Pith. sign in

REVIEW 4 major objections 5 minor 61 references

Personality can be read from text without locking to Big-5 or MBTI by learning shared latent pseudo-facets and letting an LLM reweight noisy labels.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-10 08:30 UTC pith:3JQMJBAQ

load-bearing objection Solid engineering package for cross-theory personality transfer; the theory-agnostic claim is overstated relative to the two-dataset evidence. the 4 major comments →

arxiv 2607.08374 v1 pith:3JQMJBAQ submitted 2026-07-09 cs.CL cs.AIcs.HCcs.ROcs.SI

Large-Language-Models-as-a-Judge in Theory-Agnostic Adaptive Metric-Alignment for Prototypical Networks in Personality Recognition

Jing Jie Tan , Ban-Hoe Kwan , Danny Wee-Kiat Ng , Yan-Chai Hum , Shih-Yu Lo , Po-An Chen , Noriyuki Kawarazaki , Kosuke Takano
show 1 more author
Anissa Mokraoui
This is my paper
classification cs.CL cs.AIcs.HCcs.ROcs.SI
keywords personality recognitiontheory-agnosticprototypical networksLLM-as-a-Judgecross-theory harmonizationpseudo-facetsmetric learningnatural language understanding
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Personality models usually train to match one fixed theory, so they fail when labels come from another theory and struggle with small or noisy datasets. This paper argues that personality itself is theory-invariant and that the right model can discover the shared latent structure (called pseudo-facets) that different taxonomies only partially capture. JAM does this with an attention-pooled graph prototypical network that clusters embeddings, a Cross-Theory Harmonization step that mixes human-guided linkages with machine consensus, and an LLM-as-a-Judge that marks samples as possible, ambiguous or impossible so their training weight can be adjusted. The result is a single model that trains on mixed Essays (Big-5) and Kaggle (MBTI) data, improves balanced accuracy on both, and can output a latent psychological profile at inference without any theory-specific labels. A reader who wants better personalization or low-resource theories cares because the same pipeline reduces dependence on any one questionnaire and cleans the data that currently limit progress.

Core claim

By learning unified latent pseudo-facets through prototypical clustering and Cross-Theory Harmonization (human-guided linkage plus machine-induced consensus), then reweighting samples with an LLM judge, a single network can generalize across personality theories and raise balanced accuracy without requiring theory-specific labels at inference.

What carries the argument

Cross-Theory Harmonization (CTH) built on an Attention-Pooled Graph Prototypical Network: Human-Guided Linkage soft-aligns dimensions across theories while Machine-Induced Consensus discovers shared structure from data; LLM-as-a-Judge (before- or in-the-loop) then sets per-sample weights so ambiguous or impossible labels contribute less to the metric.

Load-bearing premise

The method assumes that the partial human mappings between theories plus automatic consensus recover real shared psychological structure rather than merely aligning noisy, theory-specific label artifacts.

What would settle it

Train the full JAM pipeline on Big-5 and MBTI with the stated linkages, then evaluate zero-shot or few-shot transfer to a third independent taxonomy (for example HEXACO) that was never used in training or linkage; if performance collapses relative to a theory-specific baseline, the claimed shared pseudo-facets are not theory-invariant.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • A single model trained on mixed Big-5 and MBTI data can serve both frameworks and improve the lower-resource Essays setting without harming the larger Kaggle setting.
  • At inference the network can emit a latent psychological profile directly from text, removing the need to choose a taxonomy or supply theory-specific labels.
  • LLM-before-the-loop reweighting of impossible and ambiguous samples yields a further accuracy lift on constrained data while muting leakage risks of pure generative inference.
  • Prototypical episode training naturally mitigates class imbalance that ordinary cross-entropy suffers from.
  • The same harmonization path can in principle absorb additional low-resource theories once a modest human linkage table is supplied.

Where Pith is reading between the lines

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

  • The same CTH-plus-judge pattern could be applied to other multi-taxonomy domains such as emotion or clinical symptom checklists where rival inventories only partially overlap.
  • If the discovered pseudo-facets prove stable, they offer a natural common embedding space for privacy-preserving federated training across institutions that never share raw labels.
  • The fact that Human-Guided Linkage alone hurts performance suggests that purely expert mappings are too rigid; future work could replace the fixed table with learned soft alignments.
  • A controlled ablation that randomizes the human linkage table would quantify how much of the reported gain is genuine structure versus artifactual label alignment.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes JAM, a theory-agnostic personality recognition framework that learns latent “pseudo-facets” via an Attention-Pooled Graph Prototypical Network (Longformer layer embeddings as fully-connected GNN nodes with attention pooling), Cross-Theory Harmonization (Human-Guided Linkage from Table I mappings plus Machine-Induced Consensus), and LLM-as-a-Judge reweighting of samples (Possible/Ambiguous/Impossible) in LBL or LIL configurations. It claims the model discovers shared psychological structure across Big-5 (Essays) and MBTI (Kaggle) without constraining to any taxonomy at training or inference, can infer latent profiles from text alone, improves cross-framework generalization (reported ~9–12% BA lifts on Essays under joint training, plus ~2.4% from LAJ), supports low-resource theories, and remains competitive with prior single-theory methods while reducing FLOPs relative to decoder-only baselines.

Significance. If the central claim holds—that CTH recovers genuine shared latent structure rather than merely aligning theory-specific label correlations—the work would be a useful step toward multi-theory personality modeling under data scarcity and privacy constraints, with practical value for recommendation and low-resource settings. Strengths that should be credited include thorough ablations (HGL alone hurts, MIC mixed, full CTH recovers; LBL > LIL), McNemar tests, t-SNE visualizations, multi-LLM sensitivity (gpt/qwen/llama/ds) with hyperparameter sweeps on z, FLOPs/cost comparison, and public code/weights. Gains are modest once strong single-dataset PF baselines are considered, and validation is limited to two English public datasets/theories, so the significance is currently more engineering/methodological than a definitive psychological or theoretical advance.

major comments (4)
  1. [§II-A, Table I, Fig. 3, §III-B2] §II-A, Table I, Fig. 3, and §III-B2 (CTH): The theory-agnostic “pseudo-facet” claim is load-bearing yet rests on partial, non-equivalent mappings (O↔S/N, C↔P/J, E↔I/E, A↔T/F) that the paper itself describes as “partial conceptual overlap rather than equivalence.” HGL alone degrades performance on both datasets (Tables IV–V), while full CTH recovers/improves mainly on Essays. Without independent evidence (e.g., human facet ratings, held-out third theory, or external behavioral correlates) that the purple shared clusters in Fig. 3(f) reflect psychological structure rather than label co-occurrence artifacts, the reported cross-framework BA gains and the “unified latent pseudo-facets” narrative remain under-supported.
  2. [Abstract, §I, Eq. (6)] Abstract, §I, and §III-B1/Eq. (6): The repeated claim that the model “can infer an individual’s latent psychological profile directly from the textual samples, without requiring theory-specific labels” is overstated. Prototypes are still formed from labeled support sets (D^(S) with y), evaluation remains multi-label accuracy against theory-specific ground truth, and inference of a usable profile still maps back to the training taxonomies. Clarify what is truly label-free at test time versus what still depends on theory-labeled supports for few-shot prototypes, and report a pure unsupervised or zero-label transfer protocol if that is the intended claim.
  3. [Tables IV–V, §IV-B] Tables IV–V and §IV-B: Joint training with CTH+JAM improves Essays substantially relative to the best single-theory baseline, but Kaggle gains are small or mixed and sometimes below the strong PF [Kaggle] baseline. Given that Kaggle is larger and higher-quality, the “supporting low-resource personality theories” claim needs a clearer quantification of when joint training helps versus hurts the high-resource theory, plus statistical tests against the strongest single-dataset PF baselines rather than only against weaker CE/PO or partial ablations.
  4. [§III-B3, Eq. (9), Fig. 7] §III-B3, Eq. (9), Fig. 7: The LLM-as-a-Judge reweighting (z=1/0.2/0) is shown to be sensitive; muting Impossible samples helps Essays but can hurt Kaggle, and the controlled 80% minor-class removal experiment produces large drops. This indicates that LLM judgments can inject new bias or noise. The paper should either provide human agreement rates / inter-judge reliability for the Possible/Impossible/Ambiguous labels or treat LAJ more cautiously as a noisy filter whose net benefit is dataset-dependent rather than a general robustness mechanism.
minor comments (5)
  1. [Eq. (10), Table III] Eq. (10) and Table III: The annealing schedule for ψ (HGL) is described only as “ψ_{e+1} ≤ ψ_e”; exact functional form, initial values, and ρ/ϕ settings used for the reported CTH and JAM runs should be stated for reproducibility.
  2. [Fig. 5, Tables IV–V] Fig. 5 and Tables IV–V: Baseline “Representative” rows are the best of several settings; make the selection criterion explicit and report the same metric suite (BA/F1/RA) for all CE/PO/PF variants to avoid cherry-picking.
  3. [Fig. 2, §III-B1] Fig. 2 and §III-B1: The fully-connected adjacency over Longformer layers is simple; a short ablation against mean-pooling or last-layer only would strengthen the claim that the GNN+attention pooling is necessary.
  4. [Throughout] Notation: z is used both for sample confidence and (in places) as a generic weight; distinguish sample weight w from LLM judgment ζ more consistently. Minor typos (e.g., “myPersonality”, “discon-tinued”, “Ambigous” in Fig. 7) should be cleaned.
  5. [§II-B] Related work: Several recent multi-theory or LLM-based personality papers are cited mainly as baselines; a clearer positioning of what is new relative to the authors’ own PICEPR/EERPD line would help readers.

Circularity Check

1 steps flagged

No load-bearing circularity: empirical gains from standard prototypical meta-learning + CTH ablations on public splits; self-cites are baselines/comparators only.

specific steps
  1. self citation load bearing [§IV-D / Fig. 10 and related-work citations [40],[49]]
    "Overall, the proposed method achieves the lowest inference time compared to the PICEPR (Embeddings) method [49]. ... Despite this, the method still achieves approximately 8× lower training FLOPs. ... reducing the risk of data leakage, which is difficult to achieve with purely decoder-only model-driven approaches such as PICEPR (Contents) [49]."

    PICEPR is prior work by overlapping authors; it is used as the primary computational baseline. The comparison is not load-bearing for the accuracy claims (those rest on public-test BA against independent priors and ablations), so the circularity is minor and non-central.

full rationale

The derivation chain is ordinary supervised/meta-learning: Longformer embeddings feed an attention-pooled GNN whose prototypes (Eq. 6) are centroids of labeled support embeddings; query classification is nearest-prototype under Euclidean distance; the joint loss (Eq. 10) simply sums PF + HGL + MIC terms with optional LLM-derived sample weights z (Eq. 9). All reported BA/F1/RA numbers are measured on held-out test splits of the public Essays and Kaggle corpora after training on the complementary support sets; nothing is fitted to the test metrics and then re-reported as a prediction. Table I mappings supply soft HGL supervision that is acknowledged as partial and non-equivalent; HGL alone degrades performance (Tables IV–V), so the final CTH gains are not forced by those mappings. Self-citations ([40], [49] PICEPR/EERPD) appear only as prior baselines, FLOPs comparators (Fig. 10), or data-augmentation recipes; they do not supply uniqueness theorems, forced ansätze, or the target performance numbers. LLM-as-a-Judge reweighting is an external filter, not a definitional identity with the evaluation metrics. Consequently the central claim (cross-framework BA lifts via latent pseudo-facets) remains an empirical observation rather than a tautology of the inputs. Minor residual risk of LLM pre-training contamination is noted but does not constitute circularity under the stated criteria.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 3 invented entities

The central claim rests on standard ML practice plus domain assumptions about personality structure and the reliability of LLM judges and partial theory mappings. Free parameters are ordinary training hyper-parameters and the discrete z-reweight schedule; invented entities are the pseudo-facets and the CTH construction itself, which lack independent psychological validation outside the reported accuracy gains.

free parameters (4)
  • z-reweight schedule (Possible=1.0, Ambiguous=0.2, Impossible=0) = 1.0 / 0.2 / 0 (default)
    Hand-chosen sample weights that directly control prototype formation and loss contribution; sensitivity plots show performance depends on these values.
  • loss coefficients ϕ, ψ, ρ and HGL annealing schedule = ϕ=1, ρ=1, ψ annealed
    Balance PF, HGL and MIC terms; ψ is annealed, values chosen experimentally.
  • learning rate, batch size, max episodes, seed = 1e-5, 32, 30000, 42
    Standard optimizer hyper-parameters fixed at 1e-5 / 32 / 30k / 42.
  • LIL loss threshold τ
    Determines which query samples are sent to the LLM judge; not derived from first principles.
axioms (5)
  • domain assumption Personality is theory-invariant and can be recovered as shared latent pseudo-facets in embedding space across heterogeneous label taxonomies.
    Stated in abstract and §I; underpins the entire CTH and theory-agnostic claim.
  • domain assumption Partial conceptual mappings between Big-5, MBTI and HEXACO (Table I) supply useful soft supervision for Human-Guided Linkage.
    Used to construct HGL; paper itself notes the mappings are only partial overlaps.
  • ad hoc to paper An LLM prompted with CoT can reliably classify a text–label pair as Possible / Impossible / Ambiguous for reweighting.
    Core of the LAJ component; no external gold standard for the judgments is provided.
  • domain assumption Personality dimensions may be treated as multi-label (independent enough for separate prototypes) while still allowing gradient-shared correlations.
    Eq. 1 and surrounding text; standard but contested in psychology literature the paper cites.
  • ad hoc to paper Fully-connected weighted graph over Longformer layer embeddings plus attention pooling yields useful cross-layer fusion for personality.
    Architectural choice in §III-B1; left as future work to learn adaptive graphs.
invented entities (3)
  • latent pseudo-facets no independent evidence
    purpose: Theory-invariant clusters in embedding space that replace predefined trait taxonomies at inference.
    Central modeling object; existence is inferred from improved accuracy and t-SNE structure, not independently measured.
  • Cross-Theory Harmonization (CTH = HGL + MIC) no independent evidence
    purpose: Unify heterogeneous theory-labeled datasets into a shared representation without requiring a single taxonomy.
    New composite module introduced by the paper; validated only by the reported ablations.
  • JAM (Judge for Adaptive Metric-Alignment) pipeline no independent evidence
    purpose: End-to-end system combining graph prototypes, CTH and dual-mode LLM judges.
    The named contribution; no external replication yet.

pith-pipeline@v1.1.0-grok45 · 34054 in / 3583 out tokens · 42940 ms · 2026-07-10T08:30:03.947682+00:00 · methodology

0 comments
read the original abstract

Personality recognition has traditionally been constrained by theory-dependent formulations, where models are trained to fit predefined psychological taxonomies rather than uncovering shared underlying behavioral structure. This limits generalization, as personality itself is better understood as theory-invariant, while existing annotations reflect only partial and sometimes inconsistent views of the same latent traits. In this work, we introduce JAM ((J)udge for (A)daptive (M)etric-Alignment), a theory-agnostic framework that shifts learning from adapting to predefined personality theories toward discovering unified latent pseudo-facets that capture shared psychological structure. Rather than constraining the model to any personality taxonomy during training or inference, the framework learns generalizable psychological representations and can infer an individual's latent psychological profile directly from the textual samples, without requiring theory-specific labels. JAM achieves this through an Attention-Pooled Graph Prototypical Network that learns structured representations via clustering in embedding space, together with a Cross-Theory Harmonization (CTH) approach that integrates (i) Human-Guided Linkage and (ii) Machine-Induced Consensus to unify heterogeneous datasets without relying on predefined labels. To further improve robustness and data quality, we incorporate an LLM-as-a-Judge mechanism operating in two configurations, (i) LLM-before-the-loop and (ii) LLM-in-the-loop which identifies ambiguous samples to guide adaptive metric learning. Experiments show that JAM improves cross-framework generalization and performance, establishing a strong step toward theory-agnostic personality inference and supporting low-resource personality theories. The related code repository, model weights, and artifacts are available at https://research.jingjietan.com/JAM

Figures

Figures reproduced from arXiv: 2607.08374 by Anissa Mokraoui, Ban-Hoe Kwan, Danny Wee-Kiat Ng, Jing Jie Tan, Kosuke Takano, Noriyuki Kawarazaki, Po-An Chen, Shih-Yu Lo, Yan-Chai Hum.

Figure 1
Figure 1. Figure 1: Overview of the relationship, terminology, and theoretical structure [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed JAM architecture for theory-agnostic personality recognition. The framework integrates a language embedding backbone with [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The conceptual schematic visualization comparing the effects of [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The CoT System Prompt generates output in a structured JSON [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: illustrates the distribution of Possible and Impossible judgments across personality traits for both datasets. The Essays dataset contains a higher proportion of Possible judg￾ments (85.7%) compared to the Kaggle dataset (70.8%). Given that only a limited proportion of the dataset is estimated to be noisy (approximately > 8%), the observed 2% improvement is within a reasonable range. This indicates that th… view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of balanced accuracy for regular classification using [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: The Sankey diagram illustrates the transitions between 4 approaches: [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of different large language models ( [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of Cost and FLOPs Comparison. In this analysis, [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗

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

61 extracted references · 61 canonical work pages · 12 internal anchors

  1. [1]

    A survey on personality-aware recommendation systems,

    S. Dhelim, N. Aung, M. A. Bouras, H. Ning, and E. Cambria, “A survey on personality-aware recommendation systems,”Artificial Intelligence Review, vol. 55, no. 3, p. 2409–2454, Sep. 2021. [Online]. Available: http://dx.doi.org/10.1007/s10462-021-10063-7

    work page doi:10.1007/s10462-021-10063-7 2021

  2. [2]

    The power of personalization: A systematic review of personality-adaptive chatbots,

    T. Ait Baha, M. El Hajji, Y . Es-Saady, and H. Fadili, “The power of personalization: A systematic review of personality-adaptive chatbots,” SN Computer Science, vol. 4, no. 5, Aug. 2023. [Online]. Available: http://dx.doi.org/10.1007/s42979-023-02092-6

    work page doi:10.1007/s42979-023-02092-6 2023

  3. [3]

    Exploring Personality-Driven Personalization in XAI: Enhancing User Trust in Gameplay

    Z. Li, S. Yang, and S. Wang, “Exploring personality-driven personalization in xai: Enhancing user trust in gameplay,” 2024. [Online]. Available: https://arxiv.org/abs/2408.04778

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  4. [4]

    Tackling cold-start with deep personalized transfer of user preferences for cross-domain recommendation,

    S. Omidvar and T. Tran, “Tackling cold-start with deep personalized transfer of user preferences for cross-domain recommendation,” International Journal of Data Science and Analytics, Nov. 2023. [Online]. Available: http://dx.doi.org/10.1007/s41060-023-00467-9

    work page doi:10.1007/s41060-023-00467-9 2023

  5. [5]

    Big-Five, MPTI, Eysenck or HEXACO: The Ideal Personality Model for Personality-aware Recommendation Systems

    S. Dhelim, L. L. Chen, N. Aung, W. Zhang, and H. Ning, “Big-five, mpti, eysenck or hexaco: The ideal personality model for personality-aware recommendation systems,” 2021. [Online]. Available: https://arxiv.org/abs/2106.03060

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  6. [6]

    mypersonality project website,

    D. Stillwell and M. Kosinski, “mypersonality project website,” 2015

    work page 2015

  7. [7]

    Personality computing: New frontiers in personality assessment,

    L. V . Phan and J. F. Rauthmann, “Personality computing: New frontiers in personality assessment,”Social and Personality Psychology Compass, vol. 15, no. 7, Jun. 2021. [Online]. Available: http: //dx.doi.org/10.1111/spc3.12624

    work page doi:10.1111/spc3.12624 2021

  8. [8]

    Social media addiction and empathy: Moderating impact of personality traits among high school students,

    M. Dalvi-Esfahani, A. Niknafs, Z. Alaedini, H. Barati Ahmadabadi, D. J. Kuss, and T. Ramayah, “Social media addiction and empathy: Moderating impact of personality traits among high school students,” Telematics and Informatics, vol. 57, p. 101516, Mar. 2021. [Online]. Available: http://dx.doi.org/10.1016/j.tele.2020.101516

    work page doi:10.1016/j.tele.2020.101516 2021

  9. [9]

    The impact of personality traits on social media use and engagement: An overview

    G. Lampropoulos, T. Anastasiadis, K. Siakas, and E. Siakas, “The impact of personality traits on social media use and engagement: An overview.” International Journal on Social and Education Sciences, vol. 4, no. 1, pp. 34–51, 2022

    work page 2022

  10. [10]

    Social media addiction, personality traits, and disorders: an overview of recent literature,

    E. Ahmed and S. Ahmed, “Social media addiction, personality traits, and disorders: an overview of recent literature,”Current Opinion in Psychiatry, vol. 38, no. 1, p. 72–77, Sep. 2024. [Online]. Available: http://dx.doi.org/10.1097/YCO.0000000000000969

    work page doi:10.1097/yco.0000000000000969 2024

  11. [11]

    Personality assessment,

    R. M. Spielman, K. Dumper, W. Jenkins, A. Lacombe, M. Lovett, and M. Perlmutter, “Personality assessment,”Introduction to Psychology (A critical approach), 2021

    work page 2021

  12. [12]

    What kinds of methods do personality psychologists use?: A survey of journal editors and editorial board members,

    R. W. Robins, J. L. Tracy, and J. W. Sherman, “What kinds of methods do personality psychologists use?: A survey of journal editors and editorial board members,” 2023

    work page 2023

  13. [13]

    Comparative content analysis of self-report scales for level of personality functioning,

    M. H. Waugh, C. M. McClain, E. C. Mariotti, A. L. Mulay, E. N. DeV ore, K. A. Lenger, A. N. Russell, A. R. Florimbio, K. C. Lewis, J. M. Ridenour, and L. G. Beevers, “Comparative content analysis of self-report scales for level of personality functioning,”Journal of Personality Assessment, vol. 103, no. 2, p. 161–173, Jan. 2020. [Online]. Available: http:...

    work page doi:10.1080/00223891.2019.1705464 2020

  14. [14]

    The study of projective methods in psychology,

    I. MUKHTASAR and M. MA VLUDA, “The study of projective methods in psychology,”JournalNX, vol. 7, no. 02, pp. 66–69, 2021

    work page 2021

  15. [15]

    Applicant faking on personality tests: Good or bad and why should we care?

    R. Tett and D. Simonet, “Applicant faking on personality tests: Good or bad and why should we care?”Personnel Assessment and Decisions, vol. 7, no. 1, May 2021. [Online]. Available: http://dx.doi.org/10.25035/pad.2021.01.002

    work page doi:10.25035/pad.2021.01.002 2021

  16. [16]

    Personality psychology,

    B. W. Roberts and H. J. Yoon, “Personality psychology,”Annual Review of Psychology, vol. 73, no. 1, p. 489–516, Jan. 2022. [Online]. Available: http://dx.doi.org/10.1146/annurev-psych-020821-114927

    work page doi:10.1146/annurev-psych-020821-114927 2022

  17. [17]

    Cervone and L

    D. Cervone and L. A. Pervin,Personality: Theory and research. John Wiley & Sons, 2022

    work page 2022

  18. [18]

    The myers-briggs type indicator: Manual (1962)

    I. B. Myers, “The myers-briggs type indicator: Manual (1962).” 1962

    work page 1962

  19. [19]

    Myers-briggs type indicator (mbti),

    A. Furnham, “Myers-briggs type indicator (mbti),”Encyclopedia of Personality and Individual Differences, pp. 1–4, 2017

    work page 2017

  20. [20]

    The structure of phenotypic personality traits

    L. R. Goldberg, “The structure of phenotypic personality traits.”Amer- ican Psychologist, vol. 48, pp. 26–34, 1993

    work page 1993

  21. [21]

    The association between the big five personality traits and driving behaviors: A systematic review and meta-analysis,

    X. Luo, Y . Ge, and W. Qu, “The association between the big five personality traits and driving behaviors: A systematic review and meta-analysis,”Accident Analysis and Prevention, vol. 183, p. 106968, 2023. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0001457523000155

    work page 2023

  22. [22]

    A six-factor structure of personality- descriptive adjectives: Solutions from psycholexical studies in seven languages

    M. C. Ashton, K. Lee, M. Perugini, P. Szarota, R. E. de Vries, L. D. Blas, K. Boies, and B. D. Raad, “A six-factor structure of personality- descriptive adjectives: Solutions from psycholexical studies in seven languages.”Journal of Personality and Social Psychology, vol. 86, pp. 356–366, 2004

    work page 2004

  23. [23]

    Analysis of personality traits from text based answers using hexaco model,

    P. William and A. Badholia, “Analysis of personality traits from text based answers using hexaco model,” in2021 International Conference on Innovative Computing, Intelligent Communication and Smart Elec- trical Systems (ICSES), 2021, pp. 1–10

    work page 2021

  24. [24]

    Psychology- informed natural language understanding: Integrating personality and emotion-aware features for comprehensive sentiment analysis and depression detection,

    J. J. Tan, B.-H. Kwan, D. W.-K. Ng, and Y . C. Hum, “Psychology- informed natural language understanding: Integrating personality and emotion-aware features for comprehensive sentiment analysis and depression detection,”Pertanika Journal of Science and Technology, vol. 33, no. S4, Jun. 2025. [Online]. Available: http://dx.doi.org/10. 47836/pjst.33.S4.04

    work page 2025

  25. [25]

    Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transformers,

    W. Wang, F. Wei, L. Dong, H. Bao, N. Yang, and M. Zhou, “Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transformers,” inAdvances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 5776–5788. [Online]. Available: ht...

    work page 2020

  26. [26]

    Mpnet: Masked and permuted pre-training for language understanding,

    K. Song, X. Tan, T. Qin, J. Lu, and T.-Y . Liu, “Mpnet: Masked and permuted pre-training for language understanding,” in Advances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 16 857–16 867. [Online]. Available: https://proceedings.neurips.cc/paper_...

    work page 2020

  27. [27]

    Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models

    J. Ni, G. H. Ábrego, N. Constant, J. Ma, K. B. Hall, D. Cer, and Y . Yang, “Sentence-t5: Scalable sentence encoders from pre-trained text-to-text models,” 2021. [Online]. Available: https://arxiv.org/abs/2108.08877

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  28. [28]

    Fine-tuning sentence-RoBERTa to construct word embeddings for low-resource languages from bilingual dictionaries,

    D. Bear and P. Cook, “Fine-tuning sentence-RoBERTa to construct word embeddings for low-resource languages from bilingual dictionaries,” inProceedings of the Workshop on Natural Language Processing for Indigenous Languages of the Americas (AmericasNLP), M. Mager, A. Ebrahimi, A. Oncevay, E. Rice, S. Rijhwani, A. Palmer, and K. Kann, Eds. Toronto, Canada: ...

    work page 2023

  29. [29]

    Sentence embedding generation framework based on kullback–leibler divergence optimization and roberta knowledge distillation,

    J. Han and L. Yang, “Sentence embedding generation framework based on kullback–leibler divergence optimization and roberta knowledge distillation,”Mathematics, vol. 12, no. 24, p. 3990, Dec. 2024. [Online]. Available: http://dx.doi.org/10.3390/math12243990

    work page doi:10.3390/math12243990 2024

  30. [30]

    Zero-shot emotion detection for semi-supervised sentiment analysis using sentence transformers and ensemble learning,

    S. G. Tesfagergish, J. Kapo ˇci¯ut˙e-Dzikien˙e, and R. Damaševi ˇcius, “Zero-shot emotion detection for semi-supervised sentiment analysis using sentence transformers and ensemble learning,”Applied Sciences, vol. 12, no. 17, 2022. [Online]. Available: https://www.mdpi.com/ 2076-3417/12/17/8662

    work page 2022

  31. [31]

    An empirical evaluation of the zero-shot, few-shot, and traditional fine-tuning based pretrained language models for sentiment analysis in software engineering,

    M. Shafikuzzaman, M. R. Islam, A. C. Rolli, S. Akhter, and N. Seliya, “An empirical evaluation of the zero-shot, few-shot, and traditional fine-tuning based pretrained language models for sentiment analysis in software engineering,”IEEE Access, vol. 12, pp. 109 714–109 734, 2024

    work page 2024

  32. [32]

    Eerpd: Leveraging emotion and emotion regulation for improving personality detection,

    Z. Li, D. Zhu, Q. Ma, W. Xiong, and S. Li, “Eerpd: Leveraging emotion and emotion regulation for improving personality detection,”

    work page

  33. [33]

    EERPD: Leveraging Emotion and Emotion Regulation for Improving Personality Detection

    [Online]. Available: https://arxiv.org/abs/2406.16079

    work page internal anchor Pith review Pith/arXiv arXiv

  34. [34]

    LLaMA: Open and Efficient Foundation Language Models

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azhar, A. Rodriguez, A. Joulin, E. Grave, and G. Lample, “Llama: Open and efficient foundation language models,” 2023. [Online]. Available: https: //arxiv.org/abs/2302.13971

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  35. [35]

    Mistral 7B

    A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. d. l. Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier, L. R. Lavaud, M.-A. Lachaux, P. Stock, T. L. Scao, T. Lavril, T. Wang, T. Lacroix, and W. E. Sayed, “Mistral 7b,” 2023. [Online]. Available: https://arxiv.org/abs/2310.06825

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  36. [36]

    Qwen Technical Report

    J. Baiet al., “Qwen technical report,” 2023. [Online]. Available: https://arxiv.org/abs/2309.16609

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  37. [37]

    GPT-4 Technical Report

    OpenAIet al., “Gpt-4 technical report,” 2023. [Online]. Available: https://arxiv.org/abs/2303.08774

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  38. [38]

    Evaluation of openai o1: Opportunities and challenges of agi,

    T. Zhonget al., “Evaluation of openai o1: Opportunities and challenges of agi,” 2024. [Online]. Available: https://arxiv.org/abs/2409.18486 15

    work page arXiv 2024

  39. [39]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    DeepSeek-AIet al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,” 2025. [Online]. Available: https://arxiv.org/abs/2501.12948

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  40. [40]

    Cross-lingual attention distillation with personality-informed generative augmentation for multilingual personality recognition,

    J. J. Tan, B.-H. Kwan, D. W.-K. Ng, Y .-C. Hum, N. Kawarazaki, K. Takano, and A. Mokraoui, “Cross-lingual attention distillation with personality-informed generative augmentation for multilingual personality recognition,”IEEE Transactions on Cognitive and Developmental Systems, p. 1–16, 2026. [Online]. Available: http://dx.doi.org/10.1109/TCDS.2026.3682672

    work page doi:10.1109/tcds.2026.3682672 2026

  41. [41]

    Adaptive focal loss with personality stratification for stably mitigating hard class imbalance in multi-dimensional personality recognition,

    J. J. Tan, B.-H. Kwan, D. W.-K. Ng, and Y .-C. Hum, “Adaptive focal loss with personality stratification for stably mitigating hard class imbalance in multi-dimensional personality recognition,”Scientific Reports, vol. 15, no. 1, Nov. 2025. [Online]. Available: http: //dx.doi.org/10.1038/s41598-025-22853-y

    work page doi:10.1038/s41598-025-22853-y 2025

  42. [42]

    Personality classification from text using bidirectional long short-term memory model,

    A. Khattak, N. Jellani, M. Z. Asghar, and U. Asghar, “Personality classification from text using bidirectional long short-term memory model,”Multimedia Tools and Applications, vol. 83, no. 10, p. 28849–28873, Sep. 2023. [Online]. Available: http://dx.doi.org/10.1007/ s11042-023-16661-7

    work page 2023

  43. [43]

    Personality2vec: Network representation learning for personality,

    Z. Guan, B. Wu, B. Wang, and H. Liu, “Personality2vec: Network representation learning for personality,” in2020 IEEE Fifth International Conference on Data Science in Cyberspace (DSC), 2020, pp. 30–37

    work page 2020

  44. [44]

    Pushing on personality detection from verbal behavior: A transformer meets text contours of psycholinguistic features,

    E. Kerz, Y . Qiao, S. Zanwar, and D. Wiechmann, “Pushing on personality detection from verbal behavior: A transformer meets text contours of psycholinguistic features,” inProceedings of the 12th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis, J. Barnes, O. De Clercq, V . Barriere, S. Tafreshi, S. Alqahtani, J. Sedo...

    work page 2022

  45. [45]

    Text-based automatic personality prediction using kgrat-net: a knowledge graph attention network classifier,

    M. Ramezani, M.-R. Feizi-Derakhshi, and M.-A. Balafar, “Text-based automatic personality prediction using kgrat-net: a knowledge graph attention network classifier,”Scientific Reports, vol. 12, no. 1, Dec. 2022. [Online]. Available: http://dx.doi.org/10.1038/s41598-022-25955-z

    work page doi:10.1038/s41598-022-25955-z 2022

  46. [46]

    Transentgat: A sentiment-based lexical psycholinguistic graph attention network for personality prediction,

    S. S. Bajestani, M. M. Khalilzadeh, M. Azarnoosh, and H. R. Kobravi, “Transentgat: A sentiment-based lexical psycholinguistic graph attention network for personality prediction,”IEEE Access, vol. 12, pp. 59 630– 59 642, 2024

    work page 2024

  47. [47]

    Heterogeneous graph contrastive learning with adaptive data augmentation for semi-supervised short text classification,

    M. Wu, Z. Xu, and L. Zheng, “Heterogeneous graph contrastive learning with adaptive data augmentation for semi-supervised short text classification,”Expert Systems, vol. 42, no. 2, Oct. 2024. [Online]. Available: http://dx.doi.org/10.1111/exsy.13744

    work page doi:10.1111/exsy.13744 2024

  48. [48]

    Enhancing textual personality detection toward social media: Integrating long-term and short-term perspectives,

    H. Zhu, X. Zhang, J. Lu, Y . Wu, Z. Bai, C. Min, L. Yang, B. Xu, D. Zhang, and H. Lin, “Enhancing textual personality detection toward social media: Integrating long-term and short-term perspectives,” 2024. [Online]. Available: https://arxiv.org/abs/2404.15067

    work page arXiv 2024

  49. [49]

    Is ChatGPT a Good Personality Recognizer? A Preliminary Study

    Y . Ji, W. Wu, H. Zheng, Y . Hu, X. Chen, and L. He, “Is chatgpt a good personality recognizer? a preliminary study,” 2023. [Online]. Available: https://arxiv.org/abs/2307.03952

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  50. [50]

    Prompting-in-a-series: Psychology-informed contents and embeddings for personality recognition with decoder-only models,

    J. J. Tan, B.-H. Kwan, D. W.-K. Ng, Y .-C. Hum, A. Mokraoui, and S.-Y . Lo, “Prompting-in-a-series: Psychology-informed contents and embeddings for personality recognition with decoder-only models,” IEEE Transactions on Computational Social Systems, p. 1–15, 2025. [Online]. Available: http://dx.doi.org/10.1109/TCSS.2025.3593323

    work page doi:10.1109/tcss.2025.3593323 2025

  51. [51]

    Linguistic styles: Language use as an individual difference

    J. W. Pennebaker and L. A. King, “Linguistic styles: Language use as an individual difference.”Journal of Personality and Social Psychology, vol. 77, pp. 1296–1312, 1999

    work page 1999

  52. [52]

    (mbti) myers-briggs personality type dataset,

    M. J., “(mbti) myers-briggs personality type dataset,” 2017. [Online]. Available: https://www.kaggle.com/datasets/datasnaek/mbti-type

    work page 2017

  53. [53]

    The nomological net of the hexaco model of personality: A large-scale meta-analytic investigation,

    I. Zettler, I. Thielmann, B. E. Hilbig, and M. Moshagen, “The nomological net of the hexaco model of personality: A large-scale meta-analytic investigation,”Perspectives on Psychological Science, vol. 15, no. 3, p. 723–760, Apr. 2020. [Online]. Available: http://dx.doi.org/10.1177/1745691619895036

    work page doi:10.1177/1745691619895036 2020

  54. [54]

    On the comparability of basic personality models: Meta- analytic correspondence, scope, and orthogonality of the big five and hexaco dimensions,

    I. Thielmann, M. Moshagen, B. Hilbig, and I. Zettler, “On the comparability of basic personality models: Meta- analytic correspondence, scope, and orthogonality of the big five and hexaco dimensions,”European Journal of Personality, vol. 36, no. 6, p. 870–900, Jun. 2021. [Online]. Available: http://dx.doi.org/10.1177/08902070211026793

    work page doi:10.1177/08902070211026793 2021

  55. [55]

    Longformer: The Long-Document Transformer

    I. Beltagy, M. E. Peters, and A. Cohan, “Longformer: The long- document transformer,” 2020. [Online]. Available: https://arxiv.org/abs/ 2004.05150

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  56. [56]

    Bottom-up and top-down: Predicting personality with psy- cholinguistic and language model features,

    Y . Mehta, S. Fatehi, A. Kazameini, C. Stachl, E. Cambria, and S. Eetemadi, “Bottom-up and top-down: Predicting personality with psy- cholinguistic and language model features,” in2020 IEEE International Conference on Data Mining (ICDM), 2020, pp. 1184–1189

    work page 2020

  57. [57]

    Psycholinguistic tripartite graph network for personality detection,

    T. Yang, F. Yang, H. Ouyang, and X. Quan, “Psycholinguistic tripartite graph network for personality detection,” inProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), C. Zong, F. Xia, W. Li, and R. Navigli, Eds. Online: A...

    work page 2021

  58. [58]

    Llm vs small model? large language model based text augmentation enhanced personality detection model,

    L. Hu, H. He, D. Wang, Z. Zhao, Y . Shao, and L. Nie, “Llm vs small model? large language model based text augmentation enhanced personality detection model,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 16, pp. 18 234–18 242, Mar

    work page

  59. [59]

    Available: https://ojs.aaai.org/index.php/AAAI/article/ view/29782

    [Online]. Available: https://ojs.aaai.org/index.php/AAAI/article/ view/29782

    work page

  60. [60]

    Orders are unwanted: Dynamic deep graph convolutional network for personality detection,

    T. Yang, J. Deng, X. Quan, and Q. Wang, “Orders are unwanted: Dynamic deep graph convolutional network for personality detection,”

    work page

  61. [61]

    [Online]. Available: https://arxiv.org/abs/2212.01515 AUTHORBIOGRAPHIES Jing Jie Tanreceived his Bachelor of Computer Science (Hons) and PhD in Engineering, specializing in machine learning, from Universiti Tunku Abdul Rahman (UTAR). He is currently a Research Fellow at the National University of Singapore (NUS). His research interests include affective c...

    work page internal anchor Pith review Pith/arXiv arXiv 1989