work- shop on natural language processing for human resources (nlp4hr 2024). In: Proceedings of the First Workshop on Natural Language Processing for Human Resources (NLP4HR 2024). "Association for Compu...

2024
[18]

https://ai4hrpes.github.io/ecmlpkdd2023/program/, accessed: 2024-10-17

Kang, B., De Bie, T., Sebag, M., Largeron, C.: Ai for human re- sources and people analytics workshop program, ecml pkdd 2023. https://ai4hrpes.github.io/ecmlpkdd2023/program/, accessed: 2024-10-17

2023
[19]

In: Kaya, M., Bogers, T., Graus, D., Verbert, K., Gutiérrez, F

Lavi, D., Medentsiy, V ., Graus, D.: consultantbert: Fine-tuned siamese sentence- bert for matching jobs and job seekers. In: Kaya, M., Bogers, T., Graus, D., Verbert, K., Gutiérrez, F . (eds.) Proceedings of the Workshop on Recommender Systems for Human Resources (RecSys in H...

2021
[20]

CoRR abs/2308.03281 (2023)

Li, Z., Zhang, X., Zhang, Y., Long, D., Xie, P ., Zhang, M.: Towards general text embeddings with multi-stage contrastive learning. CoRR abs/2308.03281 (2023). https://doi.org/10.48550/ARXIV .2308.03281, https://doi.org/10.48550/arXiv.2308.03281

2023 doi arXiv
[21]

LinkedIn: Work change report: Ai is coming to work (January 2025), https://economicgraph.linkedin.com/content/dam/me/economicgraph/en- us/PDF/Work-Change-Report.pdf, accessed: 2025-05-18

2025
[22]

In: CLEF (Working Notes) (2025)

Llamas, J., Puertas, E., Serrano, J., Martinez, J.: COTECMAR–UTB at TalentCLEF 2025: Linking Job Titles and ESCO Skills with Sentence Transformer Embed- dings. In: CLEF (Working Notes) (2025)

2025
[23]

ManpowerGroup: 2024 global talent shortage (2024), https://go.manpowergroup.com, accessed: 2025-05-19

2024
[24]

Database 2023, 1–24 (2023)

Miranda-Escalada, A., Mehryary, F ., Luoma, J., Estrada-Zavala, D., Gasco, L., Pyysalo, S., Valencia, A., Krallinger, M.: Overview of drugprot task at biocreative vii: data and methods for large-scale text mining and knowledge graph generation of heterogenous chemical–protein ...

2023 doi
[25]

In: CLEF (Working Notes) (2025)

Moreno Novoa, M., Martínez-Santos, J.C., Serrano, J., Puertas, E.: VerbaNex at TalentCLEF2025: Semantic Matching of Multilingual Job Titles through a Frame- work Integrating ESCO Taxonomy. In: CLEF (Working Notes) (2025)

2025
[26]

In: Experimental IR Meets Multilinguality, Multimodality, and Interaction

Nentidis, A., Katsimpras, G., Vandorou, E., Krithara, A., Miranda-Escalada, A., Gasco, L., Krallinger, M., Paliouras, G.: Overview of bioasq 2022: the tenth bioasq challenge on large-scale biomedical semantic indexing and question answering. In: Experimental IR Meets Multiling...

2022
[27]

CoRR abs/1807.03748 (2018), http://arxiv.org/abs/1807.03748

van den Oord, A., Li, Y., Vinyals, O.: Representation learning with contrastive pre- dictive coding. CoRR abs/1807.03748 (2018), http://arxiv.org/abs/1807.03748

2018 arXiv
[28]

In: Kaya, M., Bogers, T., Graus, D., Johnson, C., Decorte, J

Retyk, F ., Fabregat, H., Aizpuru, J., Taglio, M., Zbib, R.: Résumé parsing as hierarchical sequence labeling: An empirical study. In: Kaya, M., Bogers, T., Graus, D., Johnson, C., Decorte, J. (eds.) Proceedings of the 3rd Workshop on Recommender Systems for Human Resources (R...

2023
[29]

In: Kaya, M., Bogers, T., Graus, D., Johnson, C., Decorte, J., Bie, T.D

Retyk, F ., Gascó, L., Carrino, C.P ., Deniz, D., Zbib, R.: MELO: an evaluation benchmark for multilingual entity linking of occupations. In: Kaya, M., Bogers, T., Graus, D., Johnson, C., Decorte, J., Bie, T.D. (eds.) Proceedings of the 4th Workshop on Recommender Systems for ...

2024
[30]

CoRR abs/2408.00118 (2024)

Rivière, M., Pathak, S., Sessa, P .G., Hardin, C., Bhupatiraju, S., Hussenot, L., Mesnard, T., Shahriari, B., Ramé, A., Ferret, J., Liu, P ., Tafti, P ., Friesen, A., Casbon, M., Ramos, S., Kumar, R., Lan, C.L., Jerome, S., Tsitsulin, A., Vieil- lard, N., Stanczyk, P ., Girgin...

2024 doi arXiv
[31]

In: CLEF (Working Notes) (2025)

Rodríguez, M., Perez-de Viñaspre, O., Perez, N.: A Two-Stage Multilingual Job Title Matching System: Combining Expert Knowledge and LLM-based Ranking. In: CLEF (Working Notes) (2025)

2025
[32]

In: CLEF (Working Notes) (2025)

Rodríguez-Vidal, J., López-Vargas, A., Vigara Gallego, P .M., Del Álamo, F .J., García-Beltrán, A.: UDII-UPM at TalentCLEF 2025: Task A-Multilingual Job Ti- tle Matching. In: CLEF (Working Notes) (2025)

2025
[33]

Association for Computational Linguistics

Senger, E., Zhang, M., van der Goot, R., Plank, B.: Deep learning-based computa- tional job market analysis: A survey on skill extraction and classification from job postings. In: Proceedings of the First Workshop on Natural Language Processing for Human Resources (NLP4HR 2024...

2024
[34]

Smit, Sven and Tacke, Tilman and Lund, Susan and Manyika, James and Thiel, Lea: The future of work in Europe: Automation, workforce transitions, and the shifting geography of employment. Tech. rep., McKinsey & Company (2020)

2020
[35]

CoRR abs/2402.16829 (2024)

Solatorio, A.V .: Gistembed: Guided in-sample selection of training negatives for text embedding fine-tuning. CoRR abs/2402.16829 (2024). https://doi.org/10.48550/ARXIV .2402.16829, https://doi.org/10.48550/arXiv.2402.16829

2024 doi arXiv
[36]

International Journal of Computer Vision 130(6), 1526–1565 (2022)

Soviany, P ., Ionescu, R.T., Rota, P ., Sebe, N.: Curriculum learning: A sur- vey. International Journal of Computer Vision 130(6), 1526–1565 (2022). https://doi.org/10.1007/S11263-022-01611-X, https://doi.org/10.1007/s11263- 022-01611-x

2022 doi
[37]

CoRR abs/2411.05934 (2024)

Tahmid, S., Sarker, S.: Qwen2.5-32b: Leveraging self-consistent tool- integrated reasoning for bengali mathematical olympiad problem solving. CoRR abs/2411.05934 (2024). https://doi.org/10.48550/ARXIV .2411.05934, https://doi.org/10.48550/arXiv.2411.05934

2024 doi arXiv
[38]

In: CLEF (Working Notes) (2025)

Tejera Villar, A., Segura Bedmar, I.: HULAT-UC3M at TalentCLEF: Artificial Intel- ligence and Natural Language Processing applied to HR Management. In: CLEF (Working Notes) (2025)

2025
[39]

In: CLEF (Working Notes) (2025)

Uddin, A., Nizami, M.H., Salani, M.T., Saeed, A.: Enhancing Human Capital Man- agement: AI Techniques for Candidate Matching and Skill Extraction. In: CLEF (Working Notes) (2025)

2025
[40]

In: CLEF (Working Notes) (2025)

Vachharajani, P .: pjmathematician at TalentCLEF 2025: Enhancing Job Title and Skill Matching with GISTEmbed and LLM-Augmented Data. In: CLEF (Working Notes) (2025)

2025
[41]

In: CLEF (Working Notes) (2025)

Vázquez García, I.X., Sedano Puente, R., González González, S., Sedano Franco, J.: Beyond Titles: Semantic Matching of Jobs and Skills Using LLMs and S-BERT. In: CLEF (Working Notes) (2025)

2025
[42]

CoRR abs/2402.05672 (2024)

Wang, L., Yang, N., Huang, X., Yang, L., Majumder, R., Wei, F .: Multilingual E5 text embeddings: A technical report. CoRR abs/2402.05672 (2024). https://doi.org/10.48550/ARXIV .2402.05672, https://doi.org/10.48550/arXiv.2402.05672

2024 doi arXiv
[43]

ACM Transactions on Information Systems (2010)

Webber, W ., Moffat, A., Zobel, J.: A similarity measure for indefinite rankings. ACM Transactions on Information Systems (2010)

2010
[44]

CoRR abs/2207.00494 (2022)

Zbib, R., Alvarez, L.L., Retyk, F ., Poves, R., Aizpuru, J., Fabregat, H., Simkus, V ., Casademont, E.G.: Learning job titles similarity from noisy skill labels. CoRR abs/2207.00494 (2022). https://doi.org/10.48550/ARXIV .2207.00494, https://doi.org/10.48550/arXiv.2207.00494

2022 doi arXiv
[45]

Contrastive vs

Zhang, M., van der Goot, R.: NLPnorth @ TalentCLEF 2025: Discriminative vs. Contrastive vs. Prompting for Job Title and Job-Skill Matching. In: CLEF (Work- ing Notes) (2025)

2025
[46]

In: Rogers, A., Boyd-Graber, J.L., Okazaki, N

Zhang, M., van der Goot, R., Plank, B.: ESCOXLM-R: multilingual taxonomy- driven pre-training for the job market domain. In: Rogers, A., Boyd-Graber, J.L., Okazaki, N. (eds.) Proceedings of the 61st Annual Meeting of the As- sociation for Computational Linguistics (Volume 1: L...

2023 doi
[47]

In: Kaya, M., Bogers, T., Graus, D., Mesbah, S., Johnson, C., Gutiérrez, F

Zhang, M., Jensen, K.N., van der Goot, R., Plank, B.: Skill extraction from job postings using weak supervision. In: Kaya, M., Bogers, T., Graus, D., Mesbah, S., Johnson, C., Gutiérrez, F . (eds.) Proceedings of the 2nd Workshop on Recom- mender Systems for Human Resources (Re...

2022
[48]

In: Carpuat, M., de Marn- effe, M., Ruíz, I.V .M

Zhang, M., Jensen, K.N., Sonniks, S.D., Plank, B.: Skillspan: Hard and soft skill extraction from english job postings. In: Carpuat, M., de Marn- effe, M., Ruíz, I.V .M. (eds.) Proceedings of the 2022 Conference of the North American Chapter of the Association for Computationa...

2022 doi
[49]

In: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

Zhu, H., Ge, Y., Xiong, H., Lim, E.P .: The 5th international work- shop on talent and management computing (tmc’2024). In: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. p. 6759–6760. KDD ’24, Association for Computing Machin- ery, New Y...

2024
[2023]

https://www

In: World Economic Forum, Geneva, Switzerland. https://www. weforum. org/reports/the-future-of-jobs-report-2023. World Economic Forum (2023)

2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

In: CLEF (Working Notes) (2025)

Ali, M.: Enhancing Job-Skill Matching with LLM-Driven Data Augmentation and Fine-Tuned Bi-Encoders. In: CLEF (Working Notes) (2025)

work page 2025

[2] [2]

In: CLEF (Working Notes) (2025)

Barakat, R., Mokhtar, O., Torki, M., Elmakky, N.: AlexU-NLP at TalentCLEF 2025: Curriculum-Driven Hybrid Retrieval for Multilingual Job Title Matching. In: CLEF (Working Notes) (2025)

work page 2025

[3] [3]

In: CLEF (Working Notes) (2025)

Bhangale, C., Gabhane, P ., Kumar M, A.: Fine-Tuned Sentence Transformer for Multilingual Job Title Matching. In: CLEF (Working Notes) (2025)

work page 2025

[4] [4]

In: Proceedings of the 18th ACM Conference on Recommender Systems

Bogers, T., Graus, D., Kaya, M., Johnson, C., Decorte, J.J., De Bie, T.: Fourth workshop on recommender systems for human resources (recsys in hr 2024). In: Proceedings of the 18th ACM Conference on Recommender Systems. p. 1222–1226. RecSys ’24, Association for Computing Machin- ery, New York, NY, USA (2024). https://doi.org/10.1145/3640457.3687109, https...

work page arXiv 2024

[5] [5]

In: CLEF (Working Notes) (2025)

Brikman, A., Sana, M., Ruegger, H.: Multilingual Job Title Matching with MPNet- Based Sentence Transformers. In: CLEF (Working Notes) (2025)

work page 2025

[6] [6]

In: Ku, L., Martins, A., Srikumar, V

Chen, J., Xiao, S., Zhang, P ., Luo, K., Lian, D., Liu, Z.: M3-embedding: Multi- linguality, multi-functionality, multi-granularity text embeddings through self- knowledge distillation. In: Ku, L., Martins, A., Srikumar, V . (eds.) Findings of the Association for Computational Linguistics, ACL 2024, Bangkok, Thailand and virtual meeting, August 11-16, 202...

doi:10.18653/v1/2024.findings-acl.137 2024

[7] [7]

CoRR abs/2412.03223 (2024)

Choi, C., Kim, J., Lee, S., Kwon, J., Gu, S., Kim, Y., Cho, M., Sohn, J.: Linq-embed-mistral technical report. CoRR abs/2412.03223 (2024). https://doi.org/10.48550/ARXIV .2412.03223, https://doi.org/10.48550/arXiv.2412.03223

arXiv doi:10.48550/arxiv.2412.03223 2024

[8] [8]

In: CLEF (Working Notes) (2025)

Decorte, J.J., De Lange, M., Van Hautte, J.: TechWolf at TalentCLEF 2025: Mul- tilingual JobBERT-V2 for Cross-Lingual Job Title Matching. In: CLEF (Working Notes) (2025)

work page 2025