Pith. sign in

REVIEW 3 major objections 6 minor 59 references

Personality Prediction from Life Stories using Language Models

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 A two-step model of fine-tuned RoBERTa plus an attention RNN predicts five personality traits from long life-story interviews with $R^2$ values of 0.30 to 0.52, ahead of much larger models at a fraction of the compute.

desk verdict Valuable two-step architecture for long-text personality prediction, but the unstated train/test split unit makes the headline numbers unverifiable. read the letter →

arxiv 2506.19258 v1 pith:2FFRDTYT submitted 2025-06-24 cs.CL cs.LG

Rasiq Hussain , Jerry Ma , Rithik Khandelwal , Joshua Oltmanns , Mehak Gupta This is my paper
classification cs.CLcs.LG
keywords personalitypredictionFive-FactorModellong-narrativeinterviewssliding-windowfine-tuningRNNattentioninterpretabilityRoBERTalifestories
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 tries to establish that personality can be read from long transcribed life narratives—interviews averaging more than 2,000 tokens—more accurately and far more cheaply by a two-step hybrid than by any single large language model. Step one fine-tunes RoBERTa on overlapping 512-token windows to predict the five-factor personality scores; step two freezes that model and trains a two-layer GRU with attention on the ordered sequence of window embeddings, so the final prediction sees the whole narrative. In five-fold cross-validation the authors report $R^2$ from 0.30 to 0.52 across the five traits, the lowest mean squared error and highest $R^2$ among fine-tuned RoBERTa, XLM-RoBERTa, XLNet, Longformer, and LLaMA-3.1-8B, at roughly one-fortieth of LLaMA's compute. The paper also argues that the attention weights make the model interpretable: the windows it weights most heavily match what psychologists expect for each trait, and removing those windows lowers the predicted score. A sympathetic reader should care because it points toward language-based personality assessment that is both accurate enough for research use and transparent enough for clinical review.

What carries the argument

The load-bearing mechanism is the stop-gradient, two-stage pipeline. A sliding window of size $w$ and stride $s$ cuts each transcript into overlapping 512-token segments; RoBERTa-L is fine-tuned on these segments with an MSE loss to predict the five scores, then frozen. Each window's [CLS] token yields a 1024-dimensional embedding, and the ordered embedding sequence is fed to a two-layer GRU (hidden size 256) whose hidden states are combined by a softmax attention weight $\alpha_t = \exp(w^\top h_t) / \sum_{j} \exp(w^\top h_j)$ into a context vector $c_i$; a fully connected layer maps that vector to the predicted scores. Because gradients during the second stage flow only through the RNN, the language model's representations stay stable and the attention weights give a per-window explanation of the prediction.

What would settle it

Re-run the five-fold evaluation twice—once with all windows of a single transcript confined to the same fold and once with windows randomly assigned to folds—and report the sliding-window stride; if the random-window version gives materially higher $R^2$ than the transcript-level version, overlapping text leaked across the split and inflated the reported numbers.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that decoupling representation learning from sequence modeling is what makes long-narrative personality prediction work. After RoBERTa is fine-tuned on sliding windows and frozen, its per-window [CLS] embeddings already carry trait-relevant local cues, and a two-layer GRU with additive attention reassembles those cues into a full-narrative context vector that drives the regression head. The authors report that this two-step 'fine-tune, freeze, aggregate' design gives the minimum MSE and highest $R^2$ across all five traits when compared with five fine-tuned baselines and five ablations; the continuous-flow variants that train RoBERTa and the aggregator jointly converge worse, and replacing the RNN with a feed-forward network or transformer does not match its accuracy. The attention weights double as an interpretability mechanism: trait-relevant content can appear anywhere in the transcript, the model locates it, and topic analysis of the highest-attention windows aligns with established personality psychology.

Load-bearing premise

The reported accuracy rests on the assumption that the 80:20 train/test split was done per interview rather than per text window, so no person's words appear in both training and testing, yet the paper does not state the split unit or the spacing between successive windows, and a window-level split would let overlapping text leak across folds and inflate the $R^2$ values.

Editorial extensions

If this is right

  • A 335-million-parameter model with a lightweight RNN can outperform an 8-billion-parameter LLM on long-narrative regression while using roughly 40 times fewer FLOPs per sample, making clinical-scale personality assessment computationally feasible.
  • The fine-tune-then-freeze-then-aggregate recipe is a general template for any long-document prediction task where inputs exceed a language model's context window and interpretability matters.
  • Because attention is computed at the window level, practitioners can trace a predicted score to specific life-story topics, which is the transparency that questionnaire-based or black-box LLM assessment lacks.
  • The reported $R^2$ values exceed earlier language-based personality prediction from social media posts, suggesting richer, longer narratives carry more trait signal than short status updates.
  • Aggregating mean window embeddings already improves prediction over a single window, so the personality signal in life stories is distributed across the narrative, and the RNN's ordering adds further accuracy.

Reading between the lines

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

  • If the train/test split turns out to have been made at the window level rather than the transcript level, the headline $R^2$ values should be treated as upper bounds; a transcript-level re-evaluation is the natural first check.
  • The attention-drop and topic analyses are suggestive rather than causal; a stricter test would erase matched control windows across all test transcripts to measure the true contribution of high-attention segments.
  • The efficiency comparison depends on the specific configuration of the large-model baselines, particularly LoRA fine-tuning of LLaMA; other ways of adapting an LLM might narrow the gap.
  • The same two-step architecture could be probed on other long-form clinical texts, such as psychiatric intake narratives, where a transparent per-segment explanation is as important as the score itself.
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 addresses personality prediction from transcribed life-narrative interviews (average over 2,000 tokens) using a two-step architecture: first fine-tune a 512-token transformer (RoBERTa-Large) on sliding windows with a regression head, freeze it, and then train a GRU with attention over the resulting per-window [CLS] embeddings to produce Five-Factor Model scores. The authors compare this method with fine-tuned RoBERTa, XLM-RoBERTa, XLNet, Longformer, and LLaMA-3.1-8B with LoRA, and report MSE/R2 over five folds, showing the proposed method with the lowest MSE and highest R2 for all five traits (R2 0.30-0.52). They also present ablations (continuous flow vs. two-step, RNN vs. transformer vs. feed-forward aggregator, fine-tuned vs. frozen embeddings), efficiency comparisons, attention-based interpretability with expert-reviewed thematic validation, attention-removal score drops, and BERTopic topic correlations.

Significance. If the reported results are valid, the paper makes a useful practical contribution: a relatively lightweight, interpretable alternative to long-context LLMs for narrative-level psychological assessment, with R2 values well above those reported for social-media personality prediction. The strengths of the paper include a moderately large real clinical dataset (1,408 older adults), a clear ablation isolating the effect of the two-step training and the RNN aggregator, an explicit efficiency comparison, and a serious attempt at interpretability validation through domain-expert review and attention-removal experiments. The central technical risk is the experimental protocol: the split unit and sliding-window parameters are not specified, and without a transcript-level split the reported improvements could be inflated by leakage across overlapping windows.

major comments (3)
  1. [Section 5] The evaluation protocol does not state the unit of the train/test split. The text says only "An 80:20 train:test split was applied and kept consistent across both stages of the two-step method," and elsewhere says 5-fold cross-validation was used. Because Section 4.3 generates overlapping sliding windows from each transcript, a split performed at the window level would place segments of the same interview in both training and test sets. The RNN and the fine-tuned RoBERTa could then memorize speaker-specific phrasing, inflating the R2 values in Tables 2 and 3. The authors must state explicitly that the split is at the transcript/participant level, and if it was not, re-run the experiments with grouped splits. This is load-bearing because the headline claim in Section 5.2 rests on those numbers.
  2. [Section 4.3] The sliding-window parameters w and s are introduced only symbolically in the equations, and their concrete values are never reported. The stride determines the degree of overlap between windows, which affects both the effective number of segments per transcript and the potential for information leakage between train and test folds. The authors should report the actual values used for w, s, and the resulting maximum sequence length of 200 windows.
  3. [Tables 2 and 3] The central comparative claim is made from five-fold means with standard deviations, but no significance tests are reported. Some differences are modest (e.g., Openness R2: 0.19 for RoBERTa-L vs. 0.30 for FT-RoBERTa+RNN), and the per-fold structure of cross-validation permits paired tests. The authors should add paired per-fold comparisons or equivalent significance testing before claiming that the proposed method achieves the minimum MSE and highest R2 across all traits.
minor comments (6)
  1. [Section 4.1] The text says transcripts average 2,992 tokens, while Table 1 reports a mean length of 2,513 words; the relationship between the two numbers (e.g., tokenizer or word/token conversion) should be clarified.
  2. [Equation (3)] The attention formula uses w^T h_t, but the learned parameter w is not defined in the surrounding text; please define it explicitly.
  3. [Section 6.2 / Figure 3] The score-drop percentages (including the negative drop for one neuroticism transcript and the 80% drop for agreeableness) are not defined precisely; the text should state the baseline used and whether negative values correspond to an increase after removal.
  4. [Section 5.3] The ridge regression R2 values in Figure 2 appear to be computed on the training set; the text should say explicitly that these are in-sample numbers so that readers do not mistake them for held-out performance.
  5. [Section 5] The paper mentions both 5-fold cross-validation and an 80:20 split without explaining how they relate; clarify whether each fold uses an 80/20 split or whether the 80:20 split is a separate final evaluation.
  6. [Table 1] The caption says the table shows "length of transcripts in words and kb," but only a single length row is present; correct the caption or add the requested units.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all reported predictions are supervised regressions against external NEO-PI-R labels, and no fitted constant or self-citation is relabeled as a predicted result.

full rationale

I walked the paper's derivation chain from data through the two-step model to the comparisons in Tables 2 and 3. The target variables are external NEO-PI-R scores, and every reported MSE/R2 value compares model outputs against those external labels. The two-step architecture (Eqs. 1-3) is a standard stacked supervised pipeline: RoBERTa is fine-tuned on windowed transcripts with MSE loss, frozen, and its [CLS] embeddings are aggregated by a GRU with attention trained on the same external labels. No quantity that is fitted is later renamed as a prediction; the RNN is not trained on residuals or on RoBERTa's own outputs as labels. The interpretability analyses use the model's attention weights, but validate them against external psychological themes and domain-expert topics, not against the model's own predictions. The only potentially author-overlapping reference is Widiger and Oltmanns (2017), cited for the substantive claim that neuroticism is associated with emotional struggles; it is not load-bearing for the method or for any comparison. The paper explicitly attributes its hierarchical transformer-plus-RNN design to prior work (Pappagari et al. 2019; Lynn et al. 2020), so no result is smuggled in via self-citation. I found no equation or construction that equates an output to an input. Concerns about the unstated split unit and unreported window stride are experimental-validity risks, not circularity, and do not affect this score.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method's performance depends on the validity of self-reported FFM scores, on 512-token windows being informative enough to supervise the first stage, and on the train/test split being done at the participant level. The listed hyperparameters (window size, stride, GRU hidden size, LoRA settings) are chosen for this dataset rather than derived.

free parameters (4)
  • Sliding-window size = 512 tokens
    Set by RoBERTa's maximum sequence length, not tuned.
  • Sliding-window stride = not reported
    Determines overlap and number of embeddings per transcript (up to 200), but is never specified.
  • GRU hidden size = 256
    Hyperparameter for the RNN aggregator; no tuning curve is given.
  • LoRA rank and alpha for LLaMA = rank 64, alpha 16
    Fixed PEFT hyperparameters for the LLaMA baseline.
assumptions (4)
  • domain assumption NEO-PI-R self-report scores are treated as ground truth personality labels
    All models are trained and evaluated against these scores; the paper acknowledges self-report biases in Section 9.
  • domain assumption Each 512-token window of a transcript is informative enough to be supervised by the transcript-level FFM score
    First-stage fine-tuning assigns the full interview's score to every window, a noisy but necessary premise.
  • ad hoc to paper The train/test split isolates participants so no overlapping windows leak across folds
    Section 5 does not state the split unit; results depend on transcript-level splitting.
  • standard math Standard GRU recurrence and softmax attention are valid sequence models
    Equations 1-3 use standard operations; no proof is needed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Personality Prediction from Life Stories using Language Models." pith.science (2026). https://pith.science/paper/2FFRDTYT

@misc{pith2026250619258,
  author       = {Pith},
  title        = {Pith review of: Personality Prediction from Life Stories using Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2FFRDTYT}},
  note         = {Machine review of arXiv:2506.19258}
}
read the original abstract

Natural Language Processing (NLP) offers new avenues for personality assessment by leveraging rich, open-ended text, moving beyond traditional questionnaires. In this study, we address the challenge of modeling long narrative interview where each exceeds 2000 tokens so as to predict Five-Factor Model (FFM) personality traits. We propose a two-step approach: first, we extract contextual embeddings using sliding-window fine-tuning of pretrained language models; then, we apply Recurrent Neural Networks (RNNs) with attention mechanisms to integrate long-range dependencies and enhance interpretability. This hybrid method effectively bridges the strengths of pretrained transformers and sequence modeling to handle long-context data. Through ablation studies and comparisons with state-of-the-art long-context models such as LLaMA and Longformer, we demonstrate improvements in prediction accuracy, efficiency, and interpretability. Our results highlight the potential of combining language-based features with long-context modeling to advance personality assessment from life narratives.

Figures

Figures reproduced from arXiv: 2506.19258 by the authors.

Figure 1
Figure 1. RoBERTa+RNN architecture. Dashed arrows show backpropagation with 1) fine-tuning RoBERTa with sliing-windows on FFM scores and 2) training RNN with embeddings from fine-tuned RoBERTa. 3 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. 2-D visualization of [CLS] embeddings of [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. It shows attention given to different text windows in the transcript for two random examples from the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Context Anywhere: Attention plots of one [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 47 canonical work pages

  1. [1]

    Shengnan An, Zexiong Ma, Zeqi Lin, Nanning Zheng, and Jian-Guang Lou. 2024. Make your llm fully utilize the context. arXiv preprint arXiv:2404.16811

    arXiv 2024

  2. [2]

    Shlomo Argamon, Moshe Koppel, James W Pennebaker, and Jonathan Schler. 2007. Mining the blogosphere: Age, gender and the varieties of self-expression. First Monday

    work page 2007

  3. [3]

    Jonah Berger and Grant Packard. 2022. Using natural language processing to understand people and culture. American Psychologist, 77(4):525

    work page 2022

  4. [4]

    Ryan L Boyd and H Andrew Schwartz. 2021. Natural language analysis and the psychology of verbal behavior: The past, present, and future states of the field. Journal of Language and Social Psychology, 40(1):21--41

    work page 2021

  5. [5]

    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):179--198

    work page 2008

  6. [6]

    Brenda Curtis, Salvatore Giorgi, Anneke EK Buffone, Lyle H Ungar, Robert D Ashford, Jessie Hemmons, Dan Summers, Casey Hamilton, and H Andrew Schwartz. 2018. Can twitter be used to predict county excessive alcohol consumption rates? PloS one, 13(4):e0194290

    work page 2018

  7. [7]

    Cutler and D.M

    A. Cutler and D.M. Condon. 2023. Deep lexical hypothesis: Identifying personality structure in natural language. Journal of Personality and Social Psychology, 125(1):173--197

    work page 2023

  8. [8]

    Z Dai, Z Yang, Y Yang, J Carbonell, Q Le, and R Transformer-XL Salakhutdinov. 2019. Attentive language models beyond a fixed-length context. 2019. ACL

    work page 2019

Show all 59 references
  1. [9]

    Luna De Bruyne, Pepa Atanasova, and Isabelle Augenstein. 2022. Joint emotion label space modeling for affect lexica. Computer Speech & Language, 71:101257

    2022

  2. [10]

    Ming Ding, Chang Zhou, Hongxia Yang, and Jie Tang. 2020 a . Cogltx: Applying bert to long texts. Advances in Neural Information Processing Systems, 33:12792--12804

    2020

  3. [11]

    Siyu Ding, Junyuan Shang, Shuohuan Wang, Yu Sun, Hao Tian, Hua Wu, and Haifeng Wang. 2020 b . Ernie-doc: A retrospective long-document modeling transformer. arXiv preprint arXiv:2012.15688

    2020 arXiv

  4. [12]

    Johannes C Eichstaedt, Margaret L Kern, David B Yaden, H Andrew Schwartz, Salvatore Giorgi, Gregory Park, Courtney A Hagan, Victoria A Tobolsky, Laura K Smith, Anneke Buffone, et al. 2021. Closed-and open-vocabulary approaches to text analysis: A review, quantitative compariso...

    2021

  5. [13]

    Johannes C Eichstaedt, Robert J Smith, Raina M Merchant, Lyle H Ungar, Patrick Crutchley, Daniel Preo t iuc-Pietro, David A Asch, and H Andrew Schwartz. 2018. Facebook language predicts depression in medical records. Proceedings of the National Academy of Sciences, 115(44):112...

    2018

  6. [14]

    Adithya V Ganesan, Yash Kumar Lal, August H kan Nilsson, and H Andrew Schwartz. 2023 a . Systematic evaluation of gpt-3 for zero-shot personality estimation. arXiv preprint arXiv:2306.01183

    2023 arXiv

  7. [15]

    Adithya V Ganesan, Matthew Matero, Aravind Reddy Ravula, Huy Vu, and H Andrew Schwartz. 2021. Empirical evaluation of pre-trained transformers for human-level nlp: The role of sample size and dimensionality. In Proceedings of the conference. Association for Computational Lingu...

    2021

  8. [16]

    Ganesan et al

    K. Ganesan et al. 2023 b . Zero-shot personality assessment using large language models. In Proc. AAAI Conf. Artificial Intelligence

    2023

  9. [17]

    Binzong Geng, Zhaoxin Huan, Xiaolu Zhang, Yong He, Liang Zhang, Fajie Yuan, Jun Zhou, and Linjian Mo. 2024. Breaking the length barrier: Llm-enhanced ctr prediction in long textual user behaviors. In Proceedings of the 47th International ACM SIGIR Conference on Research and De...

    2024

  10. [18]

    Hongyu Gong, Yelong Shen, Dian Yu, Jianshu Chen, and Dong Yu. 2020. Recurrent chunking mechanisms for long-text machine reading comprehension. arXiv preprint arXiv:2005.08056

    2020 arXiv

  11. [19]

    William G Graziano and Ren \'e e M Tobin. 2017. Agreeableness and the five factor model. The Oxford handbook of the five factor model, 1:105--131

    2017

  12. [20]

    Derek R Hatfield and Benjamin M Ogles. 2004. The use of outcome measures by psychologists in clinical practice. Professional Psychology: Research and Practice, 35(5):485

    2004

  13. [21]

    Rumen Iliev, Morteza Dehghani, and Eyal Sagi. 2015. Automated text analysis in psychology: Methods, applications, and future developments. Language and cognition, 7(2):265--290

    2015

  14. [22]

    Joshua Conrad Jackson, Joseph Watts, Johann-Mattis List, Curtis Puryear, Ryan Drabble, and Kristen A Lindquist. 2022. From text to thought: How analyzing language can advance psychological science. Perspectives on Psychological Science, 17(3):805--826

    2022

  15. [23]

    Joshua J Jackson and Brent W Roberts. 2015. Conscientiousness. Journal Name

    2015

  16. [24]

    Dipika Jain, Akshi Kumar, and Rohit Beniwal. 2022. Personality bert: A transformer-based model for personality detection from textual data. In Proceedings of international conference on computing and communication networks: ICCCN 2021, pages 515--522. Springer

    2022

  17. [25]

    Archa Joshy and Sumod Sundar. 2022. Analyzing the performance of sentiment analysis using bert, distilbert, and roberta. In 2022 IEEE international power and renewable energy conference (IPRECON), pages 1--6. IEEE

    2022

  18. [26]

    Oscar Kjell, Daiva Daukantait \.e , and Sverker Sikstr \"o m. 2021. Computational language assessments of harmony in life—not satisfaction with life or rating scales—correlate with cooperative behaviors. Frontiers in psychology, 12:601679

    2021

  19. [27]

    Oscar NE Kjell, Sverker Sikstr \"o m, Katarina Kjell, and H Andrew Schwartz. 2022. Natural language analyzed with ai-based transformers predict traditional subjective well-being measures approaching the theoretical upper limits in accuracy. Scientific reports, 12(1):3918

    2022

  20. [28]

    Peter J Kwantes, Natalia Derbentseva, Quan Lam, Oshin Vartanian, and Harvey HC Marmurek. 2016. Assessing the big five personality traits with latent semantic analysis. Personality and Individual Differences, 102:229--233

    2016

  21. [29]

    Jiwei Li, Xinlei Chen, Eduard Hovy, and Dan Jurafsky. 2015. Visualizing and understanding neural models in nlp. arXiv preprint arXiv:1506.01066

    2015 arXiv

  22. [30]

    Wenxiong Liao, Bi Zeng, Xiuwen Yin, and Pengfei Wei. 2021. An improved aspect-category sentiment analysis model for text sentiment analysis based on roberta. Applied Intelligence, 51:3522--3533

    2021

  23. [31]

    Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V. Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692

    2019 arXiv

  24. [32]

    Edward Loper and Steven Bird. 2002. https://doi.org/10.3115/1118108.1118117 NLTK : The natural language toolkit . In Proceedings of the ACL -02 Workshop on Effective Tools and Methodologies for Teaching Natural Language Processing and Computational Linguistics , pages 63--70, ...

    2002

  25. [33]

    Veronica Lynn, Niranjan Balasubramanian, and H Andrew Schwartz. 2020. Hierarchical modeling for user personality prediction: The role of message-level attention. In Proceedings of the 58th annual meeting of the association for computational linguistics, pages 5306--5316

    2020

  26. [34]

    Sean MacAvaney, Anjali Mittu, Glen Coppersmith, Jeff Leintz, and Philip Resnik. 2021. Community-level research on suicidality prediction in a secure environment: Overview of the clpsych 2021 shared task. In Proceedings of the Seventh Workshop on Computational Linguistics and C...

    2021

  27. [35]

    Matthew Matero, Albert Hung, and H Andrew Schwartz. 2021. Evaluating contextual embeddings and their extraction layers for depression assessment. arXiv preprint arXiv:2112.13795

    2021 arXiv

  28. [36]

    D.P. McAdams. 1993. The Stories We Live By: Personal Myths and the Making of the Self. Guilford Press

    1993

  29. [37]

    Yash Mehta, Samin Fatehi, Amirmohammad Kazameini, Clemens Stachl, Erik Cambria, and Sauleh Eetemadi. 2020. Bottom-up and top-down: Predicting personality with psycholinguistic and language model features. In 2020 IEEE international conference on data mining (ICDM), pages 1184-...

    2020

  30. [38]

    Raghavendra Pappagari, Piotr Zelasko, Jes \'u s Villalba, Yishay Carmiel, and Najim Dehak. 2019. Hierarchical transformers for long document classification. In 2019 IEEE automatic speech recognition and understanding workshop (ASRU), pages 838--844. IEEE

    2019

  31. [39]

    Park, H.A

    G. Park, H.A. Schwartz, J.C. Eichstaedt, M.L. Kern, M. Kosinski, D.J. Stillwell, L.H. Ungar, and M.E.P. Seligman. 2015. Automatic personality assessment through social media language. Journal of Personality and Social Psychology, 108(6):934--952

    2015

  32. [40]

    DL Paulhus and S Vazire. 2007. The self-report method. handbook of research methods in personality psychology

    2007

  33. [41]

    Heinrich Peters and Sandra C Matz. 2024. Large language models can infer psychological dispositions of social media users. PNAS nexus, 3(6):pgae231

    2024

  34. [42]

    Chris Piotrowski. 1999. Assessment practices in the era of managed care: Current status and future directions. Journal of Clinical Psychology, 55(7):787--796

    1999

  35. [43]

    Jack W Rae, Anna Potapenko, Siddhant M Jayakumar, and Timothy P Lillicrap. 2019. Compressive transformers for long-range sequence modelling. arXiv preprint arXiv:1911.05507

    2019 arXiv

  36. [44]

    Ted Schwaba. 2019. The structure, measurement, and development of openness to experience across adulthood. Handbook of personality development, pages 185--200

    2019

  37. [45]

    H Andrew Schwartz, Johannes C Eichstaedt, Margaret L Kern, Lukasz Dziurzynski, Stephanie M Ramones, Megha Agrawal, Achal Shah, Michal Kosinski, David Stillwell, Martin EP Seligman, et al. 2013. Personality, gender, and age in the language of social media: The open-vocabulary a...

    2013

  38. [46]

    H Andrew Schwartz and Lyle H Ungar. 2015. Data-driven content analysis of social media: A systematic overview of automated methods. The ANNALS of the American Academy of Political and Social Science, 659(1):78--94

    2015

  39. [47]

    Simchon, A

    A. Simchon, A. Sutton, M. Edwards, and S. Lewandowsky. 2023. Online reading habits can reveal personality traits: Towards detecting psychological microtargeting. PNAS Nexus, 2(6):pgad191

    2023

  40. [48]

    Christopher J Soto. 2019. How replicable are links between personality traits and consequential life outcomes? the life outcomes of personality replication project. Psychological Science, 30(5):711--727

    2019

  41. [49]

    Jessie Sun, H Andrew Schwartz, Youngseo Son, Margaret L Kern, and Simine Vazire. 2020. The language of well-being: Tracking fluctuations in emotion experience through everyday speech. Journal of Personality and Social Psychology, 118(2):364

    2020

  42. [50]

    Yla R Tausczik and James W Pennebaker. 2010. The psychological meaning of words: Liwc and computerized text analysis methods. Journal of language and social psychology, 29(1):24--54

    2010

  43. [51]

    Zhiguo Wang, Patrick Ng, Xiaofei Ma, Ramesh Nallapati, and Bing Xiang. 2019. Multi-passage bert: A globally normalized bert model for open-domain question answering. arXiv preprint arXiv:1908.08167

    2019 arXiv

  44. [52]

    T. A. Widiger and C. Crego. 2019. https://doi.org/10.1002/wps.20658 The five factor model of personality structure: an update . World Psychiatry, 18(3):271--272

    2019 doi

  45. [53]

    Thomas A Widiger and Joshua R Oltmanns. 2017. Neuroticism is a fundamental domain of personality with enormous public health implications. World psychiatry, 16(2):144

    2017

  46. [54]

    Joshua Wilt and William Revelle. 2015. Extraversion. Journal Name

    2015

  47. [55]

    Caroline Vaile Wright, Shannon G Beattie, Daniel I Galper, Abere Sawaqdeh Church, Lynn F Bufka, Virginia M Brabender, and Bruce L Smith. 2017. Assessment practices of professional psychologists: Results of a national survey. Professional psychology: research and practice, 48(2):73

    2017

  48. [56]

    Qingyang Wu, Zhenzhong Lan, Jing Gu, and Zhou Yu. 2020. Memformer: The memory-augmented transformer. Journal Name

    2020

  49. [57]

    Ruixuan Zhang, Zhuoyu Wei, Yu Shi, and Yining Chen. 2020. Bert-al: Bert for arbitrarily long document understanding. Journal Name

    2020

  50. [58]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  51. [59]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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