Interpretative Analysis Research Articles

Performance Evaluation of Large Language Models in Cervical Cancer Management Based on a Standardized Questionnaire: Comparative Study.

Cervical cancer remains the fourth leading cause of death among women globally, with a particularly severe burden in low-resource settings. A comprehensive approach-from screening to diagnosis and treatment-is essential for effective prevention and management. Large language models (LLMs) have emerged as potential tools to support health care, though their specific role in cervical cancer management remains underexplored. This study aims to systematically evaluate the performance and interpretability of LLMs in cervical cancer management. Models were selected from the AlpacaEval leaderboard version 2.0 and based on the capabilities of our computer. The questions inputted into the models cover aspects of general knowledge, screening, diagnosis, and treatment, according to guidelines. The prompt was developed using the Context, Objective, Style, Tone, Audience, and Response (CO-STAR) framework. Responses were evaluated for accuracy, guideline compliance, clarity, and practicality, graded as A, B, C, and D with corresponding scores of 3, 2, 1, and 0. The effective rate was calculated as the ratio of A and B responses to the total number of designed questions. Local Interpretable Model-Agnostic Explanations (LIME) was used to explain and enhance physicians' trust in model outputs within the medical context. Nine models were included in this study, and a set of 100 standardized questions covering general information, screening, diagnosis, and treatment was designed based on international and national guidelines. Seven models (ChatGPT-4.0 Turbo, Claude 2, Gemini Pro, Mistral-7B-v0.2, Starling-LM-7B alpha, HuatuoGPT, and BioMedLM 2.7B) provided stable responses. Among all the models included, ChatGPT-4.0 Turbo ranked first with a mean score of 2.67 (95% CI 2.54-2.80; effective rate 94.00%) with a prompt and 2.52 (95% CI 2.37-2.67; effective rate 87.00%) without a prompt, outperforming the other 8 models (P<.001). Regardless of prompts, QiZhenGPT consistently ranked among the lowest-performing models, with P<.01 in comparisons against all models except BioMedLM. Interpretability analysis showed that prompts improved alignment with human annotations for proprietary models (median intersection over union 0.43), while medical-specialized models exhibited limited improvement. Proprietary LLMs, particularly ChatGPT-4.0 Turbo and Claude 2, show promise in clinical decision-making involving logical analysis. The use of prompts can enhance the accuracy of some models in cervical cancer management to varying degrees. Medical-specialized models, such as HuatuoGPT and BioMedLM, did not perform as well as expected in this study. By contrast, proprietary models, particularly those augmented with prompts, demonstrated notable accuracy and interpretability in medical tasks, such as cervical cancer management. However, this study underscores the need for further research to explore the practical application of LLMs in medical practice.

Deep learning interpretability analysis for carbon star identification in Gaia DR3

A large fraction of asymptotic giant branch (AGB) stars develop carbon-rich atmospheres during their evolution. Based on their color and luminosity, these carbon stars can easily be distinguished from many other kinds of stars. However, numerous G, K, and M giants also occupy the same region as carbon stars on the HR diagram. Despite this fact, their spectra exhibit differences, especially in the prominent CN molecular bands. We aim to distinguish carbon stars from other kinds of stars using Gaia's XP spectra while providing attributional interpretations of key features that are necessary for identification and even discovering new key spectral features. We propose a classification model named "GaiaNet," an improved one-dimensional convolutional neural network specifically designed for handling Gaia's XP spectra. We utilized SHapley Additive exPlanations (SHAP), an approach for interpretability based on game theoretic, to determine SHAP values for each feature in a spectrum, enabling us to explain the output of the GaiaNet model and provide further meaningful analysis. Compared to four traditional machine learning methods, the GaiaNet model exhibits an average classification accuracy improvement of approximately 0.3% on the validation set, with the highest accuracy reaching 100%. Utilizing the SHAP model, we present a clear spectroscopic heatmap highlighting molecular band absorption features primarily distributed around CN_ and CN_ and we summarize five key feature regions for carbon star identification. Upon applying the trained classification model to the CSTAR sample with Gaia "xp_sampled_mean" spectra, we obtained 451 new candidate carbon stars as a by-product. Our algorithm is capable of discerning subtle feature differences from low-resolution spectra of Gaia, thereby assisting us in effectively identifying carbon stars with typically higher temperatures and weaker CN features while providing compelling attributive explanations. The interpretability analysis of deep learning holds significant potential in spectral identification.

The COVID-19 pandemic art of Anna Szpakowska-Kujawska: from pastels to spatial compositions

In this article, Karolina Tomczak is concerned with the latest and as yet unstudied works of renowned Wrocław artist Anna Szpakowska-Kujawska (born in 1931) created during the pandemic. These works include a series of pastels Kwadraty zaokienne (Squares outside the Window, or “pandemic squares”) from 2020 and spatial objects made of polystyrene foam and cardboard executed in the years 2021–2022. Tomczak gives an analytical interpretation of the modernist element in Szpakowska-Kujawska’s artistic record of her experience of the plague as it developed from painting at the point of realistic departure to unconventional sculpture. Tomczak shows that the reworking of difficult realities in the artistic medium finally takes place in spatial objects that are transgressively modern in that they reinterpret the model of avant-garde sculpture, selected postmodern artistic solutions and Szpakowska-Kujawska’s original art. At the same time, Tomczak sees Szpakowska-Kujawska’s works as expressive objects and personal psychograms encoding reality in the subjective iconography of a special period, which stylistically determines the coexistence of individualized biomorphism and simplified figuration, abstract universalization and symbolic representation. Being an example of creative adaptation to the contemporary threat, the works are affirmations of artistic activity in everyday conditions, as well as a continuation of Szpakowska-Kujawska’s distinct version of occasionally transgressive modernism. In her methodologically concretized approach, Tomczak uses stylistic analysis and compositional interpretation, as well as contextual and partly psychoanalytic method.

Analysis of student interpreters’ self-assessment from the perspective of self-regulated learning

Abstract This study aims to examine the significance of self-assessment in interpreter training from the perspective of self-regulated learning. To this end, this study first explores the three cyclical stages of self-regulated learning as conceptualized by Zimmerman, while also examining the significance of self-evaluation in interpreter training. Based on this theoretical framework, the study analyzes self-evaluations from both high-performing and low-performing students at a graduate school of translation and interpretation. The findings of the study reveal three key points: first, the effective use of self-assessment enhances learning effectiveness in interpreter training; second, significant differences exist between high-performing and low-performing students in their self-assessment practices; third, these differences highlight the need for incorporating self-assessment into interpreter training programs. Finally, the study suggests that enhancing self-regulated learning capabilities by actively incorporating self-evaluation into interpreter training can improve educational outcomes in interpreter training, leveraging the cyclical and positive impact of self-regulated learning processes.

Improving Identification of Drug-Target Binding Sites Based on Structures of Targets Using Residual Graph Transformer Network

Improving identification of drug-target binding sites can significantly aid in drug screening and design, thereby accelerating the drug development process. However, due to challenges such as insufficient fusion of multimodal information from targets and imbalanced datasets, enhancing the performance of drug-target binding sites prediction models remains exceptionally difficult. Leveraging structures of targets, we proposed a novel deep learning framework, RGTsite, which employed a Residual Graph Transformer Network to improve the identification of drug-target binding sites. First, a residual 1D convolutional neural network (1D-CNN) and the pre-trained model ProtT5 were employed to extract the local and global sequence features from the target, respectively. These features were then combined with the physicochemical properties of amino acid residues to serve as the vertex features in graph. Next, the edge features were incorporated, and the residual graph transformer network (GTN) was applied to extract the more comprehensive vertex features. Finally, a fully connected network was used to classify whether the vertex was a binding site. Experimental results showed that RGTsite outperformed the existing state-of-the-art methods in key evaluation metrics, such as F1-score (F1) and Matthews Correlation Coefficient (MCC), across multiple benchmark datasets. Additionally, we conducted interpretability analysis for RGTsite through the real-world cases, and the results confirmed that RGTsite can effectively identify drug-target binding sites in practical applications.

STATISTICS EDUCATION IN A CRITICAL PERSPECTIVE FOR THE STUDY OF INTERSECTIONALITIES

This study examines, with focus on statistical education, how intersectionality appears in didactic tasks for Brazilian school education, developed by IBGEeduca, an educational platform organized by IBGE (Brazilian Institute of Geography and Statistics). In this study, intersectionality was used as an analytical tool to help understand and explain the complexities of the world that impose social injustices through interrelationships of power categories. The documentary methodological approach adopted in this research is part of a doctoral research project. The criteria for choosing the analyzed tasks were based on the authors' perception of the intersectional categories in the examined pedagogical proposals. We present an interpretative analysis of an IBGEeduca proposal from the theoretical perspective of critical statistics education. This investigation revealed several themes related to contemporary problems present sporadically in the proposed activities, requiring greater scrutiny of the hidden intersections among the categories exposed in the proposals. The findings of this Brazilian case study could spark conversations about how statistics teachers can use tasks to help students understand concepts of inequity and social injustice.

Optimization and Interpretability Analysis of Machine Learning Methods for ZnO Colloid Particle Size Prediction

Optimization and Interpretability Analysis of Machine Learning Methods for ZnO Colloid Particle Size Prediction

Disentangled global and local features of multi-source data variational autoencoder: An interpretable model for diagnosing IgAN via multi-source Raman spectral fusion techniques.

A single Raman spectrum reflects limited molecular information. Effective fusion of the Raman spectra of serum and urine source domains helps to obtain richer feature information. However, most of the current studies on immunoglobulin A nephropathy (IgAN) based on Raman spectroscopy are based on small sample data and low signal-to-noise ratio. If a multi-source data fusion strategy is directly adopted, it may even reduce the accuracy of disease diagnosis. To this end, this paper proposes a data enhancement and spectral optimization method based on variational autoencoders to obtain reconstructed Raman spectra with doubled sample size and improved signal-to-noise ratio. In the diagnosis of IgAN in multi-source domain Raman spectra, this paper builds a global and local feature decoupled variational autoencoder (DMSGL-VAE) model based on multi-source data. First, the statistical features after spectral segmentation are extracted, and the latent variables obtained by the variational encoder are decoupled through the decoupling module. The global representation and local representation obtained represent the global shared information and local unique information of the serum and urine source domains, respectively. Then, the cross-source reconstruction loss and decoupling loss are used to constrain the decoupling, and the effectiveness of the decoupling is proved quantitatively and qualitatively. Finally, the features of different source domains were integrated to diagnose IgAN, and the results were analyzed for important features using the SHapley Additive exPlanations algorithm. The experimental results showed that the AUC value of the DMSGL-VAE model for diagnosing IgAN on the test set was as high as 0.9958. The SHAP algorithm was used to further prove that proteins, hydroxybutyrate, and guanine are likely to be common biological fingerprint substances for the diagnosis of IgAN by serum and urine Raman spectroscopy. In summary, the DMSGL-VAE model designed based on Raman spectroscopy in this paper can achieve rapid, non-invasive, and accurate screening of IgAN in terms of classification performance. And interpretable analysis may help doctors further understand IgAN and make more efficient diagnostic measures in the future.

Correlation based-graph neural network for health prognosis of non-fully charged and discharged lithium-ion batteries

The growing adoption of lithium-ion batteries in various real-world applications creates an urgent need for efficient battery management technology, particularly in monitoring battery health conditions. While efforts have been made to achieve battery health prognosis under practical cycling conditions, there are still challenges as these methods are mainly modeled with high-quality battery degradation data or limited to specific application scenarios. The lack of extrapolation and generalisation of existing methods motivated this paper to develop a framework capable of utilising data from complex and practical operating condition for battery health prognosis. This study developed a correlation based-graph neural network (GNN) estimation framework. Two correlation analysis methods, Pearson’s correlation coefficient and the Mantel test, were applied to extract valid information and identify critical health features. Regression modeling was then used to estimate the current state of health (SOH) of a battery. The SHAP (Shapley Additive exPlanations) values derived from the interpretability analysis were utilised to select the most important health features from the regression model. Batteries from two datasets with various working conditions are chosen to validate the framework’s effectiveness. Experimental results show that the framework, designed for practical applications, demonstrates exceptional capability in accurately and rapidly estimating battery health conditions.

Sequencing Silicates in the Spitzer Infrared Spectrograph Debris Disk Catalog. I. Methodology for Unsupervised Clustering

Abstract Debris disks, which consist of dust, planetesimals, planets, and gas, offer a unique window into the mineralogical composition of their parent bodies, especially during the critical phase of terrestrial planet formation spanning 10 yr to a few hundred million years. Observations from the Spitzer Space Telescope have unveiled thousands of debris disks, yet systematic studies remain scarce, let alone those with unsupervised clustering techniques. This study introduces CLustering UnsupErvised with Sequencer (CLUES), a novel, nonparametric, fully interpretable machine learning spectral analysis tool designed to analyze and classify the spectral data of debris disks. CLUES combines multiple unsupervised clustering methods with multiscale distance measures to discern new groupings and trends, offering insights into compositional diversity and geophysical processes within these disks. Our analysis allows us to explore a vast parameter space in debris disk mineralogy and also offers broader applications in fields such as protoplanetary disks and solar system objects. This paper details the methodology, implementation, and initial results of CLUES, setting the stage for more detailed follow-up studies focusing on debris disk mineralogy and demographics.

Intelligent generation and interpretability analysis of shear wall structure design by learning from multidimensional to high-dimensional features

Intelligent generation and interpretability analysis of shear wall structure design by learning from multidimensional to high-dimensional features

Comprehensive analysis of compost maturity differences across stages and materials with statistical models.

Aerobic composting is an environmentally friendly and effective approach to treating organic solid waste. The variability in material composition introduces complex interactions between environmental factors and materials, which in turn affects compost maturity. This study uses multiple statistical analyses to systematically compare key indicators across composting processes for kitchen waste, livestock manure, and sludge. The results show that material type and composting stage have a significant impact on compost maturity (p<0.001). High-precision modeling (R2>0.90) was achieved using a Stacking model on the composting dataset, with interpretability analysis highlighting the important roles of temperature, moisture content, and nitrogen content across different composting materials. The combined effects of environmental and material changes jointly influence the composting progression. In kitchen waste composting, strong interactions between multiple indicators were observed, while moisture shifts in livestock manure and sludge composting primarily influenced compost maturity by promoting decomposition and enhancing nitrogen retention, respectively. Partial dependence analysis quantified the relationships between key indicators and compost maturity scores. These findings offer a scientific basis for identifying key factors and optimization paths in various composting processes, supporting the development of more effective composting strategies.

Improved Reservoir Porosity Estimation Using an Enhanced Group Method of Data Handling with Differential Evolution Model and Explainable Artificial Intelligence

Summary Reservoir characterization is critical to the oil and gas industry, influencing field development, production optimization, hydraulic fracturing, and reserves estimation decisions. Accurately estimating porosity is crucial for reservoir characterization, well planning, and production optimization in the oil and gas industry. Traditional porosity determination methods, such as porosimetry, geostatistical, and core analysis, often involve complex geological and geophysical models, which are expensive and time-consuming. This study used the integrated machine learning and optimization model of differential evolution (DE) with group method of data handling (GMDH-DE) to estimate the porosity using integrated well log and core data from the Mpyo oil field, Uganda. The GMDH-DE demonstrates superior performance compared with conventional GMDH, support vector regression (SVR), and random forest (RF), achieving a coefficient of determination (R2) of 0.9925 and a root mean square error (RMSE) of 0.0017 during training, an R² of 0.9845 with an RMSE of 0.0121 during testing, and when validated the R2 was 0.9825 with RMSE of 0.00018. A key novelty of this work is the integration of Shapley additive explanations (SHAP), which provides an interpretable analysis of the model’s input features. SHAP reveals that bulk density (RHOB) and neutron porosity (NPHI) are the most critical parameters for porosity estimation, offering valuable insight into features importance. The proposed GMDH-DE model and SHAP analysis represent a novel and independent approach for accurate porosity estimation and interpretability, significantly enhancing the efficiency and reliability of hydrocarbon exploration and development.

Machine learning-based prediction for airflow velocity in unpressured water-conveyance tunnels

Spillway and drainage tunnels have an open-channel flow pattern when operating under unpressured condition, above which air flow is driven and resisted by water flow, wall friction, and pressure difference. Unpressured tunnels present many airflow-related safety and environmental issues, including water flow fluctuation, gate vibration, shaft cover blow-off, and odor emission; therefore, it is valuable to study and predict their airflow velocity. Given the difficulty in accurate prediction of airflow velocity in unpressured tunnels and complicated influences of hydraulic, structural, and boundary pressure parameters, this study focuses on establishing high-performance prediction models and understanding the importance and independent and coupled influences of each parameter using machine learning. It is found that the water Froude number, ratio of free-surface width to unwetted perimeter, relative ventilation area, and relative tunnel length are four key parameters. By including these parameters in the input parameter combination, the machine learning models can well predict the airflow velocity in unpressured tunnels, achieving significantly higher performance than the existing empirical and theoretical models. Among these models, the models built by Random Forest and XGBoost demonstrate best performance with R2 ≥ 0.911. The interpretability analysis reveals the highest importance of the water Froude number and the ratio of free-surface width to unwetted perimeter, increases in which generally result in enhancement of the airflow velocity. The water Froude number plays a dominant role when it is ≤11.5, and a continuous increase exhibits a significantly marginal effect. The relative ventilation area and relative length of tunnels have close importances, with an increase in either generally promoting the airflow velocity. To help researchers and engineers unfamiliar with machine learning to easily and accurately predict the airflow velocity in unpressured tunnels, GPlearn algorithm is employed to establish explicit expressions, which is validated to have good performance with R2 close to 0.900.

Developing an interpretable machine learning model for diagnosing gout using clinical and ultrasound features.

To develop a machine learning (ML) model using clinical data and ultrasound features for gout prediction, and apply SHapley Additive exPlanations (SHAP) for model interpretation. This study analyzed 609 patients' first metatarsophalangeal (MTP1) joint ultrasound data from two institutions. Institution 1 data (n=571) were split into training cohort (TC) and internal testing cohort (ITC) (8:2 ratio), while Institution 2 data (n=92) served as external testing cohort (ETC). Key predictors were selected using Random Forest (RF), Least Absolute Shrinkage and Selection Operator (LASSO), and Extreme Gradient Boosting (XGBoost) algorithms. Six ML models were evaluated using standard performance metrics, with SHAP analysis for model interpretation. Five key predictors were identified: serum uric acid (SUA), deep learning (DL) model predictions, tophus, bone erosion, and double contour sign (DCs). The logistic regression (LR) model demonstrated optimal performance, achieving Area Under the Curve (AUC) values of 0.870 (95% CI: 0.820-0.920) in ITC and 0.854 (95% CI: 0.804-0.904) in ETC. The model showed good calibration with Brier scores of 0.138 and 0.159 in ITC and ETC, respectively. This study developed an interpretable ML model for gout prediction and utilized SHAP to elucidate feature contributions, establishing a foundation for future applications in clinical decision support for gout diagnosis.

Assessing impact of a community-based screening campaign to address social determinants of cervical cancer.

Screening represents a cornerstone of cervical cancer control strategy. However, disparities in social determinants of health (SoDoH) have perpetuated gaps in screening among racial/ethnic minorities. SoDoH including cultural stigma and lack of insurance have contributed to decreased screening among Hispanic women. To increase cancer screening in this population, community-academic partnerships and culturally-tailored media have emerged as promising strategies. This study assessed the impact of a culturally-tailored cervical cancer screening campaign implemented through academic-community-government partnership. Intercept surveys, conducted from 2015-2018 in eastern neighborhoods of Los Angeles, assessed campaign recall, interpretation and screening intention among Hispanic women aged 21-65 after exposure to the campaign. Screening intention was evaluated using Chi-square tests and logistic regression by participant characteristics, with thematic analysis for campaign interpretation. Out of 673 participants, 26.1% were uninsured and 85.9% primarily spoke Spanish at home. Campaign recall was 25.1%, with 64.5% interpreting the campaign's message as cervical cancer screening or health checkups. The campaign's most-liked aspect was emphasis on family (cited by 37.1% of participants). Post-campaign, 89.5% of participants overall were likely or extremely likely to schedule a Pap test, including 83.5% of women who had not had a Pap test in the last three years. Our findings underscore several important strategies to reduce cervical cancer disparities: 1) associating positive cultural values with screening to decrease stigma, 2) combining culturally-tailored outreach with interventions that target other known screening barriers, 3) facilitating long-term community relationships and 4) leveraging academic-community-government partnerships.

Detection of Oral Fluid Stains on Fabric via Solution Extraction Combined with Deep Ultraviolet Raman Spectroscopy.

DNA phenotyping plays a central role in modern practical forensics, yet an overwhelming amount of evidence creates significant backlogs in all major crime laboratories. A fast nondestructive test of a potential biological stain prior to DNA phenotyping should reduce the number of irrelevant samples for the analysis and increase the efficiency of the overall process. Evidence items recovered from the crime scene can often include body fluid traces, such as oral fluid (OF). This proof-of-concept study demonstrates the effectiveness of Deep-UV Raman spectroscopy in identifying OF stains on substrates such as cotton, polyester, and blue denim, commonly encountered in forensic investigations. Through spectral interpretation and statistical analysis, this study compares Raman spectra from OF extracted from substrates with pure OF spectra. Additionally, comparative analysis with near-infrared (NIR) Raman spectroscopy to deep-UV Raman spectroscopy was performed. Distinct advantages of deep-UV Raman spectroscopy were determined, including reduced sample preparation requirements and the absence of fluorescence background, enhancement of the signal-to-noise ratio, and simplified data preprocessing. Using statistical analysis methods like principal component analysis and partial least-squares discriminate analysis, differentiation between OF and non-OF samples was possible. Overall, this study underscores the versatility and potential of deep-UV Raman spectroscopy as a valuable tool in forensic science.

An explainable transformer model integrating PET and tabular data for histologic grading and prognosis of follicular lymphoma: a multi-institutional digital biopsy study.

Pathological grade is a critical determinant of clinical outcomes and decision-making of follicular lymphoma (FL). This study aimed to develop a deep learning model as a digital biopsy for the non-invasive identification of FL grade. This study retrospectively included 513 FL patients from five independent hospital centers, randomly divided into training, internal validation, and external validation cohorts. A multimodal fusion Transformer model was developed integrating 3D PET tumor images with tabular data to predict FL grade. Additionally, the model is equipped with explainable modules, including Gradient-weighted Class Activation Mapping (Grad-CAM) for PET images, SHapley Additive exPlanations analysis for tabular data, and the calculation of predictive contribution ratios for both modalities, to enhance clinical interpretability and reliability. The predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC) and accuracy, and its prognostic value was also assessed. The Transformer model demonstrated high accuracy in grading FL, with AUCs of 0.964-0.985 and accuracies of 90.2-96.7% in the training cohort, and similar performance in the validation cohorts (AUCs: 0.936-0.971, accuracies: 86.4-97.0%). Ablation studies confirmed that the fusion model outperformed single-modality models (AUCs: 0.974 - 0.956, accuracies: 89.8%-85.8%). Interpretability analysis revealed that PET images contributed 81-89% of the predictive value. Grad-CAM highlighted the tumor and peri-tumor regions. The model also effectively stratified patients by survival risk (P < 0.05), highlighting its prognostic value. Our study developed an explainable multimodal fusion Transformer model for accurate grading and prognosis of FL, with the potential to aid clinical decision-making.

The Client in Intrinsic Agency-Based Evaluation: Implications for Social Work Pedagogy

ABSTRACT The author addresses the importance of the evaluation client within evaluation theory and practice and offers a qualitatively derived framework for conceptualizing this role in social work evaluation. Based on an interpretative qualitative analysis of graduate students’ evaluation reports authored over 11 offerings of a specific capstone course on intrinsic agency-based evaluation, the author undertook an analysis of 96 reports selected from 220. Using open coding, the author surfaced four alternative ways of conceiving of the evaluation client, and through interpretation organized those into a typology incorporating two dimensions: the client’s strength of evaluation perspective, and the client’s level of engagement. The author discusses each of the four cells, identifies issues facing social work evaluators when they interact with each type of evaluation client, and draws conclusions for the role of each of the four types of evaluation client in intrinsic agency-based evaluation. The author draws implications for the use of the typology in undertaking client-driven evaluation.

Willy Wonka and the boy-entrepreneur archetype: the changing faces of entrepreneurial masculinity, 1971–2023

PurposeThis paper uses the cultural figure of Willy Wonka to explore the archetype of the “boy-entrepreneur”, and what the various film manifestations of Wonka (1971–2023) say about changes in entrepreneurial masculinity. We (1) develop an original conceptualisation of boyhood as creatively, socially and gender liminal, (2) analyse the entrepreneurial archetype using literary theories and (3) provide a novel interpretation of Apollonian and Dionysian masculinity to aid future cultural analysis of founder/innovator depictions in children’s media.Design/methodology/approachOur study conducts a three-stage dramaturgical analysis of the major film adaptations of Roald Dahl’s Willy Wonka (1971, 2005 and 2023). After supplementing the limited literature on boyhood in entrepreneurship with literary and feminist art theory, we identify the significant narrative features which frame the Wonka dramas and produce a framework which tracks thematic changes across the films in terms of the creative, social and gender liminal elements of Wonka’s boy-entrepreneur identity.FindingsOur interpretive analysis reveals a gradual shift in values expressed through Wonka’s boy-entrepreneurialism away from a more hegemonic, Apollonian style of masculinity towards a more Dionysian style embracing emotional expression, intimacy with female characters and kin, and collaborations with nature. Such shifts reflect the growing influence of non-hegemonic entrepreneurial gender expressions, value-driven and relational approaches to new venture creation.Originality/valueThe study contributes to the hugely unexplored area of the “boy-entrepreneur”/“boy-genius”, demonstrating its durability in reality, imagination and popular culture. We provide an in-depth character portrait to provoke further attention to children’s multi-media ways of experiencing early entrepreneurial impressions. We also expand the methodological scope of research on entrepreneurial masculinity beyond real-life founders.