From LSA to LLM: Evolution and limitations of topic modelling methods for biodiversity conservation

The recent declines in youth mental health highlight the need for research into the factors underlying distress and those that foster well-being. Open-ended text responses offer the potential to reveal novel insights, but remain an underused resource. Advances in natural language processing (NLP) offer powerful tools for efficiently analyzing text in large samples. This study aimed to extract meaningful themes from young adults' open-text responses about their future hopes and worries, and to test associations of these themes with internalizing symptoms. Data came from an urban community sample of 24-year-olds (N=1113) who provided brief written responses about their greatest hopes and worries. A total of 3973 text-segments were analyzed using topic modeling with the Python library BERTopic. Associations of the themes with internalizing symptoms were tested using regression analyses. Thirteen thematic topics for both hopes and worries emerged. Young adults' hopes and worries spanned personal, interpersonal, work-life-finances and broader systemic and global domains. Many themes overlapped, but hopes tended to center more on interpersonal relationships, whereas worries were focused more on systemic and global challenges. Higher levels of internalizing symptoms were associated with more interpersonal and work-, life-, and financial worries, but with fewer systemic concerns, such as climate- or war and conflict- related worries. Our work exemplifies diverse, normative, and broadly shared hopes and worries among young adults that are important for youth-focused public policy and psychosocial support. Specific topics associated with internalizing symptoms including financial or interpersonal concerns constitute concrete interventional targets for alleviating distress.

Remote rural communities often remain trapped in asset-based poverty because rural land functions as “dead capital” that cannot be easily monetized for more profitable uses. One potential solution is transferable development rights (TDR), a market-based redistribution instrument that monetizes rural development quotas and channels part of urban expansion gains to disadvantaged rural areas. Yet evidence on whether TDR alleviates poverty is mixed, and prior research has emphasized material outcomes while paying less attention to the social and political processes that generate unequal outcomes and to spatial heterogeneity within rural areas. We therefore apply a trivalent spatial justice framework—distributive, procedural, and recognitional justice—to assess China's TDR and explain why impacts differ between remote hinterland and peri-urban sending areas. By applying Latent Dirichlet Allocation topic modeling and spatial analysis to examine online citizen–government interactions from a Chinese participatory platform, we find that while TDR programs provide short-term economic gains for rural residents, these gains are frequently offset by longer-term livelihood losses. Procedural and recognitional injustices are central: a government-centered alliance marginalizes farmers' voices, while relocation reshapes landscapes, erodes rural culture, and reproduces discrimination. Moreover, these justice outcomes are spatially uneven—peri-urban areas exhibit stronger rights-claiming capacity and relatively better distributive outcomes, whereas remote areas face deeper constraints and greater livelihood risks. We conclude that poverty reduction cannot rely on land reform alone. The path to revitalizing the countryside lies in institutional reforms, particularly in rural political governance and the empowerment of rural communities. • This study uses a spatial justice framework to assess the effectiveness of transferable development rights on revitalizing rural land in Guangdong Province, China. • We apply a topic modeling algorithm to analyze citizen-government interactions on an online participation platform. • The transferable development rights program often delivers short-term monetary compensation, yet is frequently associated with under-cultivated/idle land and longer-term livelihood insecurity. • Rural land reform-oriented solutions to poverty alleviation must involve institutional reforms, particularly in rural political governance and the empowerment of local communities. • Justice outcomes are spatially uneven. Remote hinterlands face structural constraints and suffer from long-term livelihood losses, whereas peri-urban areas benefit from higher administrative capacity, thereby securing relatively better distributive outcomes.

Malignant temporal bone tumors (1941-2025): A bibliometric analysis of publication trends, key contributors, and thematic evolution.

Rice is fundamental to global food and nutritional security, yet the evolution of rice-related research has not been systematically mapped at scale. This study applied an AI-driven bibliometric framework to 99,011 peer-reviewed articles indexed in Scopus (1970 to 2024), integrating natural language processing using term frequency-inverse document frequency (TF-IDF feature extraction) with topic modelling via Latent Dirichlet Allocation (LDA) and network analysis using graph-based clustering. This enabled both thematic structuring of research and identification of global collaboration patterns. Five dominant knowledge domains emerged: (1) soil contamination and heavy metal uptake in rice systems, (2) agricultural productivity and environmental impact, (3) nutritional and functional applications of rice by-products, (4) genotypic diversity and stress adaptation, and (5) genomic and molecular strategies for rice improvement. Temporal dynamics revealed a shift from agronomic yield and soil management research (1970s to 1990s) toward molecular genetics, stress resilience and environmental sustainability in the post-2000 era, with nutritional functionality and by-product utilization emerging only in the last decade. Collaboration mapping showed Asia being led by India, China and Japan as the primary research hubs, while Western institutions frequently connected regional clusters. Although progress was achieved, thematic compartmentalization remained, with limited interdisciplinary collaboration across molecular, agronomic and nutritional domains. By integrating machine learning (ML) and large-scale bibliometrics, this study provides the first systems-level evidence base of rice science, aimed at prioritizing areas for cross-disciplinary research and policy engagement to enhance and accelerate innovations towards resilient, sustainable and nutrition-sensitive food systems.

Derivation topic propagation prediction model based on topic attractiveness and dynamic temporal perception

Artificial intelligence in psychiatry: A global perspective on research status, trends and clinical applications.

Mapping global research on international students’ cross-cultural adaptation: A structural topic modelling review

Policy feedback in crowd-safety crises: a dynamic topic modeling approach to South Korea’s Post-Itaewon crisis

Firms investing in market-upstream R&D face governance challenges that vary systematically with the stage of technological development. This study examines how technology maturity shapes the design of interfirm R&D contracts, particularly in balancing appropriation and coordination concerns. Drawing on Transaction Cost Economics (TCE) and incomplete contracting theory, we view contracts as inherently incomplete and use the discriminating alignment principle as a lens to investigate which governance mechanisms are most effective at different levels of technology maturity. Leveraging a large-scale, unique dataset of R&D contracts in the semiconductor industry, we apply a correlated topic model (CTM) to systematically analyze and compare contractual mechanisms across multiple levels of technology maturity. We find that appropriation mechanisms are more prominent at low to intermediate maturity levels, where knowledge remains tacit and appropriation concerns are high. Conversely, coordination mechanisms become increasingly critical at intermediate and high maturity levels, where demands related to integration, execution, and commercialization intensify. We also demonstrate that different levels of technology maturity are associated with distinct framings of key contractual elements—such as intellectual property rights and project management clauses. By introducing the technology readiness level (TRL) framework as an empirical lens for studying contract design, this study advances alliance governance and incomplete contracting theory by revealing stage-contingent variation in contractual governance across the innovation process. • Technology maturity systematically shapes the design of interfirm R&D contracts. • We investigated a unique set of R&D contracts using correlated topic modelling. • The extent of appropriation and coordination mechanisms varies across the stages. • But also the framing of key contractual clauses evolves with technology maturity. • Technology Readiness Levels offer a structured basis for explaining stage-contingent governance needs.

Prioritizing accurate cost contingency estimation through risk identification is essential for the success of construction projects. Traditional methods for identifying and classifying risk factors, such as workshops, interviews, and referencing similar projects, are predominantly manual, subjective, and time-consuming. To overcome these challenges, this study introduces a novel deep learning approach that leverages the BERTopic algorithm to extract cost-related risk factors from extensive project risk registers. The methodology consists of three key steps: (1) identifying risk factor topics; (2) visualizing topics, documents, and terms; and (3) revealing dynamic features of the topics. The effectiveness and practicality of this approach were demonstrated using risk register data from 277 public works projects in Hong Kong, with a comparative analysis against traditional topic modeling techniques, such as latent Dirichlet allocation (LDA) and Top2Vec. This analysis, validated by a panel of project planning experts, successfully identified critical cost-related risk factors, such as design changes, market conditions, project delays, and underground conditions. The findings offer valuable insight for project planners, enabling more effective assessment and prioritization of cost risk factors in future construction projects.

• Analyzed 25,302 questions on Mocking from StackOverflow. • Applied LDA for topic modelling and pyLDAvis for topic visualizations. • Identified 30 topics, performed categorization, constructed topic hierarchy. • Analyzed category and topic-wise question trends, question types, Q&A popularity and difficulty. Mocking is a common unit testing technique that is used to simplify tests, reduce flakiness, and improve coverage by replacing real dependencies with simplified implementations. Despite its widespread use in Open Source Software (OSS) projects, there is limited understanding of how and why developers use mocks and the challenges they face. In this study, we have analyzed 25,302 questions related to Mocking on StackOverflow to identify the challenges faced by developers. We have used Latent Dirichlet Allocation (LDA) for topic modeling, identified 30 key topics, and grouped the topics into five key categories. Consequently, we analyzed the annual and relative probabilities of each category to understand the evolution of mocking-related discussions. Trend analysis reveals that categories such as Mocking Techniques and External Services have remained consistently dominant, highlighting evolving developer priorities and ongoing technical challenges. While the questions on Theoretical category declined after 2010, posts regarding Error Handling grew notably from 2009. Our findings also show an inverse relationship between a topic’s popularity and its difficulty. Popular topics like Framework Selection tend to have lower difficulty and faster resolution times, while complex topics like HTTP Requests and Responses are more likely to remain unanswered and take longer to resolve. Additionally, we evaluated questions based on the answer status- successful, ordinary, or unsuccessful, and found that topics such as Framework Selection have higher success rates, whereas tool setup and Android-related issues are more often unresolved. A classification of questions into How, Why, What , and Other revealed that over 64 % are How questions, particularly in practical domains like file access, APIs, and databases, indicating a strong need for implementation guidance. Why questions are more prevalent in error-handling contexts, reflecting conceptual challenges in debugging, while What questions are rare and mostly tied to theoretical discussions. These insights offer valuable guidance for improving developer support, tooling, and educational content in the context of mocking and unit testing.

The introduction of the new 30% conservation target under the Convention on Biological Diversity, along with emerging governance mechanisms for biodiversity protection in areas beyond national jurisdiction, has intensified the need to evaluate how marine protected areas (MPAs) are governed. While many assessment frameworks exist, each emphasizes different aspects of governance. This diversity underscores the importance of harmonizing these frameworks to ensure consistent and comparable evaluations. Using a systematic literature review combined with structural topic modeling, our study aims to identify common criterion topics shared across existing assessment frameworks for MPA governance. We uncovered 15 recurring topics, with “stakeholder participation” being the most prevalent, highlighting an equity dimension of governance. Based on the correlations among the 15 topics, we organized them into five overarching categories: (1) management resources; (2) management measures or day-to-day management activities; (3) social-ecological system (SES) approaches; (4) equity practices; and (5) cross-cutting measures such as cooperation and coordination. Management resources form the foundation that enables effective management activities. Through these routine activities, SES approaches are integrated and equity practices are strengthened, ultimately contributing to social, ecological, and equity-oriented conservation outcomes. Cross-cutting measures, relevant across all categories, further reinforce the quality and coherence of governance. Together, these five categories capture the core elements of effective MPA governance. The 15 topics provide a set of standardized criteria that can be used to assess and compare MPAs under different governance structures and social-ecological conditions. Establishing such standardized criteria is essential for providing coherent evaluation schemes for MPA governance. • Adequate human and financial resources are fundamental to MPA management. • Solid MPA management upholds effective MPA governance. • Integrating social-ecological system approaches into MPA governance is essential. • Equity is a core component of the social-ecological system. • Standardized criteria for assessing MPA governance help advance conservation efforts.

Document modelling using topics based on meaningful-interesting patterns for information filtering

Traditional topic modeling methods, such as Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF), are limited by their context ignorance, static nature, and low interpretability.Building upon the hybrid approach LDA+NMF+class-based Term Frequency-Inverse Document Frequency (c-TF-IDF), a new formalized framework -Dynamic Contextual Topic Modeling with Large Language Model (LLM) Refinement (DCTM-LLM) -is presented.This LLM-refined framework integrates transformer embeddings for the detection of dynamic semantic clusters and leverages an LLM for their subsequent refinement and the synthesis of high-level narratives.Experiments on a corpus of 35,000 arXiv abstracts (cs.AI (Computer Science -Artificial Intelligence), 2015-2025) showed that DCTM-LLM achieves a Normalized Pointwise Mutual Information (NPMI) of 0.53, a Silhouette score of 0.62, an Adjusted Rand Index (ARI) of 0.55, and Topic Diversity at 10 of 0.88.Crucially, with a Bidirectional Encoder Representations from Transformers (BERT)-based score (BERTScore) F1 of 0.89, the method significantly outperforms Dynamic BERTopic (0.62) and the hybrid LDA, NMF, and c-TF-IDF approach (0.65).Thus, the proposed approach shifts the paradigm of topic modeling from keyword extraction toward automated knowledge synthesis.

Abstract - Online platforms such as YouTube generate vast volumes of user-generated textual content that reflect public opinion, emotions, and behavioural patterns. Analysing this content is essential for understanding audience sentiment, dominant discussion themes, and the presence of toxic or harmful speech. Existing Natural Language Processing (NLP) approaches typically address sentiment analysis, topic modeling, and toxicity detection as independent tasks, results in fragmented insights and increased system complexity. This paper presents The Public Pulse, an integrated and interpretable NLP framework that unifies sentiment analysis, topic modeling, and toxicity detection within a single analytical pipeline. The system processes YouTube comments using lexicon-based sentiment analysis, Latent Dirichlet Allocation (LDA) for topic extraction, and rule-based toxicity detection, with results visualised through an interactive Streamlit dashboard. Experimental results demonstrate that the proposed approach provides coherent insights into public discourse while remaining computationally efficient and suitable for academic and resource constrained environments. Key Words: Sentiment Analysis, Topic Modeling, Toxicity Detection, YouTube Comments, NLP, LDA, TextBlob

Abstract - Text visualization plays a critical role in extracting meaningful insights from large textual datasets. Among various techniques, word clouds, word-level visualizations, and document visualization methods are widely used for summarization, exploration, and knowledge discovery. This paper presents a comprehensive review of these techniques, covering their methodologies, design principles, applications, advantages, and limitations. The study also highlights recent advancements such as semantic word clouds, interactive visualization systems, and graph-based document representations. Finally, future research directions are discussed to enhance interpretability and analytical effectiveness. Key Words: text visualization, word cloud, tag cloud, semantic visualization, document visualization, visual analytics, natural language processing (NLP), topic modeling, word embeddings, graph visualization

Abstract This study systematically examines the thematic and conceptual evolution of the Potato Research journal between 1970 and 2024 using a multi-layered text mining and topic modelling framework. A total of 1967 articles were analyzed across titles, abstracts, and keywords to capture surface-level and latent themes. Word frequency analysis, trend analysis, co-occurrence networks, and thematic mapping were integrated with Latent Dirichlet Allocation (LDA) and Structural Topic Modeling (STM) to identify dominant research themes and to evaluate their temporal dynamics. In addition, Sustainable Development Goals (SDGs) mapping was conducted using three independent SDG identification frameworks to assess the alignment of potato research with global sustainability agendas. The findings reveal a clear transformation in the journal’s scientific orientation, shifting from an early focus on agronomic production and plant pathology toward sustainability-oriented, climate-resilient, and data-intensive research paradigms. LDA identified five core thematic domains, namely post-harvest pathology, genetic resistance and molecular breeding, abiotic stress and physiological responses, plant growth and productivity, and agricultural management, which were further validated through STM-based inferential analysis. Temporal trends indicate statistically significant increases in themes related to climate change, water management, food quality, and analytical modelling, alongside a relative decline in conventional agronomic practices. SDG mapping demonstrates strong alignment with SDG 2 (Zero Hunger), SDG 3 (Good Health and Well-Being), and particularly SDG 13 (Climate Action). The findings highlight the role of Potato Research as both a historical record of disciplinary development and a scientific publishing platform reflecting sustainability-oriented agricultural research.

The Mediterranean Sea is both a global biodiversity hotspot and the world’s most heavily invaded marine region, where non-indigenous species arrivals are accelerating under intensifying shipping, Suez Canal traffic, aquaculture, and climate warming. Yet, despite rapidly growing research activity, a comprehensive synthesis of the scientific literature on Mediterranean marine invasions has been lacking. This study provides the first Mediterranean-wide combined bibliometric and topic-modeling analysis of invasive marine species research, using 3521 unique documents retrieved from Scopus and Web of Science. We quantify temporal growth in publications and citations, map the conceptual structure of the field through co-citation, co-word, and topic modeling, and reveal pronounced regional and thematic biases. Latent Dirichlet Allocation resolves 13 coherent topics, dominated by first records of non-native species, invasive macroalgae, alien species diversity, and ecological impacts, with strong signals for Lessepsian migration and climate-driven range shifts, particularly in the Eastern Mediterranean. Spatial and thematic analyses reveal pronounced regional biases, with invasion hotspots in the Aegean and Levantine seas contrasted by comparatively sparse coverage of western and central sub-basins, and notable gaps in predictive modeling and socioeconomic assessments. The results underscore the need to rebalance effort toward under-studied regions and themes, while leveraging existing collaboration networks and methodological advances to support MSFD (Marine Strategy Framework Directive) implementation, International Maritime Organization (IMO) instruments, and broader ecosystem-based management. The reproducible framework presented here offers a baseline for periodically tracking research evolution and guiding adaptive, transboundary governance of Mediterranean marine bio-invasions.

Understanding consumer behavior in the context of online shopping is critical for businesses to adapt to evolving market trends. Customer reviews serve as a rich source of information reflecting consumer sentiments and preferences. Sentiment analysis of these reviews has become a powerful tool to uncover underlying consumer emotions and purchasing trends. However, traditional methods relying on shallow lexical features and classical machine learning algorithms often fall short in capturing the intricate and contextual patterns present in textual data. In this study, we propose the use of the large language model RoBERTa-Large to enhance sentiment classification performance by imposing its advanced contextual embeddings and attention mechanisms. This approach enables the capture of complex semantic relationships beyond surface-level word frequencies. Alongside sentiment analysis, we apply topic modeling using Latent Dirichlet Allocation (LDA) on publicly available datasets to identify prevalent themes and topics within consumer feedback. We perform a comprehensive comparison of RoBERTa against traditional machine learning and ensemble models using TF-IDF features, as well as deep learning architectures utilizing sentence embeddings and transformer-based models. Experimental results demonstrate that RoBERTa-Large achieves the highest accuracy of 93.59%, significantly outperforming baseline models. To enhance model transparency and trustworthiness, we apply SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) interpretability techniques, providing meaningful explanations of model predictions at both global and local levels.