pith. sign in

arxiv: 2605.24155 · v1 · pith:2EBBWEAWnew · submitted 2026-05-22 · 💻 cs.IR · cs.AI· cs.LG

An Interpretable CF-RL-TOPSIS Fusion Model for Skills-Aware Talent Recommendation

\"Ozkan Canay This is my paper

Pith reviewed 2026-06-30 14:17 UTC · model grok-4.3

classification 💻 cs.IR cs.AIcs.LG
keywords talent recommendationcollaborative filteringreinforcement learningTOPSISlate fusionskills-aware recommendationinterpretable modelsNDCG
0
0 comments X

The pith

A late-fusion model combining collaborative filtering, reinforcement learning and TOPSIS improves talent recommendations on one benchmark while remaining competitive on another through auditable branch weights.

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

The paper proposes CF-RL-TOPSIS, a model that combines a transition-aware collaborative filtering branch, a reinforcement-style occupation bandit, and an entropy-weighted TOPSIS branch using six semantic proxies. It evaluates this on two frozen ICT talent-history benchmarks using chronological top-5 ranking. On JobHop the full hybrid reaches NDCG@5 of 0.3040 and significantly beats repeat-last, Markov models, GRU4Rec and SASRec. On Karrierewege it stays competitive without exceeding the strongest baseline, indicating the bandit branch shrinks in persistence-heavy settings. This provides evidence on when such fusion adds value with inspectable components.

Core claim

The CF-RL-TOPSIS model integrates three branches whose validation-selected fusion coefficients remain auditable, and on JobHop it significantly surpasses multiple baselines while on Karrierewege the adaptive branch correctly becomes inactive, showing the architecture adapts to different regimes through its collaborative backbone.

What carries the argument

The CF-RL-TOPSIS late-fusion architecture that combines transition-aware collaborative filtering, a compact reinforcement-style occupation-family bandit, and an entropy-weighted TOPSIS branch constructed from six semantic proxies, with validation-selected fusion coefficients.

If this is right

  • In semantically rich talent-history settings the three branches reinforce one another to improve ranking quality.
  • In persistence-dominated regimes the same model remains competitive by relying on its collaborative filtering component.
  • Branch scores, criterion weights, and rank shifts can be inspected for individual recommendations.
  • Proxy-sensitivity and family-level deep Q-network checks support the interpretation of when each branch contributes.

Where Pith is reading between the lines

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

  • Similar late-fusion approaches could be tested in other recommendation domains where interpretability is required alongside performance.
  • The use of frozen benchmarks allows direct comparison but may limit claims about generalization to live systems.
  • Extending the semantic proxies or bandit to other domains might reveal additional conditions for fusion value.

Load-bearing premise

The validation-selected fusion coefficients and the six semantic proxies in the TOPSIS branch produce stable, auditable rankings that generalize beyond the two frozen benchmarks without post-hoc adjustment.

What would settle it

A new talent-history benchmark where the hybrid model fails to match or exceed the strongest baseline or where the selected fusion weights lead to unstable rankings.

Figures

Figures reproduced from arXiv: 2605.24155 by \"Ozkan Canay.

Figure 1
Figure 1. Figure 1: End-to-end study design from public talent-history sources to repeated chronological evaluation. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: CF-RL-TOPSIS late-fusion scoring pipeline. [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Repeated-split NDCG@5 for selected baselines and hybrids on the two ICT benchmarks. [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mean validation-selected full-hybrid branch weights on each benchmark. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
read the original abstract

Effective skills-aware talent recommendation must balance behavioral transition patterns, trajectory-sensitive adaptation, and inspectable occupation-level criteria. Evidence from public benchmarks on how these signals interact, however, remains limited. This study proposes CF-RL-TOPSIS, an interpretable late-fusion model that integrates a transition-aware collaborative branch, a compact reinforcement-style occupation-family bandit, and an entropy-weighted TOPSIS branch constructed from six semantic proxies; the validation-selected fusion coefficients remain auditable. The model is evaluated on two frozen public ICT talent-history benchmarks, JobHop and Karrierewege, using repeated chronological top-5 ranking and paired Wilcoxon tests. On JobHop the full hybrid attains NDCG@5 = 0.3040 +/- 0.0073 and significantly surpasses repeat-last, item Markov, transition-aware collaborative filtering, the CF+TOPSIS hybrid, GRU4Rec, and SASRec (p <= 0.0039 across planned comparisons). On Karrierewege the hybrid remains competitive but does not significantly exceed the strongest Markov baseline, revealing a persistence-dominated setting in which the bandit branch appropriately shrinks to near-zero weight. Proxy-sensitivity, family-level deep Q-network, and runtime checks support this interpretation, and a worked user-level case shows how branch scores, criterion weights, and rank shifts can be inspected for an individual recommendation. The contribution is not a benchmark-agnostic superiority claim, but a reproducible account of the conditions under which transparent late fusion adds value beyond simple continuation heuristics. In semantically rich, non-saturating talent-history regimes the three branches reinforce one another; in persistence-dominated regimes the same architecture remains competitive through its collaborative backbone, with the adaptive branch correctly inactive.

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 / 2 minor

Summary. The paper proposes CF-RL-TOPSIS, an interpretable late-fusion model for skills-aware talent recommendation that combines a transition-aware collaborative filtering branch, a reinforcement learning occupation-family bandit, and an entropy-weighted TOPSIS branch constructed from six semantic proxies. Fusion coefficients are selected on validation splits and remain auditable. The model is evaluated on two frozen public benchmarks (JobHop and Karrierewege) using repeated chronological top-5 ranking and paired Wilcoxon tests. On JobHop the hybrid reports NDCG@5 = 0.3040 +/- 0.0073 and significantly outperforms repeat-last, item Markov, transition-aware CF, CF+TOPSIS, GRU4Rec, and SASRec (p <= 0.0039). On Karrierewege the hybrid is competitive but does not significantly exceed the strongest Markov baseline, with the RL branch weight shrinking to near-zero. The contribution is scoped to the observed conditions under which the branches reinforce one another, supported by proxy-sensitivity checks and a worked user-level interpretability case.

Significance. If the reported gains hold under the stated conditions, the work supplies a reproducible demonstration of when transparent late fusion of behavioral, adaptive, and semantic signals adds value beyond continuation heuristics in talent-history data. Explicit credit is due for the use of chronological splits on public benchmarks, planned statistical comparisons with p-values, proxy-sensitivity and runtime checks, and the inclusion of a concrete user-level case showing branch scores, criterion weights, and rank shifts.

major comments (2)
  1. [Evaluation and Results sections] The two late-fusion coefficients and the six semantic proxies for the TOPSIS branch are selected on validation performance within each benchmark. No ablation or stability results are reported under alternative chronological validation folds or under transfer of the JobHop-selected coefficients to Karrierewege (or vice versa). This directly affects attribution of the NDCG@5 = 0.3040 gain on JobHop to intrinsic branch synergy rather than post-selection tuning.
  2. [Results on Karrierewege and proxy-sensitivity analysis] The observation that the RL branch weight collapses on the persistence-dominated Karrierewege set is noted, yet no experiment verifies whether the validation-selected coefficients from JobHop would produce comparable shrinkage or performance when applied without re-tuning. This is load-bearing for the claim that the architecture 'appropriately' adapts across regimes.
minor comments (2)
  1. [Model description] The abstract refers to a 'family-level deep Q-network' in the RL branch; a concise description of its state representation, reward definition, and training procedure would improve reproducibility without lengthening the main text.
  2. [Experimental setup] Table or figure presenting the exact numerical values of the validation-selected fusion coefficients and TOPSIS entropy weights for each dataset would make the 'auditable' claim more concrete.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments, which help clarify the scope of our evaluation claims. We address each major point below and commit to revisions that strengthen the attribution of results to branch synergy.

read point-by-point responses

  1. Referee: [Evaluation and Results sections] The two late-fusion coefficients and the six semantic proxies for the TOPSIS branch are selected on validation performance within each benchmark. No ablation or stability results are reported under alternative chronological validation folds or under transfer of the JobHop-selected coefficients to Karrierewege (or vice versa). This directly affects attribution of the NDCG@5 = 0.3040 gain on JobHop to intrinsic branch synergy rather than post-selection tuning.

    Authors: We agree that the per-benchmark validation selection of coefficients and proxies leaves open the possibility that reported gains partly reflect tuning rather than intrinsic synergy. To address this, the revised manuscript will include ablations across multiple alternative chronological validation folds on each benchmark, reporting the stability of the selected coefficients and resulting NDCG@5 values. We will also add explicit transfer experiments that apply the JobHop-selected coefficients (and proxies) directly to Karrierewege without re-tuning, and vice versa, to quantify how much performance depends on benchmark-specific selection. revision: yes

  2. Referee: [Results on Karrierewege and proxy-sensitivity analysis] The observation that the RL branch weight collapses on the persistence-dominated Karrierewege set is noted, yet no experiment verifies whether the validation-selected coefficients from JobHop would produce comparable shrinkage or performance when applied without re-tuning. This is load-bearing for the claim that the architecture 'appropriately' adapts across regimes.

    Authors: We acknowledge that the current results do not directly test whether the observed RL weight collapse on Karrierewege is reproducible under transferred coefficients. In revision we will add the requested transfer experiment: apply the full set of JobHop validation-selected coefficients to the Karrierewege test folds without any re-optimization, and report the resulting branch weights, NDCG@5, and whether the bandit weight again shrinks toward zero. This will provide direct evidence on the architecture's adaptation behavior across regimes. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical results on held-out data

full rationale

The paper reports measured NDCG@5 and statistical comparisons on chronologically held-out test portions of two public benchmarks (JobHop, Karrierewege). Fusion coefficients are explicitly chosen on a validation split and the contribution is scoped to 'reproducible account of the conditions' rather than any derivation or general claim. No equations, self-citations, or ansatzes are presented that reduce the reported performance numbers to the inputs by construction. This matches standard ML benchmark evaluation with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The model relies on standard ranking metrics and statistical tests; the only fitted elements are the late-fusion coefficients chosen on validation data and the entropy weights inside TOPSIS. No new physical or mathematical entities are postulated.

free parameters (2)
  • late-fusion coefficients
    Validation-selected weights that combine the three branch scores; their values are not reported but are described as remaining auditable.
  • TOPSIS entropy weights
    Weights derived from the six semantic proxies; these are data-dependent and part of the branch construction.
axioms (2)
  • standard math NDCG@5 and paired Wilcoxon signed-rank test assumptions hold for the chronological splits
    Used to declare statistical significance on both benchmarks.
  • domain assumption The six semantic proxies are sufficient to capture occupation-level criteria
    Invoked in the construction of the TOPSIS branch.

pith-pipeline@v0.9.1-grok · 5838 in / 1516 out tokens · 35119 ms · 2026-06-30T14:17:12.526477+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

55 extracted references · 55 canonical work pages · 1 internal anchor

  1. [1]

    Sustainability 13, 10786

    An e-commerce recommendation system based on dynamic analysis of customer behavior. Sustainability 13, 10786. doi:10.3390/su131910786. Al-bashiri, H., Abdulgabber, M.A., Romli, A.,

    work page doi:10.3390/su131910786

  2. [2]

    PLOS ONE 13, e0204434

    An improved memory-based collaborative filtering method based on the topsis technique. PLOS ONE 13, e0204434. doi:10.1371/journal.pone.0204434. Al-Hasan, T.M., Sayed, A.N., Bensaali, F., Himeur, Y., Varlamis, I., Dimitrakopoulos, G.,

    work page doi:10.1371/journal.pone.0204434

  3. [3]

    Alsaif, S.A., Hidri, M.S., Eleraky, H.A.,

    doi:10.3390/bdcc8040036. Alsaif, S.A., Hidri, M.S., Eleraky, H.A.,

    work page doi:10.3390/bdcc8040036

  4. [4]

    Azri, M., Haw, S.C., Ng, K.W.,

    doi:10.3390/computers11110161. Azri, M., Haw, S.C., Ng, K.W.,

    work page doi:10.3390/computers11110161

  5. [5]

    Bączkiewicz, A., Kizielewicz, B., Shekhovtsov, A.,

    doi:10.62527/joiv.9.2.3021. Bączkiewicz, A., Kizielewicz, B., Shekhovtsov, A.,

    work page doi:10.62527/joiv.9.2.3021

  6. [6]

    Journal of Theoretical and Applied Electronic Commerce Research 16, 2192–2229

    Methodical aspects of mcdm based e-commerce recommender system. Journal of Theoretical and Applied Electronic Commerce Research 16, 2192–2229. doi:10.3390/jtaer16060122. Bangari, S., Nayak, S., Patel, L., Rashmi, K.T.,

    work page doi:10.3390/jtaer16060122

  7. [7]

    A review on reinforcement learning based news recommendation systems and its challenges, in: 2021 International Conference on Artificial Intelligence and Smart Systems, pp. 260–265. doi:10.1109/ICAIS50930.2021.9395812. Bied, G., Nathan, S., Pérennès, E., Deleu, J., Duflo, G., Piwowarski, B., Soulier, L.,

    work page doi:10.1109/icais50930.2021.9395812 2021

  8. [8]

    5906–5914

    Toward job recommendation for all, in: Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, pp. 5906–5914. doi:10.24963/ijcai.2023/655. Canay:Preprint submitted to Knowledge-Based SystemsPage 21 of 23 CF-RL-TOPSIS Fusion for Talent Recommendation Burke, R.,

    work page doi:10.24963/ijcai.2023/655 2023

  9. [9]

    User Modeling and User-Adapted Interaction 12, 331–370

    Hybrid recommender systems: Survey and experiments. User Modeling and User-Adapted Interaction 12, 331–370. doi:10.1023/A:1021240730564. Caro-Martínez, M., Jiménez-Díaz, G., Recio-García, J.A.,

    work page doi:10.1023/a:1021240730564

  10. [10]

    Journal of Artificial Intelligence Research 71, 557–589

    Conceptual modeling of explainable recommender systems: An ontological formalization to guide their design and development. Journal of Artificial Intelligence Research 71, 557–589. doi:10.1613/jair.1.12789. Carraro, D., Bridge, D.,

    work page doi:10.1613/jair.1.12789

  11. [11]

    Journal of Intelligent Information Systems 58, 311–336

    A sampling approach to debiasing the offline evaluation of recommender systems. Journal of Intelligent Information Systems 58, 311–336. doi:10.1007/s10844-021-00651-y. Channabasamma, A., Suresh, Y.,

    work page doi:10.1007/s10844-021-00651-y

  12. [12]

    Journal of Computer Science 18, 612–621

    A recommendation-based contextual model for talent acquisition. Journal of Computer Science 18, 612–621. doi:10.3844/jcssp.2022.612.621. Chen, W.H., Hsu, C.C., Lai, Y.A.,

    work page doi:10.3844/jcssp.2022.612.621 2022

  13. [13]

    Cui, S., Sun, Y., Zhang, Y., Meng, Q., Zhu, H.,

    doi:10.3389/fdata.2019.00049. Cui, S., Sun, Y., Zhang, Y., Meng, Q., Zhu, H.,

    work page doi:10.3389/fdata.2019.00049 2019

  14. [14]

    ACM Transactions on Management Information Systems doi:10.1145/3787466

    LLM-enhanced career knowledge graph understanding for job mobility prediction. ACM Transactions on Management Information Systems doi:10.1145/3787466. Cureton, E.E.,

    work page doi:10.1145/3787466

  15. [15]

    Psychometrika 21, 287–290

    Rank-biserial correlation. Psychometrika 21, 287–290. doi:10.1007/BF02289138. Dhameliya, J., Desai, N.P.,

    work page doi:10.1007/bf02289138

  16. [16]

    International Journal of Soft Computing and Engineering 9, 8–13

    Job recommendation system using content and collaborative filtering based techniques. International Journal of Soft Computing and Engineering 9, 8–13. doi:10.35940/ijsce.c3266.099319. Du, Y., Luo, D., Yan, R., Li, Z., Ma, Y., Liu, Y., Zhang, R., Zhang, X., Zhang, L.,

    work page doi:10.35940/ijsce.c3266.099319

  17. [17]

    Proceedings of the AAAI Conference on Artificial Intelligence 38, 8363–8371

    Enhancing job recommendation through llm-based generative adversarial networks. Proceedings of the AAAI Conference on Artificial Intelligence 38, 8363–8371. doi:10.1609/aaai.v38i8.28678. Feng, J., Yang, J., Li, S., Miao, Q., Xi, Y., Xia, Z.,

    work page doi:10.1609/aaai.v38i8.28678

  18. [18]

    Expert Systems with Applications 300, 130413

    Enhancing person-job fit through multi-temporal career trajectory modeling. Expert Systems with Applications 300, 130413. doi:10.1016/j.eswa.2025.130413. Fu, M., Huang, L., Rao, A., Irissappane, A.A., Zhang, J., Qu, H.,

    work page doi:10.1016/j.eswa.2025.130413 2025

  19. [19]

    IEEE Transactions on Industrial Informatics 19, 2049–2061

    A deep reinforcement learning recommender system with multiple policies for recommendations. IEEE Transactions on Industrial Informatics 19, 2049–2061. doi:10.1109/TII. 2022.3209290. Guan, Z., Yang, J.Q., Yang, Y., Zhu, H., Li, W., Xiong, H.,

    work page doi:10.1109/tii 2049

  20. [20]

    ACM Transactions on Knowledge Discovery from Data doi:10.1145/3701735

    JobFormer: Skill-aware job recommendation with semantic- enhanced transformer. ACM Transactions on Knowledge Discovery from Data doi:10.1145/3701735. Gugnani, A., Misra, H.,

    work page doi:10.1145/3701735

  21. [21]

    Proceedings of the AAAI Conference on Artificial Intelligence 34, 13286–13293

    Implicit skills extraction using document embedding and its use in job recommendation. Proceedings of the AAAI Conference on Artificial Intelligence 34, 13286–13293. doi:10.1609/aaai.v34i08.7038. Hidasi, B., Karatzoglou, A., Baltrunas, L., et al.,

    work page doi:10.1609/aaai.v34i08.7038

  22. [22]

    Session-based Recommendations with Recurrent Neural Networks

    Session-based recommendations with recurrent neural networks. arXiv , arXiv:1511.06939doi:10.48550/arXiv.1511.06939. Hoque, S., Karim, A., Alam, M., Gope, N.,

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1511.06939

  23. [23]

    IEEE Access 14, 15778–15794

    When LLM meets Fuzzy-TOPSIS for personnel selection through automated profile analysis. IEEE Access 14, 15778–15794. doi:10.1109/ACCESS.2026.3658575. Intayoad, W., Kamyod, C., Temdee, P.,

    work page doi:10.1109/access.2026.3658575 2026

  24. [24]

    Reinforcement learning for online learning recommendation system, in: 2018 Global Wireless Summit (GWS), pp. 167–170. doi:10.1109/GWS.2018.8686513. Jadidinejad, A.H., Macdonald, C., Ounis, I.,

    work page doi:10.1109/gws.2018.8686513 2018

  25. [25]

    ACM Transactions on Information Systems 40, 1–22

    The simpson’s paradox in the offline evaluation of recommendation systems. ACM Transactions on Information Systems 40, 1–22. doi:10.1145/3458509. Johary, I., Romero, R., Mara, A., De Bie, T.,

    work page doi:10.1145/3458509

  26. [26]

    JobHop: A large-scale dataset of career trajectories, in: Proceedings of the 2025 IEEE International Conference on Big Data (BigData), IEEE. pp. 2184–2191. doi:10.1109/BigData66926.2025. 11402454. Kang, W.C., McAuley, J.,

    work page doi:10.1109/bigdata66926.2025 2025

  27. [27]

    Self-attentive sequential recommendation, in: 2018 IEEE International Conference on Data Mining (ICDM), pp. 197–206. doi:10.1109/ICDM.2018.00035. Li, Y., Liu, K., Satapathy, R., Wang, S., Cambria, E.,

    work page doi:10.1109/icdm.2018.00035 2018

  28. [28]

    IEEE Computational Intelligence Magazine 19, 78–95

    Recent developments in recommender systems: A survey. IEEE Computational Intelligence Magazine 19, 78–95. doi:10.1109/MCI.2024.3363984. Lin, Y., Liu, Y., Lin, F., Zou, L., Wu, P., Zeng, W., Miao, C.,

    work page doi:10.1109/mci.2024.3363984 2024

  29. [29]

    IEEE Transactions on Neural Networks and Learning Systems 35, 13164–13184

    A survey on reinforcement learning for recommender systems. IEEE Transactions on Neural Networks and Learning Systems 35, 13164–13184. doi:10.1109/TNNLS.2023.3280161. Liu, H., Ismail, F., Zhang, W., Zou, P., Hussain, T., Sharma, Y., Lilhore, U., Simaiya, S., Tekeste, L.,

    work page doi:10.1109/tnnls.2023.3280161 2023

  30. [30]

    Artificial Intelligence Review 54, 427–468

    A systematic literature review of multicriteria recommender systems. Artificial Intelligence Review 54, 427–468. doi:10.1007/s10462-020-09851-4. Mukhametzyanov, I.Z.,

    work page doi:10.1007/s10462-020-09851-4

  31. [31]

    doi:10.1109/ICOIACT.2018.8350761

    Deep reinforcement learning for recommender systems, in: 2018 InternationalConferenceonInformationandCommunicationsTechnology,pp.226–233. doi:10.1109/ICOIACT.2018.8350761. Nematollahi, M., Guitouni, A., Izadyar, N., Belacel, N., Park, A.,

    work page doi:10.1109/icoiact.2018.8350761 2018

  32. [32]

    Computers and Operations Research doi:10.1016/j.cor.2026.107426

    Multi-attribute utility deep reinforcement learning method for sequential multi-criteria decision problems: Application to human resource planning. Computers and Operations Research doi:10.1016/j.cor.2026.107426. Panda, M., Jagadev, A.K.,

    work page doi:10.1016/j.cor.2026.107426 2026

  33. [33]

    Topsis in multi-criteria decision making: a survey, in: 2018 International Conference on Data Science and Business Analytics, pp. 51–54. doi:10.1109/ICDSBA.2018.00017. Peng, S., Siet, S., Ilkhomjon, S.,

    work page doi:10.1109/icdsba.2018.00017 2018

  34. [34]

    Canay:Preprint submitted to Knowledge-Based SystemsPage 22 of 23 CF-RL-TOPSIS Fusion for Talent Recommendation Poncio, F.P.,

    doi:10.3390/app14031155. Canay:Preprint submitted to Knowledge-Based SystemsPage 22 of 23 CF-RL-TOPSIS Fusion for Talent Recommendation Poncio, F.P.,

    work page doi:10.3390/app14031155

  35. [35]

    Journal of Research in Innovative Teaching & Learning 17, 352–367

    Navigating techniques in job recommender systems on internship profile matching: A systematic review. Journal of Research in Innovative Teaching & Learning 17, 352–367. doi:10.1108/JRIT-01-2024-0016. Reusens, M., Lemahieu, W., Baesens, B.,

    work page doi:10.1108/jrit-01-2024-0016 2024

  36. [36]

    Decision Support Systems 98, 26–35

    A note on explicit versus implicit information for job recommendation. Decision Support Systems 98, 26–35. doi:10.1016/j.dss.2017.04.002. Roy, D., Dutta, M.,

    work page doi:10.1016/j.dss.2017.04.002 2017

  37. [37]

    Schellingerhout,R.,2024

    doi:10.1186/s40537-022-00592-5. Schellingerhout,R.,2024. Explainablemulti-stakeholderjobrecommendersystems,in:Proceedingsofthe18thACMConference on Recommender Systems, pp. 1318–1322. doi:10.1145/3640457.3688014. Senger, E., Campbell, Y., van der Goot, R., Plank, B.,

    work page doi:10.1186/s40537-022-00592-5 2024

  38. [38]

    arXiv preprint arXiv:2412.14612 doi:10.48550/arXiv.2412.14612

    Karrierewege: A large scale career path prediction dataset. arXiv preprint arXiv:2412.14612 doi:10.48550/arXiv.2412.14612. Senger, E., Campbell, Y., van der Goot, R., Plank, B.,

    work page doi:10.48550/arxiv.2412.14612

  39. [39]

    Frontiers in Big Data 8, 1564521

    Toward more realistic career path prediction: Evaluation and methods. Frontiers in Big Data 8, 1564521. doi:10.3389/fdata.2025.1564521. Shani, G., Heckerman, D., Brafman, R.I.,

    work page doi:10.3389/fdata.2025.1564521 2025

  40. [40]

    ACM Transactions on Information Systems 43, 1–35

    Market-aware long-term job skill recommendation with explainable deep reinforcement learning. ACM Transactions on Information Systems 43, 1–35. doi:10.1145/3704998. Sun, Y., Zhuang, F., Zhu, H., Zhang, Q., Xiong, H., He, Q.,

    work page doi:10.1145/3704998

  41. [41]

    Proceedings of The Web Conference 2021 , 3827–3838doi:10.1145/3442381.3449985

    Cost-effective and interpretable job skill recommendation with deep reinforcement learning. Proceedings of The Web Conference 2021 , 3827–3838doi:10.1145/3442381.3449985. Tang, F., Zhu, R., Yao, F., Li, Q., Guo, X.,

    work page doi:10.1145/3442381.3449985 2021

  42. [42]

    Frontiers in Artificial Intelligence 8, 1660548

    Explainable person–job recommendations: Challenges, approaches, and comparative analysis. Frontiers in Artificial Intelligence 8, 1660548. doi:10.3389/frai.2025.1660548. Tang, X., Chen, Y., Li, X., Liu, J., Ying, Z.,

    work page doi:10.3389/frai.2025.1660548 2025

  43. [43]

    British Journal of Mathematical and Statistical Psychology 72, 108–135

    A reinforcement learning approach to personalized learning recommendation systems. British Journal of Mathematical and Statistical Psychology 72, 108–135. doi:10.1111/bmsp.12144. Tiwary, N., Noah, S.A.M., Fauzi, F., Yee, T.S.,

    work page doi:10.1111/bmsp.12144

  44. [44]

    IEEE Access 12, 91999–92019

    A review of explainable recommender systems utilizing knowledge graphs and reinforcement learning. IEEE Access 12, 91999–92019. doi:10.1109/ACCESS.2024.3422416. Wilcoxon, F.,

    work page doi:10.1109/access.2024.3422416 2024

  45. [45]

    Biometrics Bulletin 1, 80–83

    Individual comparisons by ranking methods. Biometrics Bulletin 1, 80–83. doi:10.2307/3001968. Wu, L., Qiu, Z., Zheng, Z., Li, Y., Xu, H., Chen, E., Xiong, H.,

    work page doi:10.2307/3001968

  46. [46]

    Proceedings of the AAAI Conference on Artificial Intelligence 38, 9178–9186

    Exploring large language model for graph data understanding in online job recommendations. Proceedings of the AAAI Conference on Artificial Intelligence 38, 9178–9186. doi:10.1609/aaai.v38i8.28769. Xin, X., Karatzoglou, A., Arapakis, I., Jose, J.M.,

    work page doi:10.1609/aaai.v38i8.28769

  47. [47]

    Self-supervised reinforcement learning for recommender systems, in: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 931–940. doi:10.1145/3397271.3401147. Zavadskas, E.K., Mardani, A., Turskis, Z., Jusoh, A., Nor, K.M.,

    work page doi:10.1145/3397271.3401147

  48. [48]

    IEEE Transactions on Computational Social Systems doi:10.1109/TCSS.2023.3239038

    Career mobility analysis with uncertainty- aware graph autoencoders: A job title transition perspective. IEEE Transactions on Computational Social Systems doi:10.1109/TCSS.2023.3239038. Zhang, E., Ma, W., Zhang, J., Xia, X.,

    work page doi:10.1109/tcss.2023.3239038 2023

  49. [49]

    Zhang, Y., Chen, X.,

    doi:10.3390/electronics12245034. Zhang, Y., Chen, X.,

    work page doi:10.3390/electronics12245034

  50. [50]

    Foundations and Trends in Information Retrieval 14, 1–101

    Explainable recommendation: A survey and new perspectives. Foundations and Trends in Information Retrieval 14, 1–101. doi:10.1561/1500000066. Zhao, W.X., Lin, Z., Feng, Z., Wang, H., Wen, J.R., Li, J.,

    work page doi:10.1561/1500000066

  51. [51]

    ACM Transactions on Information Systems 41, 1–41

    A revisiting study of appropriate offline evaluation for top-n recommendation algorithms. ACM Transactions on Information Systems 41, 1–41. doi:10.1145/3545796. Zhao, X., Gu, C., Zhang, H., Yang, X., Liu, X., Tang, J., Liu, H.,

    work page doi:10.1145/3545796

  52. [52]

    Dear: Deep reinforcement learning for online advertising impression in recommender systems, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 750–758. doi:10.1609/aaai.v35i1.16156. Zheng, Y., Wang, D.X.,

    work page doi:10.1609/aaai.v35i1.16156

  53. [53]

    Neurocomputing 474, 141–153

    A survey of recommender systems with multi-objective optimization. Neurocomputing 474, 141–153. doi:10.1016/j.neucom.2021.11.041. Zhou, S., Dai, X., Chen, H., Zhang, W., Ren, K., Tang, R., Yu, Y.,

    work page doi:10.1016/j.neucom.2021.11.041 2021

  54. [54]

    Interactive recommender system via knowledge graph-enhanced reinforcement learning, in: Proceedings of the 43rd ACM SIGIR Conference, pp. 179–188. doi:10.1145/ 3397271.3401174. Zou, Z., Huspi, S.H., Nuar, A.N.A.,

    work page arXiv

  55. [55]

    Journal of Advanced Research in Applied Sciences and Engineering Technology 41, 113–124

    A review on job recommendation system. Journal of Advanced Research in Applied Sciences and Engineering Technology 41, 113–124. doi:10.37934/araset.41.2.113124. Canay:Preprint submitted to Knowledge-Based SystemsPage 23 of 23

    work page doi:10.37934/araset.41.2.113124