arxiv: 2605.05280 · v1 · submitted 2026-05-06 · 💻 cs.LG

Recognition: unknown

Forecasting Green Skill Demand in the Automotive Industry: Evidence from Online Job Postings

Hector G. Ceballos, Joshua N. Arrazola E., Patricia Caratozzolo, Sabur Butt

Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:06 UTC · model grok-4.3

classification 💻 cs.LG

keywords green skillsjob postingstime series forecastingautomotive industrytransformer modelsESCO taxonomyMexico labor marketsustainable transition

0 comments

The pith

A pipeline from Mexican automotive job postings uses transformer models to forecast rising green skill demand.

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

The paper develops a two-stage method that extracts green skills from online job advertisements in Mexico's automotive industry and then applies time-series forecasting to predict their future demand. It processes over 200,000 skill records to isolate 274 green skills and benchmarks fifteen models, finding that transformer architectures deliver the lowest errors. A growth classification system further separates skills into stable, emerging, and high-impact groups. If the approach holds, workforce planners could use these forecasts to anticipate training needs as the economy shifts toward sustainability. The work matters because it turns publicly available postings into actionable signals for the green transition.

Core claim

The authors compile job postings from three platforms covering July 2024 to July 2025, apply multilingual embeddings plus ESCO validation to label 274 unique green skills, and show through rolling-origin evaluation that FEDformer, Reformer, and Informer achieve the lowest forecasting errors while current demand clusters around operational sustainability and future growth centers on renewable energy, recycling, and hydrogen technologies.

What carries the argument

Two-stage pipeline that first matches skill mentions via multilingual embeddings and ESCO taxonomy then feeds the resulting time series into transformer-based forecasters evaluated under rolling origin validation.

If this is right

Demand today concentrates in operational sustainability practices.
Fastest-growing skills center on renewable energy, recycling, and hydrogen technologies.
FEDformer, Reformer, and Informer produce the most accurate forecasts with MAE near 2.5e-5 and relative RMSE below 15.
The absolute-and-relative growth classification distinguishes stable, emerging, and high-impact competencies for workforce planning.

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 applied to other manufacturing sectors to map green-skill trajectories beyond automotive.
Policymakers could use the emerging-skill list to prioritize curriculum updates in technical training programs.
Periodic re-runs on fresh postings would allow ongoing monitoring of whether the green transition accelerates demand as projected.

Load-bearing premise

Online job postings from the three platforms accurately reflect true skill demand in the automotive industry and the embedding-plus-ESCO step identifies green skills with acceptably low error rates.

What would settle it

A direct survey or audit of Mexican automotive employers that reports substantially different frequencies or identities of green skills than those extracted from the 204,373 records in the dataset.

Figures

Figures reproduced from arXiv: 2605.05280 by Hector G. Ceballos, Joshua N. Arrazola E., Patricia Caratozzolo, Sabur Butt.

**Figure 1.** Figure 1: Example of a skill extraction process from a job description using GPT-4o. view at source ↗

**Figure 2.** Figure 2: Example of a semantic query executed in Spanish and matched against the English ESCO green skill view at source ↗

**Figure 3.** Figure 3: Most frequent green skills detected in the Mexican automotive industry. view at source ↗

**Figure 4.** Figure 4: Rolling Origin evaluation scheme with two consecutive sliding windows. Each window uses the most recent view at source ↗

**Figure 5.** Figure 5: Skill classification based on absolute and relative growth. view at source ↗

read the original abstract

The global transition toward sustainable economies is reshaping labor markets, yet systematic methods for identifying and forecasting green skills remain limited. This study presents a computational framework to measure and predict green skill demand using online job postings from Mexico's automotive industry, which contributes about 4% of national GDP. We compile a dataset of job advertisements from Indeed Mexico, OCC Mundial, and LinkedIn (July 2024 to July 2025), yielding 204,373 skill records. A two-stage pipeline combining multilingual embeddings and ESCO validation identifies 274 unique green skills across 8,576 occurrences (4.22% of all skills). We benchmark 15 time series forecasting models using a rolling origin evaluation. Transformer-based models, especially FEDformer, Reformer, and Informer, achieve the best performance, with MAE around 2.5e-5 and relative RMSE below 15. We further propose a framework to classify skills by absolute and relative growth, identifying stable, emerging, and high-impact competencies. Results show current demand is concentrated in operational sustainability practices, while the fastest-growing skills relate to renewable energy, recycling, and hydrogen technologies. This pipeline supports data-driven workforce planning in the green transition.

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.

Desk Editor's Note private letter to a colleague

The paper applies standard NLP and time-series tools to a new Mexican automotive job-ad dataset to track and forecast green skills, but the unvalidated labeling step leaves the results on shaky ground.

read the letter

The main thing to know is that this work pulls 204k skill mentions from Indeed, OCC, and LinkedIn postings in Mexico's auto sector over one year, uses multilingual embeddings plus ESCO to flag 274 green skills, runs fifteen forecasters with rolling-origin evaluation, and then bins the skills into stable, emerging, and high-impact groups based on growth. Transformers such as FEDformer come out ahead on the normalized series, and the results point to operational sustainability dominating current demand while renewables, recycling, and hydrogen skills grow fastest. That concrete application to an emerging-economy industry is the clearest addition to the literature. The pipeline is laid out plainly enough that someone could replicate the steps on similar data. The classification framework also gives a simple way to summarize trends for policy use. The soft spot is the labeling step. The abstract and stress-test note give no precision, recall, or human validation numbers for the embedding-plus-ESCO match, yet every downstream count, forecast, and growth label depends on it. With roughly 31 occurrences per skill on average, even moderate false positives or negatives would shift the time series and the rankings. The one-year window is also short for transformer-based forecasting, and the extremely low MAE figures suggest heavy normalization whose details matter for interpretation. Online postings are a reasonable proxy but still miss offline hiring and may over-represent certain firms. This is useful reading for labor-market analysts or green-transition policymakers who want an example of applied skill forecasting in Latin America. A reader already working on job-ad NLP would pick up the setup quickly and see where the numbers come from. It is worth sending to peer review because the data source and sector focus are timely and the method is straightforward, provided the authors add a validation check on the green-skill labels and clarify the scaling used for the error metrics.

Referee Report

3 major / 2 minor

Summary. The manuscript presents a computational framework for identifying and forecasting green skill demand in Mexico's automotive industry using online job postings from Indeed, OCC Mundial, and LinkedIn collected between July 2024 and July 2025. From 204,373 skill records, a two-stage pipeline with multilingual embeddings and ESCO validation extracts 274 green skills appearing 8,576 times. The authors benchmark 15 time series forecasting models via rolling-origin evaluation, reporting that transformer-based models (FEDformer, Reformer, Informer) perform best with MAE around 2.5e-5 and relative RMSE below 15. They also propose a classification of skills based on absolute and relative growth rates into stable, emerging, and high-impact categories, concluding that demand is concentrated in operational sustainability while growth is in renewable energy, recycling, and hydrogen technologies.

Significance. If the green-skill identification step proves reliable, the work offers a practical, data-driven approach to tracking and predicting labor market needs during the green transition, with particular relevance to Mexico's automotive sector. The use of a large real-world dataset, systematic benchmarking of multiple forecasting models including recent transformers, and the rolling-origin evaluation protocol are strengths that enhance reproducibility and applicability. The proposed growth-based classification framework could aid workforce planning if the underlying time series are accurate.

major comments (3)

The central performance claims (MAE around 2.5e-5 and relative RMSE below 15 for transformer models) and all growth classifications rest on the two-stage green-skill identification pipeline (multilingual embeddings + ESCO validation) described in the abstract, yet no precision, recall, inter-annotator agreement, or hold-out validation metrics are reported for this step despite its load-bearing role for the 274 skills and 8,576-occurrence time series.
The abstract notes an average of ~31 occurrences per green skill; the manuscript provides no analysis of how sparsity, missing values, or potential labeling noise (false positives from generic terms or false negatives for context-specific skills) propagates into the rolling-origin forecasts or the absolute/relative growth cutoffs used for stable/emerging/high-impact classification.
The embedding similarity threshold is listed as a free parameter in the pipeline, but the abstract and evaluation description report no sensitivity analysis or ablation on this threshold, leaving the reported superiority of FEDformer, Reformer, and Informer vulnerable to changes in the green-skill labeling.

minor comments (2)

The dataset period (July 2024 to July 2025) includes future dates; clarify whether the postings are historical, projected, or collected in real time to avoid confusion about data availability.
Reporting 'relative RMSE below 15' without a table of exact values, baselines, or per-model breakdowns for all 15 models reduces transparency; a results table would strengthen the benchmarking claims.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback, which underscores the importance of rigorously validating the green-skill identification pipeline that underpins our forecasting and classification results. We address each major comment below and will revise the manuscript accordingly to strengthen its methodological transparency and robustness.

read point-by-point responses

Referee: The central performance claims (MAE around 2.5e-5 and relative RMSE below 15 for transformer models) and all growth classifications rest on the two-stage green-skill identification pipeline (multilingual embeddings + ESCO validation) described in the abstract, yet no precision, recall, inter-annotator agreement, or hold-out validation metrics are reported for this step despite its load-bearing role for the 274 skills and 8,576-occurrence time series.

Authors: We agree that quantitative validation of the identification pipeline is essential given its central role. The original submission omitted these metrics. In the revised manuscript we will report precision, recall, and F1 scores obtained from a manually annotated hold-out sample of 500 skill mentions, together with inter-annotator agreement (Cohen's kappa) between two domain experts who reviewed the ESCO-validated green skills. These additions will directly support the reliability of the 274 skills and the downstream time series. revision: yes
Referee: The abstract notes an average of ~31 occurrences per green skill; the manuscript provides no analysis of how sparsity, missing values, or potential labeling noise (false positives from generic terms or false negatives for context-specific skills) propagates into the rolling-origin forecasts or the absolute/relative growth cutoffs used for stable/emerging/high-impact classification.

Authors: We acknowledge the absence of explicit propagation analysis. The time series consist of monthly aggregated counts, and the rolling-origin protocol already evaluates temporal robustness; however, we did not quantify how labeling noise or sparsity affects growth-rate cutoffs or forecast errors. In the revision we will add a dedicated subsection that (i) characterizes the sparsity distribution across the 274 series and (ii) presents a sensitivity experiment in which we inject controlled false-positive and false-negative perturbations and recompute both the stable/emerging/high-impact classifications and the MAE/RMSE of the top models. This will clarify the robustness of our conclusions. revision: yes
Referee: The embedding similarity threshold is listed as a free parameter in the pipeline, but the abstract and evaluation description report no sensitivity analysis or ablation on this threshold, leaving the reported superiority of FEDformer, Reformer, and Informer vulnerable to changes in the green-skill labeling.

Authors: The threshold (0.75) was selected after preliminary manual inspection to balance coverage and precision, yet no systematic ablation was reported. We will include a new ablation table in the revised manuscript that varies the threshold from 0.60 to 0.90, reporting for each value the resulting number of green skills, total occurrences, and the MAE and relative RMSE of FEDformer, Reformer, and Informer under the same rolling-origin protocol. This will demonstrate that the relative ranking of the transformer models remains stable across reasonable threshold choices. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmarking via rolling-origin evaluation and data preprocessing

full rationale

The paper reports performance of standard time-series models (FEDformer, Reformer, Informer, etc.) evaluated with rolling-origin hold-out on future periods of the extracted skill-occurrence series. This is a conventional non-circular protocol that does not reduce reported MAE or relative RMSE to quantities fitted on the same data used for the headline claim. The two-stage green-skill pipeline (multilingual embeddings + ESCO validation) is presented as preprocessing that produces the input time series; no derivation or uniqueness theorem is claimed for it. The proposed growth-classification framework simply computes absolute and relative growth rates from the observed counts and assigns labels; it contains no self-referential fitting, ansatz smuggled via citation, or self-citation load-bearing step. No equations, self-citations, or prior-author uniqueness results appear in the provided text that would collapse any central result back to its own inputs by construction. The work is therefore self-contained empirical analysis against external data splits.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The work rests on standard labor-market and NLP assumptions rather than new postulates; limited abstract information prevents exhaustive enumeration of fitted hyperparameters.

free parameters (2)

Embedding similarity threshold for green-skill matching
Implicit cutoff used to decide which job-skill phrases count as green after ESCO validation.
Growth-rate cutoffs for stable/emerging/high-impact classification
Hand-chosen boundaries that determine which skills are labeled emerging or high-impact.

axioms (2)

domain assumption Online job postings from the sampled platforms are an unbiased proxy for actual skill demand in the Mexican automotive industry.
Core premise enabling both identification and forecasting steps.
domain assumption The ESCO taxonomy plus multilingual embeddings produces a sufficiently accurate green-skill label without systematic misclassification.
Invoked when the two-stage pipeline is described.

pith-pipeline@v0.9.0 · 5526 in / 1522 out tokens · 74299 ms · 2026-05-08T17:06:20.591120+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

67 extracted references · 5 canonical work pages · 1 internal anchor

[1]

Daron Acemoglu, David Autor, Jonathon Hazell, and Pascual Restrepo. 2022. Artificial intelligence and jobs: Evidence from online vacancies.Journal of Labor Economics40, S1 (2022), S293–S340

2022
[2]

Annabell Albertz and Matthias Pilz. 2025. Green alignment, green vocational education and training, green skills and related subjects: A literature review on actors, contents and regional contexts.International Journal of Training and Development29, 2 (2025), 243–254

2025
[3]

Apify Technologies s.r.o. 2025. Apify: Web Scraping and Automation Platform.https: //apify.com/. Accessed: 25-Oct-2025

2025
[4]

Esther Arenas Arroyo, Jacob Fabian, Friederike Mengel, Bernhard Schmidpeter, and Michel Serafinelli. 2025. Demand for Green Skills in an Evolving Landscape. (2025)

2025
[5]

Ed Atkins. 2024. Green jobs, just transition, and the future of work. InThe Handbook for the Future of Work. Routledge, 375–387

2024
[6]

Sabrina Aufiero, Giordano De Marzo, Angelica Sbardella, and Andrea Zaccaria. 2024. Map- ping job fitness and skill coherence into wages: an economic complexity analysis.Scientific Reports14, 1 (2024), 11752

2024
[7]

Georgeta Vidican Auktor. 2020. Green industrial skills for a sustainable future.United Nations Industrial Development Organization, Vienna(2020)

2020
[8]

Jose Daniel Azofeifa, Valentina Rueda-Castro, Luis Jose Gonzalez-Gomez, Sabur Butt, Juli- eta Noguez, Hector G Ceballos, and Patricia Caratozzolo. 2025. Insights from a dynamic ksa taxonomy framework: Top 10 wanted knowledge, skills, and abilities for the infocomm sector in mexico. In2025 Institute for the Future of Education Conference (IFE). IEEE, 1–11. 29

2025
[9]

Angélica Basulto Castillo. 2023. The automotive industry in Western Mexico and the per- formance of local suppliers in the value chains of multinational companies. InGlobal Value Chains in Latin America: Experiences of Transformations. Springer, 287–324

2023
[10]

Sabur Butt, Gustavo De los Ríos Alatorre, Luis José González Gómez, and Hector G Cebal- los. 2025. Semi-Supervised Taxonomy Expansion and Completion in Dynamic Taxonomies. Applied Sciences15, 12 (2025), 6517

2025
[11]

Sabur Butt, Hector G Ceballos, and Diana P Madera-Espíndola. 2025. Tec-Habilidad: Skill Classification for Bridging Education and Employment. InMexican International Confer- ence on Artificial Intelligence. Springer, 313–328

2025
[12]

Sabur Butt, Héctor Gibrán Ceballos Cancino, and Michael Fung. [n. d.].Shaping skills for the Mexican workforce: Data-based insights into the Mexican automotive industry.Technical Report
[13]

Patricia Caratozzolo, Uriel Cukierman, Bente Nørgaard, Katriina Schrey-Niemenmaa, Jose Daniel Azofeifa, and Valentina Rueda-Castro. 2024. Future skills forecasting: Ensur- ing quality learning for every segment of the workforce. In2024 IEEE Global Engineering Education Conference (EDUCON). IEEE, 1–5

2024
[14]

Wenshuo Chao, Zhaopeng Qiu, Likang Wu, Zhuoning Guo, Zhi Zheng, Hengshu Zhu, and Hao Liu. 2024. A cross-view hierarchical graph learning hypernetwork for skill demand- supply joint prediction. InProceedings of the AAAI Conference on Artificial Intelligence, V ol. 38. 19813–19822

2024
[15]

Xi Chen, Chuan Qin, Chuyu Fang, Chao Wang, Chen Zhu, Fuzhen Zhuang, Hengshu Zhu, and Hui Xiong. 2024. Job-SDF: A Multi-Granularity Dataset for Job Skill Demand Fore- casting and Benchmarking. arXiv:2406.11920 [cs.LG]https://arxiv.org/abs/2406. 11920

work page arXiv 2024
[16]

Xi Chen, Chuan Qin, Chuyu Fang, Chao Wang, Fuzhen Zhuang, Hengshu Zhu, Hui Xiong, et al. 2024. Job-SDF: A multi-granularity dataset for job skill demand forecasting and bench- marking.Advances in Neural Information Processing Systems37 (2024), 129329–129356

2024
[17]

Davide Consoli, Giovanni Marin, Alberto Marzucchi, and Francesco V ona. 2016. Do green jobs differ from non-green jobs in terms of skills and human capital?Research Policy45, 5 (2016), 1046–1060

2016
[18]

E Mark Curtis, Layla O’Kane, and R Jisung Park. 2024. Workers and the green-energy transition: Evidence from 300 million job transitions.Environmental and Energy Policy and the Economy5, 1 (2024), 127–161

2024
[19]

Tamíris da Costa, Lorena Irazu Aranda Lopez, Camila Perussello, Fiona Quinn, Quentin G Crowley, Helena McMahon, and Nicholas M Holden. 2025. Addressing the demand for green skills: bridging the gap between university outcomes and industry requirements.Sus- tainability17, 6 (2025), 2732. 30

2025
[20]

Simone D’Amico, Lorenzo Malandri, Fabio Mercorio, Mario Mezzanzanica, and Filippo Pallucchini. 2024. Alignment of multilingual embeddings to estimate job similarities in online labour market. In2024 IEEE 11th International Conference on Data Science and Advanced Analytics (DSAA). IEEE, 1–10

2024
[21]

Subhro Das, Sebastian Steffen, Wyatt Clarke, Prabhat Reddy, Erik Brynjolfsson, and Mar- tin Fleming. 2020. Learning occupational task-shares dynamics for the future of work. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society. 36–42

2020
[22]

De La Garza, Ismael and Banker, Global . 2024. Automotive Sector: Relevance and Main Sustainability Challenges in Mexico.https://tinyurl.com/2z49h3h3

2024
[23]

Johan De Smedt, Martin le Vrang, and Agis Papantoniou. 2015. ESCO: Towards a Semantic Web for the European Labor Market.. InLdow@ www

2015
[24]

David J Deming and Kadeem Noray. 2020. Earnings dynamics, changing job skills, and STEM careers.The Quarterly Journal of Economics135, 4 (2020), 1965–2005

2020
[25]

Johannes Emmerling, Laurent Drouet, Jessica Jewell, Lola Nacke, Sandeep Pai, and Hisham Zerriffi. 2025. Green jobs and just transition: Employment implications of Europe’s Net Zero pathway.Energy Research&Social Science127 (2025), 104292

2025
[26]

2022.Green Skills and Knowledge Concepts: Labelling the ESCO classification

European Commission. 2022.Green Skills and Knowledge Concepts: Labelling the ESCO classification. Technical Report. Publications Office of the European Com- mission.https://esco.ec.europa.eu/system/files/2025-01/Green%20Skills% 20and%20Knowledge%20-%20Labelling%20ESCO.pdf

2022
[27]

European Commission. 2024. The ESCO Classification.https://esco.ec.europa.eu/ en/classification. Accessed: 25-Jan-2026

2024
[28]

European Commission. 2025. ESCO - European Skills, Competences, Qualifications and Occupations Classification.https://esco.ec.europa.eu/en/classification/skill_ main. Accessed: 25-Oct-2025

2025
[29]

2022.Methodological issues related to the use of online labour market data

Brian Fabo, Lucia M Kureková, et al. 2022.Methodological issues related to the use of online labour market data. Number 68. ILO Working Paper.https://doi.org/10.54394/ ZZBC8484

2022
[30]

Martina Fuchs. 2024. Green skills for sustainability transitions.Geography Compass18, 10 (2024), e70003

2024
[31]

Maysa Malfiza Garcia de Macedo, Wyatt Clarke, Eli Lucherini, Tyler Baldwin, Dilermando Queiroz Neto, Rogerio Abreu de Paula, and Subhro Das. 2022. Practical skills demand forecasting via representation learning of temporal dynamics. InProceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society. 285–294. 31

2022
[32]

2024.Navigating the Green Transi- tion: Building Green Skills for a Sustainable Workforce: Skills4Dev

Daniel Garrote Sanchez and Monica Yanez Pagans. 2024.Navigating the Green Transi- tion: Building Green Skills for a Sustainable Workforce: Skills4Dev. Skills4Dev Issue 13. World Bank Group, Washington, D.C.http://documents.worldbank.org/curated/ en/099806510302420882English

work page arXiv 2024
[33]

Jerry Haar and Eric Porras. 2026. Mexico’s Role in Transforming the North American Au- tomotive Industry.Technology in Society(2026), 103282

2026
[34]

Joshua Healy, Kostas Mavromaras, and Peter J Sloane. 2015. Adjusting to skill shortages in Australian SMEs.Applied Economics47, 24 (2015), 2470–2487

2015
[35]

Z Ibrahim, CS Lai, AF Zaime, MF Lee, and NM Othman. 2020. Green skills in knowledge and attitude dimensions from the industrial perspective. InIOP conference series: Materials science and engineering, V ol. 917. IOP Publishing, 012025

2020
[36]

Indeed Inc. 2025. Indeed México: Portal de Empleo.https://www.indeed.com.mx/. Accessed: 25-Oct-2025

2025
[37]

Florence Jaumotte, Kim Jadem, David Koll, Elmer Li, Longji Li, Giovanni Melina, Alina Song, and Marina M Tavares. 2026. Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age. (2026).https://doi.org/10.5089/9798229028196.006

work page doi:10.5089/9798229028196.006 2026
[38]

Hamit Kavas, Marc Serra-Vidal, and Leo Wanner. 2025. Multilingual skill extraction for job vacancy–job seeker matching in knowledge graphs. InProceedings of the Workshop on Generative AI and Knowledge Graphs (GenAIK). 146–155

2025
[39]

Mikkel Knudsen, Marjolein Caniëls, Peter Dickinson, Michel Hery, Totti Könnölä, and Heila Lotz-Sisitka. 2023. Futures of green skills and jobs in Europe in 2050: scenario and policy implications. (2023)

2023
[40]

Jonathan D Levine and Frederick L Oswald. 2013. O* NET: The occupational information network. InThe Handbook of Work Analysis. Routledge, 281–301

2013
[41]

Shengsheng Lin, Weiwei Lin, Wentai Wu, Feiyu Zhao, Ruichao Mo, and Haotong Zhang
[42]

Segrnn: Segment recurrent neural network for long-term time series forecasting.IEEE Internet of Things Journal(2025)

2025
[43]

LinkedIn Corporation. 2025. LinkedIn México: Ofertas de Empleo.https://mx. linkedin.com/jobs. Accessed: 25-Oct-2025

2025
[44]

Yong Liu, Haixu Wu, Jianmin Wang, and Mingsheng Long. 2022. Non-stationary transform- ers: Exploring the stationarity in time series forecasting.Advances in neural information processing systems35 (2022), 9881–9893

2022
[45]

Zhaolu Liu, Jonathan M Clarke, Bertha Rohenkohl, and Mauricio Barahona. 2025. Patterns of co-occurrent skills in UK job adverts.PLOS Complex Systems2, 2 (2025), e0000028. 32

2025
[46]

Elena Magrini, Julia Nania, Mauro Pelucchi, Rudra Sett, Will Hanvey, Layla O’Kane, Dun- can Brown, and Bledi Taska. 2023. The green transition in labour markets across europe: an analysis of job postings data. InPathways of Greening Labour Markets. Nomos Verlagsge- sellschaft mbH & Co. KG, 321–346

2023
[47]

Jorge Membrillo-Hernández, Vianney Lara-Prieto, and Patricia Caratozzolo. 2021. Sustain- ability: A public policy, a concept, or a competence? efforts on the implementation of sus- tainability as a transversal competence throughout higher education programs.sustainability 13, 24 (2021), 13989

2021
[48]

Sibilla Montanari, Evi Agostini, and Denis Francesconi. 2023. Are we talking about green skills or sustainability competences? A scoping review using scientometric analysis of two apparently similar topics in the field of sustainability.Sustainability15, 19 (2023), 14142

2023
[49]

Adam Nelson-Archer, Aleia Sen, Meena Al Hasani, Sofia Davila, Jessica Le, and Omar Abbouchi. 2025. Forecasting Labor Markets with LSTNet: A Multi-Scale Deep Learning Approach.arXiv preprint arXiv:2507.01979(2025)

work page arXiv 2025
[50]

Khanh Nguyen, Mike Zhang, Syrielle Montariol, and Antoine Bosselut. 2024. Rethinking skill extraction in the job market domain using large language models. InProceedings of the First Workshop on Natural Language Processing for Human Resources (NLP4HR 2024). 27–42

2024
[51]

Y Nie. 2022. A Time Series is Worth 64Words: Long-term Forecasting with Transformers. arXiv preprint arXiv:2211.14730(2022)

work page internal anchor Pith review arXiv 2022
[52]

OCCMundial S.A. de C.V . 2025. OCC Mundial: Bolsa de Trabajo en México.https: //www.occ.com.mx/. Accessed: 25-Oct-2025

2025
[53]

Ionica Oncioiu, Mariana Man, Marius Florin Ghiberdic, and Mihaela Hortensia Hojda. 2025. Forecasting the Energy-Driven Green Transition of European Labour Markets: A Composite Readiness Index.Energies19, 1 (2025), 114

2025
[54]

OpenAI. [n. d.]. Text Embedding 3 Large.https://platform.openai.com/docs/ models/text-embedding-3-large. Consultado: 2025-10-13

2025
[55]

OECD Skills Outlook. 2023. Skills for a resilient green and digital transition.URL: https://www. oecd-ilibrary. org/sites/27452f29-en/index. html(2023)

2023
[56]

Konstantinos Pouliakas. 2021. Understanding Technological Change and Skill Needs: Tech- nology and Skills Foresight. Cedefop Practical Guide 3.Cedefop-European Centre for the Development of Vocational Training(2021)

2021
[57]

William Rasdorf, Joseph E Hummer, and Stephanie C Vereen. 2016. Data collection op- portunities and challenges for skilled construction labor demand forecast modeling.Public Works Management&Policy21, 1 (2016), 28–52. 33

2016
[58]

2020.Estimating the green potential of occupations: A new approach applied to the US labor market

Christian Rutzer, Matthias Niggli, and Rolf Weder. 2020.Estimating the green potential of occupations: A new approach applied to the US labor market. Technical Report. WWZ Working Paper

2020
[59]

Alessia Scoz and Arthur Biamouret. 2024. Mapping green and digital energy jobs: trends and insights from online job posting. InMapping green and digital energy jobs: trends and insights from online job posting: Scoz, Alessia|uBiamouret, Arthur. [Paris]: International Energy Agency

2024
[60]

Anne Marie Thake. 2025. Transitioning to a green economy-the impact on the labor market and workforce skills. InGreening our economy for a sustainable future. Elsevier, 163–175

2025
[61]

THUML. 2023. Time-Series-Library (TSLib): A Library for Advanced Deep Time Series Models for General Time Series Analysis.https://github.com/thuml/ Time-Series-Library. Accessed: 25-Oct-2025

2023
[62]

2025.Green jobs: A critique of the occupational approach to measure the employment implications of the green transition

Davide Villani, Ignacio González Vázquez, and Enrique Fernández-Macías. 2025.Green jobs: A critique of the occupational approach to measure the employment implications of the green transition. Technical Report. JRC Working Papers Series on Labour, Education and Technology.https://hdl.handle.net/10419/322064

2025
[63]

Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long. 2021. Autoformer: Decomposi- tion transformers with auto-correlation for long-term series forecasting.Advances in neural information processing systems34 (2021), 22419–22430

2021
[64]

Felix Zaussinger, Tobias S Schmidt, and Florian Egli. 2025. Skills-based and regionally explicit labor market exposure to the low-carbon transition in Europe.Joule9, 2 (2025)

2025
[65]

Ailing Zeng, Muxi Chen, Lei Zhang, and Qiang Xu. 2023. Are transformers effective for time series forecasting?. InProceedings of the AAAI conference on artificial intelligence, V ol. 37. 11121–11128

2023
[66]

Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang. 2021. Informer: Beyond efficient transformer for long sequence time-series forecast- ing. InProceedings of the AAAI conference on artificial intelligence, V ol. 35. 11106–11115

2021
[67]

Not mentioned

Tian Zhou, Ziqing Ma, Qingsong Wen, Xue Wang, Liang Sun, and Rong Jin. 2022. Fed- former: Frequency enhanced decomposed transformer for long-term series forecasting. In International conference on machine learning. PMLR, 27268–27286. 34 Appendix A. Prompt Design System:You are an information extraction system specialized in identifyingskillswithin job des...

2022