Topic modeling is a process that discovers key themes in unstructured text data by identifying the distribution of topics and words in a document, revealing hidden dimensions. Latent Dirichlet allocation is a widely used generative probabilistic topic model, but it cannot capture the dependency between topics. Generally, the topics within a document are primarily influenced by its overarching theme which naturally interrelates the topics. Thus, it is imperative to unveil such relationships between the topics. To this end, this study proposes a multilevel topic model (MTM) to unearth the hidden topic dependency in a corpus through multilevel latent structure. The MTM allows word-based topic proportions to vary across the higher-level latent structure. The parameters are estimated with a modified EM algorithm using an upward-downward approach to alleviate the computational complexity. Empirical studies on corpora have also been conducted on the multilevel topic model and the hierarchy of multilevel topic model have been interpreted. These analyses have demonstrated that the proposed multilevel topic model outperforms latent Dirichlet allocation in terms of systematic interpretability.

Topic modeling and alignment with large language models for multi-labeled text corpora

A topic modeling analysis of stigma dimensions, social, and related behavioral circumstances in clinical notes among patients with HIV.

Unveiling latent frontiers of urban and rural sustainability: a semi-automatic overview with LDA topic model

TEP-BS: Public opinion evolution prediction based on stochastic competitive learning-taking the hot case of platform X in september 2024 as an example.

Global and local context in short text neural topic model

Macau’s intangible cultural heritage (ICH) exemplifies a unique Sino-Western cultural fusion, wherein the interplay of Eastern and Western traditions complicates conventional analysis of heritage complexity and resilience. To address this challenge, we introduce FusionNet, a multimodal AI framework integrating image-based classification, an attention mechanism, identity embedding, and knowledge graph modeling for context-aware analysis of ICH. FusionNet combines image-based deep learning with an attention mechanism to focus on salient visual features in heritage imagery. This integrated architecture enables a holistic understanding of heritage elements and their adaptability to changing cultural contexts. Applied to Macau’s ICH, FusionNet reveals patterns of cultural resilience, illustrating how traditional practices persist and evolve amid centuries of EastWest influences. Our findings demonstrate the efficacy of fusing visual and knowledge-based modalities for heritage analysis, offering a robust approach for studying and preserving intangible cultural heritage in complex cultural environments. To elucidate how Macau’s intangible cultural heritage (ICH) exhibits “cultural resilience” and the mechanisms of identity (re)construction amid Sino‑Portuguese cultural interweaving; and to propose a computable multimodal framework (FusionNet + cultural‑identity embeddings + knowledge graph) that quantifies and validates these mechanisms. Materials include digital archives and historical texts (e.g., Macau Memory), social‑media text (Weibo plus ~1,000 English TripAdvisor/blog reviews), open heritage images, and structured knowledge bases (China ICH database). Methods comprise an attention‑based image classifier (FusionNet), LDA topic modeling (5‑fold cross‑validation selecting k = 3; mean coherence ≈ 0.59, compared with BERTopic), bilingual sentiment analysis, knowledge‑graph embedding and link prediction (evaluated with MRR, Hits@10), and t‑SNE visualization with clustering (three clusters; average silhouette ≈ 0.47). All implementations are in Python. LDA reveals three stable themes: (A) Chinese traditions (~45%), (B) Lusophone heritage (~30%), and (C) hybrid/local identity (~25%; e.g., Patuá and Macanese cuisine). Sentiment analysis indicates >70% positive evaluations, with ~12–15% negative. On the image side, most categories achieve diagonal accuracy >0.80, with some true‑positive rates reaching 0.95–1.00; Sino‑Portuguese architecture shows interpretable confusion. Knowledge‑graph embeddings and t‑SNE place the “hybrid/local identity” between the Chinese and Portuguese clusters, acting as a bridge (silhouette ≈ 0.47). Overall, multimodal fusion is more robust than multiple baselines on recognition and semantic association tasks, revealing a resilience pathway in which Macau ICH preserves core practices while continually absorbing exogenous elements. The proposed multimodal, knowledge‑driven framework effectively quantifies and explains identity (re)construction and cultural resilience in Macau’s ICH within a Sino‑Portuguese milieu; the “hybrid/local identity” is the key bridging mechanism. Future work can expand cross‑platform data, enhance cross‑modal alignment and knowledge reasoning, and generalize the approach to other multicultural contexts to strengthen external validity.

This study aimed to analyse patient-initiated compliment letters from a single institution, identify the key elements that patients value and offer actionable insights to enhance patient-centred care. A retrospective, single-institution study using the Healthcare Complaints Analysis Tool (HCAT), text network analysis and latent Dirichlet allocation (LDA) topic modelling on patient compliment letters to pinpoint key valued care elements. A newly established general hospital in Gwangmyeong, South Korea, opened on 22 March 2022. A total of 1213 compliment letters were collected through the hospital's feedback system, which accepted both online and on-site submissions between 25 March 2022 and 28 June 2024. Letters lacking substantive descriptive content and those containing purely administrative requests were excluded. The HCAT was adapted to categorise positive statements into clinical, management and relationship domains, along with six stages of care. Inter-rater reliability was evaluated using Gwet's AC1 statistic. A text network analysis, applying a term frequency-inverse document frequency approach, was conducted to identify prominent keywords. Subsequently, LDA was performed to extract thematic topics. Most compliments concerned the 'relationship' domain (62%), particularly during the care in the ward stage (56%). Keyword analysis indicated that the most frequently mentioned terms were 'gratitude', 'kindness', 'nurse', 'doctor' and 'heart/mind', underscoring patients' high valuation of positive interactions, professional competence and compassionate communication with medical staff. Topic modelling identified three primary topics, namely, 'appreciation of nursing care' (39%), 'professionalism in surgery and treatment' (35%) and 'effective communication during consultations' (26%). Positive relationships with medical staff, particularly kindness, professionalism and effective communication, influence patient satisfaction. Patient compliment letters serve as important indicators of exceptional care and can inform quality improvement initiatives. Healthcare institutions should leverage these insights to enhance patient-centred services by strengthening patient-provider relationships and promoting a culture of excellence.

Purpose This study aims to investigate the key factors influencing central bank digital currency (CBDC) adoption by conducting a meta-analysis of scholarly literature. It introduces a novel keyword-network analysis framework using bibliometric metadata from the Web of Science and Scopus databases, validated through artificial intelligence (AI)-based topic modeling. Design/methodology/approach Grounded in the unified theory of acceptance and use of technology (UTAUT) framework, this study identifies core and extended constructs related to CBDC adoption. It applies VOSviewer for keyword co-occurrence analysis on the bibliometric metadata and proposes a new method to calculate the relative importance of adoption factors based on link strength. This study also reveals dominant themes refined and validated through advanced AI-based topic modeling, signifying research trends on CBDC-adoption literature. Findings This study reveals dominant research themes, identified from topic keywords, demonstrating the breadth and depth of CBDC-adoption research and research trends on CBDC-adoption literature. Network-based weight calculations prioritized key adoption constructs such as performance expectancy, effort expectancy, social influence, usefulness, awareness, financial literacy, acceptance, behavior, intention, attitude, adoption intention and regulation, offering a structured understanding of CBDC-adoption dynamics. Practical implications The findings provide valuable insights for policymakers, regulators and financial institutions by highlighting the critical variables that drive or hinder CBDC adoption. The proposed bibliometric-AI hybrid methodology offers a replicable model for future digital currency and FinTech adoption studies. Originality/value This research pioneers a bibliometric and AI-integrated methodology to classify CBDC-adoption factors systematically. It extends the literature by linking thematic clusters to adoption constructs using quantitative co-occurrence analysis and advanced topic modeling.

ABSTRACT This study examines the effect of artificial intelligence (AI) adoption on financing performance in reward‐based crowdfunding. Using Kickstarter data from US projects, we find that AI projects have lower pledged amounts, receive fewer donations and attract fewer backers. The negative effect is partly driven by abnormal narrative tone and moderated by information readability, geographic conditions and backer sentiment. Additionally, subsample analysis shows that AI technology projects perform worst, while topic modelling indicates better outcomes when AI supports rather than replaces human creativity. Our findings highlight that the trust‐based frictions facing cutting‐edge innovations in crowdfunding differ from those in professional investor evaluations.

Purpose As global tourism accelerates, ecotourism has emerged as a strategic response to the environmental and social challenges posed by mass tourism. The integration of Industry 4.0 technologies into ecotourism holds the potential to support sustainable development, yet also introduces significant implementation barriers. This study aims to examine the impacts, challenges and opportunities of applying Industry 4.0 technologies in ecotourism, while mapping thematic clusters and future research directions with attention to environmental, social and economic sustainability. Design/methodology/approach A hybrid systematic literature review was conducted by using the PRISMA framework and Natural Language Processing (NLP). A total of 163 peer-reviewed articles published between 2017 and 2024 were retrieved from Scopus and Web of Science. Bibliometric analysis using VOSviewer was used to examine author collaboration and keyword co-occurrence. At the same time, topic modeling via latent Dirichlet allocation and semantic clustering (Sentence-BERT and Uniform Manifold Approximation and Projection) identified five key research clusters. Findings These clusters relate to the application of Internet of Things, artificial intelligence, blockchain, augmented reality/virtual reality and big data in ecotourism. While these technologies offer benefits such as real-time monitoring, personalized experiences and transparent certification, adoption remains limited because of infrastructure, cost and governance challenges. A four-stage digital roadmap (2020–2050) is proposed to guide the ecotourism digital transition. Research limitations/implications This study is limited by its reliance on peer-reviewed sources indexed in Scopus, which may exclude recent innovations and practitioner-led initiatives in ecotourism. The focus on publications from 2017 to 2024 could also overlook earlier foundational work. While the hybrid PRISMA and NLP approach improves thematic depth, variations in terminology and regional context may result in the omission of relevant studies. Future research should broaden data sources and explore multilingual and practice-based perspectives to strengthen the applicability of findings across diverse ecotourism settings. Practical implications The findings provide actionable insights for multiple stakeholders aiming to implement Industry 4.0 technologies in tourism destinations. For policymakers, the quadrant mapping of research clusters and roadmap visualization serve as strategic tools to prioritize digital infrastructure development, capacity building and adaptive policy design based on technology readiness and contextual needs. Tourism operators can leverage the findings to identify scalable solutions, such as real-time monitoring or AI-driven planning, while being aware of adoption barriers, such as interoperability, training gaps and visitor acceptance. Technology providers may use the thematic clusters to develop localized, low-cost applications suited to underserved ecotourism regions. Furthermore, the framework equips funding bodies and NGOs with an evidence-based foundation to align innovation investments with measurable sustainability outcomes, particularly in regions with limited digital readiness but high ecotourism potential. Social implications This study offers several theoretical contributions to the field of sustainable tourism and digital innovation. First, it extends the literature on smart ecotourism by applying a hybrid NLP bibliometric review framework, demonstrating how NLP and semantic clustering can complement traditional PRISMA-based approaches in uncovering latent research patterns. This methodology enhances the systematic analysis of large, unstructured corpora and supports the development of evidence-based research agendas in emerging tourism technologies. Originality/value This study contributes a novel hybrid NLP bibliometric framework for tourism research and provides actionable insights for policymakers, tourism operators and technology providers aiming to foster inclusive, technology-enabled sustainable tourism.

The 2000s have witnessed a significant, worldwide boom in new art museums founded by private, wealthy collectors. While the arts have long been a key arena for the remaking of elite distinction and the reproduction of inequalities, this surge in private museums has sparked much controversy. In this paper, we demonstrate how wealthy elites deploy this form of cultural philanthropy for (self)legitimation. Based on topic modelling analysis, we examine the online mission statements and 'about us' sections of 399 private museums across 59 countries to understand what forms of legitimation discourses they construct. We find that, beyond discourses of intra-elite distinction, the mission statements additionally mobilize discursive legitimation strategies that highlight private museums and their founders as reliable, institutionalized agents in the artworld and valuable philanthropic actors in society more broadly. Overall, our analysis demonstrates how the arts function as a particularly versatile and powerful tool for symbolic elite legitimation struggles, allowing wealthy elites from different backgrounds to coalesce globally around private art museums. In light of escalating wealth concentration and widening economic disparities around the world, our paper adds to sociology's critical imperative to scrutinize the formation and reproduction of contemporary elites.

Purpose Healthcare online feedback is widely used to improve service quality. This study aims to explore the determinants and evolving dynamics of healthcare employee satisfaction as reflected in employee-generated content. Design/methodology/approach This study analyzes structured (numerical ratings) and unstructured (textual feedback) data from over 300,000 online employee reviews of 9,103 US healthcare organizations. Using topic modeling, it identifies key satisfaction and dissatisfaction factors and examines their variations across job roles and tenure lengths, with a particular focus on the impact of the COVID-19 pandemic. Findings Our analysis reveals that job satisfaction determinants vary by role and tenure. During the initial phase of the COVID-19 pandemic, satisfaction temporarily increased due to a heightened sense of purpose and strong peer relationships. However, as the crisis persisted, satisfaction declined due to mounting stress, staff shortages, irregular shifts, and inadequate compensation. Practical implications These findings can guide healthcare organizations in developing targeted management strategies to enhance employee satisfaction and retention. Originality/value This study offers a novel perspective on healthcare online feedback by analyzing large-scale employee reviews from the service provider's standpoint, providing valuable insights into workplace experiences. Additionally, it contributes to employee satisfaction research by examining its dynamic changes across different phases and role-specific variations.

Spinal muscular atrophy (SMA) is a rare neuromuscular disorder caused by SMN1 mutations, leading to progressive muscle weakness. Although nusinersen improves motor function, its broader impact on daily life remains insufficiently understood. This study explored caregivers' perceptions of nusinersen treatment and identified themes reflecting its influence on everyday functioning. An observational study included 23 children with SMA (10 type 2, 7 type 3, 6 type 1) treated with nusinersen for up to 36 months. The mean age at interview was 6.3 years (SD 2.3). Caregivers completed a semi-structured questionnaire with closed and open-ended items. Narrative responses were analyzed using natural language processing (NLP) and topic modeling to identify recurring patterns. Caregivers reported consistent positive effects. NLP identified three thematic clusters: Functional and Daily Progress (34.9%), Global Improvement and Quality of Life (33.6%), and Hospitalization and Physical Recovery (31.5%). Reported benefits included improved swallowing, easier breathing, fewer hospitalizations, increased independence, and enhanced mood. Families emphasized reduced stress and improved family dynamics. Caregivers perceive nusinersen as providing meaningful improvements in daily functioning beyond motor outcomes. Incorporating caregiver-reported insights through NLP may enrich treatment evaluation and support holistic, patient- and family-centered care in SMA.

Social media has become a key venue where PhD graduates seek career information, compare experiences, and negotiate uncertainty. Drawing on information behavior and sense-making perspectives, this study examines how returnee PhDs from non-core study destinations discuss employment challenges in China’s academic labor market when credential signals are contested. Using Korean-trained PhDs as a theoretically motivated exemplary case, we collected 1149 publicly available posts from Xiaohongshu, a Chinese social media platform, and applied BERTopic to identify latent themes, followed by qualitative close reading of representative posts to interpret discourse functions. The model yielded ten topics, and semantic association analysis indicates substantial overlap among high-frequency topics, suggesting intertwined concerns rather than neatly separated issue domains. The four most prevalent topics account for 72.06% of the corpus, centering on credential recognition, job-search pathways, informal screening rules, and intersecting age- and gender-related pressures. Qualitative readings further reveal recurring discursive moves, including exposing tacit hiring heuristics, contesting stigmatizing labels (e.g., “water PhD,” a derogatory term implying low-quality credentials), and exchanging actionable strategies across regions and career tracks. Overall, the findings point to discursive convergence under evaluation uncertainty: when formal criteria are ambiguous and institutional signals are unreliable, participants turn to social media to stabilize expectations by triangulating cases and iteratively refining shared interpretations of the job market. This study contributes empirical evidence on uncertainty-driven information practices in highly educated labor markets and demonstrates the value of combining topic modeling with qualitative interpretation to capture online collective sense-making.

To address the limitations of static and coarse-grained analysis in mining online tourism reviews, this study proposes a gated attention-based multi-model fusion framework for dynamic topic evolution and complaint-driven latent issue pattern mining. Using 300,000 reviews from Ctrip and Meituan, we fuse global semantics from Sentence-BERT with attention (SBERT-Attention), local features from Bidirectional Encoder Representations from Transformers–Text Convolutional Neural Network (BERT-TextCNN), and topic distributions from the Biterm Topic Model (BTM) via a learnable gating mechanism. The fused model achieves an F1-score of 92.3% in review classification. We partition the corpus quarterly and apply Uniform Manifold Approximation and Projection (UMAP) followed by K-means++ clustering to the fused vectors, yielding interpretable topics, including Scenery, Transportation, Amenities, Management, Culture, and Value for Money, and enabling dynamic topic discovery over time. River map visualizations and negative review analysis reveal seasonal evolution patterns and recurring complaint patterns associated with specific topics. The framework enables dynamic, interpretable semantic mining, advancing intelligent processing of short-text user content and offering a generalizable approach for temporal knowledge discovery in smart tourism and beyond.

Purpose This study aims to address a critical gap in Humanitarian Logistics and Disaster Supply Chain Management (HLDSCM) scholarship by examining how academic research informs real-world policymaking. This study investigates “reverse dynamic” where scientific outputs support policy decisions worldwide and prioritize relevance to Sustainable Development Goals (SDG-3, SDG-11), thereby advancing broader science-policy dialogue. Design/methodology/approach An advanced methodological framework was used to identify and evaluate 2,132 Scopus-indexed articles and were systematically linked with policy documents in Overton database based on their citations coverage, density and intensity. The author identified most influential journal (Journal of Humanitarian Logistics and Supply Chain Management), author (Gyöngyi Kovács), institution (Hanken School of Economics, Finland) and country (United States). A machine learning-based Latent Dirichlet Allocation topic modeling approach was applied to detect core themes in the policy-cited research. This recent methodological advancement provides a more robust and scalable means to identify emergent themes and their policy relevance by enhancing the objectivity and depth of relevance assessment compared to conventional qualitative methods applied in HLDSCM research. Findings In total, 389 articles have been referenced in global policy documents, revealing an 18.24% policy citation rate. Analysis highlights key intermediaries and five dominant themes ranging from cross-sector collaboration to pandemic-driven adaptations that together contribute significantly to achieving SDGs. The study underscores growing appeal of HLDSCM research among policymakers seeking evidence-based guidance from academia. Research limitations/implications Policy citations capture visible traces of research in public policy documents but do not measure implementation or causal influence, and Overton coverage varies across regions and languages. Within these boundaries, the findings provide a benchmark for HLDSCM’s policy-document visibility; a theory-informed interpretation of why some HLDSCM research is more policy-visible than others; and actionable guidance for designing HLDSCM research and decision-support tools that are more usable for policy and operational planning aligned with SDG-3 and SDG-11. Originality/value The study combines policy-citation analysis with topic modeling to map and explain HLDSCM’s policy visibility, offering a replicable method and a theory-grounded set of recommendations for increasing the policy relevance of HLDSCM scholarship.

Purpose This study investigates the scope, intensity, and thematic structure of human capital (HC) disclosures in Islamic banks. It addresses the gap in understanding how HC narratives are constructed, benchmarked and communicated in faith-based financial institutions across diverse regulatory settings. Design/methodology/approach The study adopts a multi-method framework combining lexicon-based extraction, bidirectional encoder representations from transformers (BERT)-based sentence classification, criteria importance through intercriteria correlation (CRITIC) weighting, Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) benchmarking and BERTopic modeling. The analysis is based on 638 annual reports from 86 Islamic banks across 21 countries (2015–2023). Findings Results reveal substantial heterogeneity in disclosure intensity and thematic focus across institutions and jurisdictions. Compensation and governance-related themes dominate reporting, while diversity, equity and inclusion and employee well-being remain underdisclosed. The COVID-19 pandemic triggered a sharp increase in health and safety reporting. Country-level rankings highlight Indonesia, Malaysia and Bangladesh as consistent leaders. Originality/value An artificial intelligence/machine learning-enabled, multi-method framework is developed to measure and interpret HC disclosure in Islamic banking by integrating transformer-based sentence classification with CRITIC-weighted benchmarking, TOPSIS ranking and topic modeling. The study extends automated disclosure analytics to a Shariah-compliant setting and offers a scalable approach for cross-jurisdictional comparability and governance insight.

Can secessionist movements in Africa be conceptualized as forms of anti-(neo)colonial resistance? This article studies this question in relation to the Indigenous Peoples of Biafra (IPOB) movement in Nigeria. It adds nuance to the dominant interpretation by African secessionism and conflict scholars that characterize IPOB as ethnonationalist and destabilizing. While these are components of the movement, this article theorizes that IPOB’s secessionist project also strategically deploys a critique of the (neo)colonial Nigerian state. This critique is emancipatory in orientation but simultaneously deployed alongside polarizing and exclusionary politics. Through a content analysis that uses structural topic modeling (STM) alongside qualitative readings, this study analyzes changes in the topical themes of over 250 speeches given by movement leader Nnamdi Kanu on his Radio Biafra broadcast from 2013 to 2019. It finds that Kanu’s discursive strategies include anti-(neo)colonial narratives and recurring themes of liberation from the Nigerian state and its international allies. The findings provide a nuanced view of IPOB as a movement that uses several and, at times, contradictory rhetorical instruments in pursuit of its broader secessionist objectives. Notwithstanding, its adoption of anti-(neo)colonial resistance should not be overlooked.

Purpose The purpose of this study is to analyze publications on deceptive design indexed in the Scopus database, with a focus on identifying the yearly evolution of terminology and proposing a thematic classification that captures the most relevant research trends in this field. Design/methodology/approach The analysis was conducted in two stages. First, topic modeling was applied to article abstracts, organized by year of publication, to detect thematic structures. Second, grounded theory coding was used to classify and interpret the generated topics, enabling the construction of higher-level conceptual categories. Findings This study identifies an evolutionary model comprising four major themes: limited empirical evidence and problematic perspectives on deceptive design; reward systems and ethical practices in gaming; manipulation and deceptive design in consumer behavior; and deceptive design and user experience in online social environments. Practical implications The results can help designers, policymakers and businesses identify the evolution of deceptive design, anticipate regulatory or ethical concerns and develop strategies that promote more transparent and user-centered digital environments. Originality/value This study presents a novel evolutionary model of deceptive design research, combining computational and qualitative approaches. It not only synthesizes existing knowledge but also highlights underexplored areas, providing a roadmap for future investigations.