Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Traits Run Deep: Enhancing Personality Assessment via Psychology-Guided LLM Representations and Multimodal Apparent Behaviors

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

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

Pith's one-line read Trait-specific prompts turn a frozen LLM embedding model into a personality-aware encoder that, fused with audio-visual features, ranks first on the AVI 2025 personality assessment track.

desk verdict A competent challenge-winning system paper whose validation ablations are partly circular and whose single test result is thinner than it looks; worth reviewing, but needs an honesty pass on prompt selection and uncertainty. read the letter →

arxiv 2507.22367 v1 pith:UBXWVF7S submitted 2025-07-30 cs.CL cs.MM

Jia Li , Yichao He , Jiacheng Xu , Tianhao Luo , Zhenzhen Hu , Richang Hong , Meng Wang This is my paper
classification cs.CLcs.MM
keywords PersonalityAssessmentMulti-ModalLearningLargeLanguageModels(LLMs)PromptEngineeringHEXACOAsynchronousVideoInterviewsText-CentricFusionAVIChallenge2025
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper seeks to establish that personality traits, which are stable and often leak through language, face, and body, can be read more accurately when a large language model is steered by psychology-informed prompts tailored to each trait. On the AVI Challenge 2025 dataset, it reports that such prompted LLM text embeddings, fused with audio and visual features by a text-centric network, cut validation mean squared error by about 45 percent relative to the competition baseline (0.1003 versus 0.1796) and achieved a test MSE of 0.12284, ranking first in the Personality Assessment track. The claim matters because it suggests a cheap route to better personality computing: instead of retraining a model, one can steer a frozen LLM with a short task description, the transcribed answer, and the subject's demographic meta information. If correct, this gives practitioners a recipe for extracting trait-relevant semantics from interview data and a fusion architecture that treats text as the anchor rather than treating all modalities equally.

What carries the argument

The load-bearing object is the psychology-informed prompt that converts a general-purpose LLM embedding model into a trait-specific encoder. The prompt concatenates a personality task description targeting one HEXACO trait, the ASR text, and the subject's demographic meta information; the optimal prompt for each trait is selected on validation performance. Around that anchor, the Text-Centric Trait Fusion Network carries the cross-modal argument: a Chunk-Wise Projector splits the high-dimensional text feature into chunks and projects each separately to fight the curse of dimensionality, Cross-Modal Connectors let text attend to audio and video features, a Text-Feature Enhancer computes a gated weighted sum with a residual connection back to text, and an ensemble regression head averages 32 MLPs. This machinery is what lets the paper claim that trait semantics come from text and that audio-visual cues add complementary signal rather than noise.

What would settle it

Run the AVI pipeline again with prompts and hyperparameters frozen before seeing any validation labels, and compare the test MSE to the reported 0.12284; if a random or fixed prompt achieves a statistically indistinguishable test MSE, then the trait-specific prompt selection is not the source of the gain.

Watch

Extended reading notes

Core claim

The paper's central discovery is that personality-relevant text representations are the bottleneck, and the fix is prompt-level: using SFR-Embedding-Mistral with per-trait psychology-informed prompts—composed of a personality task description, the ASR transcript, and the subject's meta information—produces text embeddings that beat both classic BERT-family encoders and the same LLM without prompts on every HEXACO dimension. The paper further claims that anchoring audio and visual features to these text embeddings through Cross-Modal Connectors (text as queries, audio/video as keys and values) and a Text-Feature Enhancer with gated residual fusion yields the best results, and that a Chunk-Wise Projector plus an ensemble of 32 regression heads makes training stable in a data-scarce setting. On the validation set the full pipeline reaches an average MSE of 0.1003, roughly 45 percent below the 0.1796 baseline, and on the AVI 2025 test set it reports 0.12284, the best among competing teams.

Load-bearing premise

The reported 45 percent gain assumes that the per-trait prompts and hyperparameters chosen on the validation set are the right ones for the test set, because the paper reports a single test result with no confidence interval or repeated splits.

Editorial extensions

If this is right

  • Frozen LLM embedding models can be specialized to personality assessment purely through prompting, so the method does not require fine-tuning the encoder.
  • Text is the dominant modality: prompted text alone outperforms every audio-only and video-only feature set on the validation set, and adding audio-visual features improves it further.
  • Treating text as the anchor for cross-modal attention beats naive feature concatenation, which the paper reports gives an average validation MSE of 0.2074 versus 0.1003 for the full fusion.
  • The ensemble regression head stabilizes predictions: the standard deviation of prediction errors across five identical runs drops from 0.0096 to 0.0031.
  • The recipe generalizes across the four HEXACO dimensions used in the challenge, with the largest validation gains on Extraversion and Agreeableness.

Reading between the lines

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

  • A direct test the paper does not run: freeze the per-trait prompts and hyperparameters before looking at the validation set, then compare prompted versus unprompted embeddings on a fresh test split; if the margin collapses, part of the 45 percent gain is selection on validation rather than genuine trait-specific semantics.
  • The same prompting recipe could transfer to other trait taxonomies, such as the Big Five, and to other interview corpora; a cheap falsifiable prediction is that the relative ordering of modalities—text strongest, video weakest—replicates whenever the interview questions are designed to elicit the target traits.
  • Because the prompt includes subject meta information, the method may also pick up demographic confounds; an extension the paper does not address is whether the trait predictions shift when gender, age, or education are removed from the prompt.
  • The reported test result is a single number with no confidence interval; a repeated-split or bootstrap evaluation would tell whether the margin over the runner-up (0.12284 versus 0.13724) is stable.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes Traits Run Deep, a multimodal framework for HEXACO personality assessment from asynchronous video interviews. Text features are extracted with the SFR-Embedding-Mistral LLM guided by psychology-informed, per-trait prompts; audio and visual features come from Emotion2Vec and SigLIP2, respectively. A Text-Centric Trait Fusion Network (Chunk-Wise Projector, Cross-Modal Connector, Text-Feature Enhancer, and an ensemble regression head) fuses the modalities and regresses to personality scores. Experiments on the AVI Challenge 2025 dataset report a validation MSE of 0.1003 (about 45% below the 0.1796 baseline) and a test MSE of 0.12284, which ranks first in the challenge.

Significance. If the reported results are robust, the paper makes a useful contribution by demonstrating that instruction-tuned LLMs with personality-specific prompts can produce text embeddings that improve automatic personality prediction, and that a text-centric fusion architecture can integrate audio and visual cues even with limited training data. The external test-set ranking is a concrete, falsifiable outcome, and the promise of released code strengthens reproducibility. However, the lack of statistical confidence intervals and the use of the validation set for both prompt selection and hyperparameter tuning currently temper the strength of the central claims; the component ablations in particular are likely to be optimistically biased.

major comments (3)
  1. [§3.1.2, §4.2, Table 1, Table 2] The validation set is used both for empirically selecting per-trait prompt variants (Section 3.1.2) and for grid-search hyperparameters such as dropout rates, learning rate, and ensemble settings (Section 4.2), and the component ablations are reported on the same validation set (Tables 1 and 2). Consequently, the claimed "approximately 45% MSE reduction" and the component-level improvements partly reflect fitting discrete prompt choices and hyperparameters to the validation labels, which can overstate generalization. Please report the test-set performance of the ablated variants (or, if not available, state clearly that all ablations are validation-tuned), or perform a nested cross-validation over subjects to provide unbiased component-level estimates.
  2. [§4.4, Table 3] The central test-set claim rests on a single evaluation with no confidence intervals, per-subject error analysis, or repeated runs. The test MSE (0.12284) is 22.5% higher than the validation MSE (0.1003), while the margin over the runner-up is only 0.0144 MSE (about 10.5% relative). For a modest-sized test set, this gap may be within sampling variability. Please provide bootstrap or other confidence intervals over subjects, or repeated evaluations with different seeds and splits, and report per-trait test MSEs so readers can assess the statistical reliability of the ranking.
  3. [§4.1] The dataset description is incomplete. The paper states that ground-truth scores are provided for "Track 1" but does not specify the exact number of training/validation/test subjects, how the split was created, or whether the challenge organizers provide a separate hidden test set. The AVI Challenge 2025 is also not cited. Please supply these details and cite the official challenge description; without them, it is difficult to evaluate whether validation-based selections could influence the reported test result through any shared preprocessing or metadata.
minor comments (6)
  1. [§2 (Related Works)] The second paragraph of Related Works contains an unintelligible, unrelated fragment ("be in trouble for me professionally.my head. I don't want it to affect…") that appears to be a stray ASR transcript; it should be removed.
  2. [Table 3] Table 3 contains extraneous symbols such as "Ἴ61st" and "Ἴ62nd" and several team names are listed as "—"; the table should be cleaned up and all team names or affiliations provided.
  3. [§4.4] There is a typo: "As shown in In Table 3" should read "As shown in Table 3."
  4. [§3.1.1] Arc2Face is cited for face cropping, but the reference describes a face generation model, not a face detector; please clarify which face detection method was actually used.
  5. [§4.3] In Table 1, the variants "Emotion2Vec+" and "Emotion2Vec+ seed" are used but not defined; please specify what the "+" and "seed" suffixes denote.
  6. [Abstract, §4] The abstract mentions "approximately 45% reduction in MSE" without naming the baseline; Section 4.3 states the baseline is 0.1796, but the abstract should also identify it or refer to the relevant table.

Circularity Check

1 steps flagged · score 4.0 of 10

Validation ablations partly reflect prompt/hyperparameter selection on the same validation set, but the test-set ranking is an independent external benchmark; score 4.

  1. fitted input called prediction [Section 3.1.2, Section 4.2, Section 4.3, Table 1]
    "In practice, we experimented with a variety of personality-related prompts and empirically selected the optimal prompt for each trait based on validation performance. ... LLM-Embedding* indicates the variant that utilizes the most effective prompt customized for each personality trait."

    The paper's evidence for the psychology-guided prompts is the validation MSE gain of LLM-Embedding* over LLM-Embedding in Table 1, and it concludes that personality-relevant prompts significantly enhance the model. But LLM-Embedding* is defined as using the most effective prompt, and that prompt was empirically selected based on validation performance. The validation set is therefore used both to choose the prompt and to measure its benefit, making the reported improvement partly a discrete model-selection artifact rather than an independent test of the prompt hypothesis. The grid-search hyperparameter selection in Section 4.2 compounds this for the validation ablations.

full rationale

The paper's central empirical claim is the test-set result in Table 3: HFUT-VisionXL achieves MSE 0.12284 and ranks first on AVI 2025 Track 1. That result is an external challenge evaluation, independent of the authors' prompt and hyperparameter selection, so the headline competitive claim does not reduce to any fitted value. The only circularity-adjacent step is in the validation ablations: the psychology-informed prompt variant LLM-Embedding* is defined as using the prompt that performed best on the validation set, and the same validation set is then used to demonstrate the prompt's benefit. The validation MSE improvement is therefore at least partially an artifact of optimizing the prompt on the evaluation set, rather than an unbiased estimate of prompt effectiveness. The same issue applies to grid-searched hyperparameters. This is a real but partial circularity: it weakens the component-level ablation evidence, but it does not invalidate the independent test-set ranking. No load-bearing self-citation chain was found; the cited embedding models and encoders are external prior work, and no uniqueness theorem or ansatz is imported from the authors' own prior publications. The novelty claim is a claim of priority, not a derived result, so it does not enter the circularity analysis.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The reported gains rest on several validation-fitted choices and domain assumptions about personality expression and LLM transfer. No new physical or theoretical entities are introduced.

free parameters (5)
  • Per-trait personality prompt (discrete prompt variant) = Not disclosed; one optimal prompt per trait selected on AVI validation
    Section 3.1.2 states prompts were 'empirically selected... based on validation performance'; the validation-set MSE gains in Table 1 depend on this choice.
  • Dropout rates = 0.2 CWP, 0.3 CMC, 0.1 TFE
    Section 4.2 gives initial values and says grid search selects optimal values; these affect final MSE.
  • Ensemble head size = 32 subnetworks
    Equation 12 fixes the average over 32 MLPs; the paper does not justify 32 over other sizes.
  • Learning rate, batch size, epochs = 1e-4, 32, 200
    Set in Section 4.2 and grid-searched; not derived from data size or theory.
  • Chunk-Wise Projector dimensions N and d' = Not reported
    Equations 1 to 3 define chunks but the number of chunks and projected dimension are omitted; they are likely tuned on validation.
assumptions (5)
  • domain assumption AVI 2025 ground-truth labels are valid HEXACO trait scores.
    The paper uses dataset labels as targets without validating them against external psychological measurement.
  • domain assumption Personality traits are stable and are expressed in language, face, and body in asynchronous patterns.
    Introductory claim in Section 1 that motivates the multimodal setup; no evidence is supplied in this paper.
  • domain assumption Instruction-following LLM embeddings with personality prompts capture trait-relevant semantics better than generic embeddings.
    Section 3.1.2 assumes SFR-Embedding-Mistral plus prompts yields high-level personality semantics; the empirical comparison supports but does not prove transfer.
  • domain assumption Mean pooling of token embeddings preserves the information needed for regression.
    Figure 1 shows mean pooling for LLM text features; no comparison to other pooling strategies is reported.
  • domain assumption Frozen pretrained encoders (Whisper-small, Emotion2Vec, SigLIP2) provide transferable features in this domain.
    Section 3.1.1 freezes all encoders; the paper does not test whether fine-tuning or other encoders would change results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Traits Run Deep: Enhancing Personality Assessment via Psychology-Guided LLM Representations and Multimodal Apparent Behaviors." pith.science (2026). https://pith.science/paper/UBXWVF7S

@misc{pith2026250722367,
  author       = {Pith},
  title        = {Pith review of: Traits Run Deep: Enhancing Personality Assessment via Psychology-Guided LLM Representations and Multimodal Apparent Behaviors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UBXWVF7S}},
  note         = {Machine review of arXiv:2507.22367}
}
read the original abstract

Accurate and reliable personality assessment plays a vital role in many fields, such as emotional intelligence, mental health diagnostics, and personalized education. Unlike fleeting emotions, personality traits are stable, often subconsciously leaked through language, facial expressions, and body behaviors, with asynchronous patterns across modalities. It was hard to model personality semantics with traditional superficial features and seemed impossible to achieve effective cross-modal understanding. To address these challenges, we propose a novel personality assessment framework called \textit{\textbf{Traits Run Deep}}. It employs \textit{\textbf{psychology-informed prompts}} to elicit high-level personality-relevant semantic representations. Besides, it devises a \textit{\textbf{Text-Centric Trait Fusion Network}} that anchors rich text semantics to align and integrate asynchronous signals from other modalities. To be specific, such fusion module includes a Chunk-Wise Projector to decrease dimensionality, a Cross-Modal Connector and a Text Feature Enhancer for effective modality fusion and an ensemble regression head to improve generalization in data-scarce situations. To our knowledge, we are the first to apply personality-specific prompts to guide large language models (LLMs) in extracting personality-aware semantics for improved representation quality. Furthermore, extracting and fusing audio-visual apparent behavior features further improves the accuracy. Experimental results on the AVI validation set have demonstrated the effectiveness of the proposed components, i.e., approximately a 45\% reduction in mean squared error (MSE). Final evaluations on the test set of the AVI Challenge 2025 confirm our method's superiority, ranking first in the Personality Assessment track. The source code will be made available at https://github.com/MSA-LMC/TraitsRunDeep.

Figures

Figures reproduced from arXiv: 2507.22367 by the authors.

Figure 1
Figure 1. A Comparative Analysis of Traditional Text Feature [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The architecture of our Traits Run Deep framework consists of two main components: (1) Modality-Specific Feature [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison of validation loss between the Chunk [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. A Survey of Large Language Models for Perception and Measurement of Human Psychology

    cs.CY 2026-05 unverdicted novelty 5.0 of 10

    A survey proposing a three-pillar framework to evaluate LLMs as tools for measuring latent psychological constructs and reviewing applications in personality and mental health.

Reference graph

Works this paper leans on

50 extracted references · 33 canonical work pages · cited by 1 Pith paper

  1. [1]

    Mike Ashton and Kibeom Lee. 2009. HEXACO personality inventory-revised. Journal of Personality Assessment (2009)

    work page 2009

  2. [2]

    Joan-Isaac Biel and Daniel Gatica-Perez. 2012. The youtube lens: Crowdsourced personality impressions and audiovisual analysis of vlogs. IEEE Transactions on Multimedia 15, 1 (2012), 41–55

    work page 2012

  3. [3]

    Cong Cai, Shan Liang, Xuefei Liu, Kang Zhu, Zhengqi Wen, Jianhua Tao, Heng Xie, Jizhou Cui, Yiming Ma, Zhenhua Cheng, et al. 2024. Mdpe: A multimodal deception dataset with personality and emotional characteristics. arXiv preprint arXiv:2407.12274 (2024)

    work page Pith review arXiv 2024

  4. [4]

    David F Caldwell and Jerry M Burger. 1998. Personality characteristics of job applicants and success in screening interviews. Personnel Psychology 51, 1 (1998), 119–136

    work page 1998

  5. [5]

    Laura Caspari, Kanishka Ghosh Dastidar, Saber Zerhoudi, Jelena Mitrovic, and Michael Granitzer. 2024. Beyond benchmarks: Evaluating embedding model simi- larity for retrieval augmented generation systems.arXiv preprint arXiv:2407.08275 (2024)

    arXiv 2024

  6. [6]

    Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. 2024. Scaling instruction-finetuned language models. Journal of Machine Learning Research 25, 70 (2024), 1–53

    work page 2024

  7. [7]

    Paul T Costa and Robert R McCrae. 2008. The revised neo personality inventory (neo-pi-r). The SAGE handbook of personality theory and assessment 2, 2 (2008), 179–198

    work page 2008

  8. [8]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers). 4171–4186

    2019

Show all 50 references
  1. [9]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv prepri...

    2020 arXiv

  2. [10]

    Hugo Jair Escalante, Heysem Kaya, Albert Ali Salah, Sergio Escalera, Yağmur Güçlütürk, Umut Güçlü, Xavier Baró, Isabelle Guyon, Julio CS Jacques Junior, Meysam Madadi, et al. 2020. Modeling, recognizing, and explaining apparent personality from videos. IEEE Transactions on Aff...

    2020

  3. [11]

    Hugo Jair Escalante, Heysem Kaya, Albert Ali Salah, Sergio Escalera, Yağmur Güçlütürk, Umut Güçlü, Xavier Baró, Isabelle Guyon, Julio C. S. Jacques Junior, Meysam Madadi, Stephane Ayache, Evelyne Viegas, Furkan Gürpınar, Achmad- noer Sukma Wicaksana, Cynthia C. S. Liem, Marcel...

    2022

  4. [12]

    Andrea Fossati, Serena Borroni, Donatella Marchione, and Cesare Maffei. 2011. The big five inventory (BFI). European Journal of Psychological Assessment (2011)

    2011

  5. [13]

    Sina Ghassemi, Tianyi Zhang, Ward van Breda, Antonis Koutsoumpis, Janneke K Oostrom, Djurre Holtrop, and Reinout E de Vries. 2023. Unsupervised multi- modal learning for dependency-free personality recognition. IEEE transactions on affective computing 15, 3 (2023), 1053–1066

    2023

  6. [14]

    Yağmur Güçlütürk, Umut Güçlü, Marcel AJ van Gerven, and Rob van Lier. 2016. Deep impression: Audiovisual deep residual networks for multimodal apparent personality trait recognition. In European conference on computer vision . Springer, 349–358

    2016

  7. [15]

    van Gerven, and Rob van Lier

    Yağmur Güçlütürk, Umut Güçlü, Xavier Baró, Hugo Jair Escalante, Isabelle Guyon, Sergio Escalera, Marcel A.J. van Gerven, and Rob van Lier. 2018. Multimodal First Impression Analysis with Deep Residual Networks. IEEE Transactions on Affective Computing 9, 3 (2018), 316–329. doi...

    2018

  8. [16]

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick

  9. [17]

    Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. arXiv preprint arXiv:2111.09543 (2021)

    2021 arXiv

  10. [18]

    Léo Hemamou, Arthur Guillon, Jean-Claude Martin, and Chloé Clavel. 2021. Mul- timodal hierarchical attention neural network: Looking for candidates behaviour which impact recruiter’s decision. IEEE Transactions on Affective Computing 14, 2 (2021), 969–985

    2021

  11. [19]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, De- vendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thoma...

    2023 arXiv

  12. [20]

    Julio CS Jacques Junior, Yağmur Güçlütürk, Marc Pérez, Umut Güçlü, Carlos Andujar, Xavier Baró, Hugo Jair Escalante, Isabelle Guyon, Marcel AJ Van Gerven, Rob Van Lier, et al. 2019. First impressions: A survey on vision-based apparent personality trait analysis. IEEE Transacti...

    2019

  13. [21]

    SV Kedar and DS Bormane. 2015. Automatic personality assessment: A systematic review. In 2015 International Conference on Information Processing (ICIP) . IEEE, 326–331

    2015

  14. [22]

    Antonis Koutsoumpis, Sina Ghassemi, Janneke K Oostrom, Djurre Holtrop, Ward van Breda, Tianyi Zhang, and Reinout E de Vries. 2024. Beyond traditional inter- views: Psychometric analysis of asynchronous video interviews for personality and interview performance evaluation using...

    2024

  15. [23]

    Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. 2019. Albert: A lite bert for self-supervised learning of language representations. arXiv preprint arXiv:1909.11942 (2019)

    2019 arXiv

  16. [24]

    Rongfan Liao, Siyang Song, and Hatice Gunes. 2024. An open-source benchmark of deep learning models for audio-visual apparent and self-reported personality recognition. IEEE Transactions on Affective Computing 15, 3 (2024), 1590–1607

    2024

  17. [25]

    Cynthia CS Liem, Markus Langer, Andrew Demetriou, Annemarie MF Hiemstra, Achmadnoer Sukma Wicaksana, Marise Ph Born, and Cornelius J König. 2018. Psychology meets machine learning: Interdisciplinary perspectives on algorith- mic job candidate screening. In Explainable and inte...

    2018

  18. [26]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 (2019)

    2019 arXiv

  19. [27]

    Zhen-Tao Liu, Abdul Rehman, Min Wu, Wei-Hua Cao, and Man Hao. 2020. Speech personality recognition based on annotation classification using log-likelihood distance and extraction of essential audio features. IEEE transactions on multime- dia 23 (2020), 3414–3426

    2020

  20. [28]

    Jianming Lv, Chujie Chen, and Zequan Liang. 2023. Automated Scoring of Asynchronous Interview Videos Based on Multi-modal Window-Consistency Fusion. IEEE Transactions on Affective Computing (2023)

    2023

  21. [29]

    Ziyang Ma, Zhisheng Zheng, Jiaxin Ye, Jinchao Li, Zhifu Gao, Shiliang Zhang, and Xie Chen. 2023. emotion2vec: Self-supervised pre-training for speech emotion representation. arXiv preprint arXiv:2312.15185 (2023)

    2023 arXiv

  22. [30]

    François Mairesse, Marilyn A Walker, Matthias R Mehl, and Roger K Moore. 2007. Using linguistic cues for the automatic recognition of personality in conversation and text. Journal of artificial intelligence research 30 (2007), 457–500

    2007

  23. [31]

    Dena F Mujtaba and Nihar R Mahapatra. 2021. Multi-task deep neural networks for multimodal personality trait prediction. In 2021 international conference on computational science and computational intelligence (CSCI) . IEEE, 85–91

    2021

  24. [32]

    Cristina Palmero, Javier Selva, Sorina Smeureanu, Julio Junior, Jacques CS, Albert Clapés, Alexa Moseguí, Zejian Zhang, David Gallardo, Georgina Guilera, et al

  25. [33]

    Foivos Paraperas Papantoniou, Alexandros Lattas, Stylianos Moschoglou, Jiankang Deng, Bernhard Kainz, and Stefanos Zafeiriou. 2024. Arc2face: A foun- dation model for id-consistent human faces. In European Conference on Computer Vision. Springer, 241–261

    2024

  26. [34]

    Víctor Ponce-López, Baiyu Chen, Marc Oliu, Ciprian Corneanu, Albert Clapés, Isabelle Guyon, Xavier Baró, Hugo Jair Escalante, and Sergio Escalera. 2016. Chalearn lap 2016: First round challenge on first impressions-dataset and results. In European conference on computer vision...

    2016

  27. [35]

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. Robust speech recognition via large-scale weak supervision. In International conference on machine learning . PMLR, 28492–28518

    2023

  28. [36]

    Brent W Roberts. 2009. Back to the future: Personality and assessment and personality development. Journal of research in personality 43, 2 (2009), 137–145

    2009

  29. [37]

    Christopher J Soto and Oliver P John. 2017. The next Big Five Inventory (BFI-2): Developing and assessing a hierarchical model with 15 facets to enhance band- width, fidelity, and predictive power. Journal of personality and social psychology 113, 1 (2017), 117

    2017

  30. [38]

    Arulkumar Subramaniam, Vismay Patel, Ashish Mishra, Prashanth Balasubra- manian, and Anurag Mittal. 2016. Bi-modal first impressions recognition using temporally ordered deep audio and stochastic visual features. In European confer- ence on computer vision . Springer, 337–348

    2016

  31. [39]

    Chanchal Suman, Sriparna Saha, Aditya Gupta, Saurabh Kumar Pandey, and Pushpak Bhattacharyya. 2022. A multi-modal personality prediction system. Knowledge-Based Systems 236 (2022), 107715

    2022

  32. [40]

    Hao Tan and Mohit Bansal. 2019. Lxmert: Learning cross-modality encoder representations from transformers. arXiv preprint arXiv:1908.07490 (2019)

    2019 arXiv

  33. [41]

    Eggi Farkhan Tsani and Derwin Suhartono. 2023. Personality identification from social media using ensemble BERT and RoBERTa. Informatica 47, 4 (2023)

    2023

  34. [42]

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, et al. 2025. Siglip 2: Multilingual vision-language encoders Conference acronym ’XX, June 03–05, 2018, Woodstoc...

    2025 arXiv

  35. [43]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017)

    2017

  36. [44]

    Alessandro Vinciarelli and Gelareh Mohammadi. 2014. A survey of personality computing. IEEE Transactions on Affective Computing 5, 3 (2014), 273–291

    2014

  37. [45]

    Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2023. Improving text embeddings with large language models. arXiv preprint arXiv:2401.00368 (2023)

    2023 arXiv

  38. [46]

    Xiu-Shen Wei, Chen-Lin Zhang, Hao Zhang, and Jianxin Wu. 2017. Deep bimodal regression of apparent personality traits from short video sequences. IEEE Transactions on Affective Computing 9, 3 (2017), 303–315

    2017

  39. [47]

    Tianyi Zhang, Antonis Koutsoumpis, Janneke K Oostrom, Djurre Holtrop, Sina Ghassemi, and Reinout E De Vries. 2024. Can large language models assess personality from asynchronous video interviews? A comprehensive evaluation of validity, reliability, fairness, and rating pattern...

    2024

  40. [48]

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems 36 (2023), 46595–46623. Rece...

    2023

  41. [2021]

    In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Context-aware personality inference in dyadic scenarios: Introducing the udiva dataset. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 1–12

  42. [2022]

    In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 16000–16009

Pith tools

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