Scopus 2025
Permanent URI for this collectionhttps://repository.li.mahidol.ac.th/handle/123456789/102712
Browse
Recent Submissions
Item Metadata only Re-constructing “Chineseness” in the Frontiers of Statehood, Memory, and Territory: The Kuomintang Communities of Northern Thailand(2025-01-01) Ohlendorf H.; Ohlendorf H.; Mahidol UniversityDiasporic identity is not merely inherited, but is actively constructed through memory, heritage, and negotiation with shifting geopolitical forces. Yet, little attention has been paid to how communities at the margins of statehood reimagine belonging through cultural memory. This study addresses that gap by examining a distinct Kuomintang diaspora community in Northern Thailand, where such identity work is particularly visible. It discusses the reconstruction of “Chineseness” in the small village of Mae Salong, which was founded by Chinese Kuomintang (KMT) soldiers fleeing the Chinese Civil War. The article examines how geopolitical events, cross-border migration, and memory have shaped this community’s identity at the frontiers of statehood and territory. Using theories of collective and cultural memory, the article analyzes key sites of memory, including the Chinese Martyrs’ Memorial and General Xi Duanwen’s tomb, to show how these communities have utilized memory and heritage to carve out new spaces of identity and belonging. The role of tourism in influencing these memory constructions is also emphasized, with a focus on how local traditions are displayed to meet tourists’ expectations. By analyzing how historical narratives and cultural practices are preserved, adapted, and reimagined in Mae Salong, this article offers insights into the broader dynamics of identity formation in Chinese diaspora communities.Item Metadata only Time Series Interpolation of Holiday Gaps for Enhanced Stock Price Prediction(2025-01-01) Mekon J.; Wongsuwarn H.; Radeerom M.; Vephasayanant A.; Supadol T.; Songmuang P.; Mekon J.; Mahidol UniversityDiscontinuities in financial time-series data caused by market holidays can significantly affect the accuracy of predictive models. This study explores interpolation strategies to address holiday-induced gaps in daily stock prices of eight major NASDAQ technology companies (2019-2024), using data from Yahoo Finance. Three approaches-linear interpolation, nearest interpolation, and no interpolation-were compared using machine learning models to forecast next-day closing prices. Input features included 7-day lagged price data and technical indicators such as SMA, EMA, and Bollinger Bands. Model performance was evaluated using MAE and R2. Results show that linear interpolation consistently yielded the lowest mean MAE, reducing error by 23.9% compared to no interpolation, and improved mean R2 by 1.15%, demonstrating higher explanatory power. Linear interpolation outperformed both alternative methods for every company examined, highlighting its substantial benefit for forecasting accuracy and model robustness in financial time-series prediction.Item Metadata only Pale-capped Pigeons Columba punicea Consuming the Fruits of Avicennia marina(2025-01-01) Round P.D.; Round P.D.; Mahidol UniversityItem Metadata only Hybrid Recommendation System for Visual Novels Using Content and Collaborative Filtering(2025-01-01) Unjindamanee S.; Poopradit J.; Kaewmanee P.; Sujipisut P.; Phienthrakul T.; Hunchangsith K.; Unjindamanee S.; Mahidol UniversityThis paper presents a hybrid recommendation system for visual novel discovery, integrating content-based filtering and collaborative filtering within the Machine Learning Visual Novel Recognition and Recommendation (MLVNR2) platform. The system combines hierarchical tag metadata and user interaction histories to generate a ranked list of relevant visual novels. The hybrid strategy blends content and collaborative signals while safeguarding semantic relevance. A user study involving thirty-one participants achieved a perfect top-3 hit rate and an average top-10 reciprocal rank of 98.4%, while additionally revealing strong novelty discovery, with 83.9% of recommendations introducing users to unfamiliar titles. More than 80% of participants reported positive satisfaction. In terms of precision at cut-offs, the system delivered 84.9% for the top three results, 76.8% for the top five, and 61.6% for the top ten. These findings demonstrate the effectiveness of combining semantic metadata similarity with user behaviour patterns for personalized visual novel recommendation.Item Metadata only The Impact of Self-Regulated Learning on Achieving Intended Learning Outcomes(2025-01-01) Huu P.N.; Tangworakitthaworn P.; Gilbert L.; Huu P.N.; Mahidol UniversityThe primary learning objective is to equip learners with the knowledge and necessary skills to achieve the course's intended learning outcomes (ILOs). Previous studies have not comprehensively introduced a methodology for assessing learners' competence by comparing actual learning outcomes with the intended learning outcomes (ILOs). This study presents experimental results that evaluate ILO achievement by comparing actual learning outcomes with the course's ILOs, using adaptable learning paths generated based on Bloom's cognitive taxonomy and Biggs' SOLO structure. The research involved an experimental study with seventy-eight voluntary Bachelor of Science in Information Technology students. Participants accessed learning through the Self-Regulated Learning Management System, a new platform built on the ILO-based Self-Assessment Model. The findings showed that the participants who received regular updates on their ILO achievement were satisfied with the system. Statistically, a one-sample Wilcoxon signed-rank test demonstrated that the median differed significantly from the expected median, z = 7.35, p<0.01, with a strong effect size (r = 0.83).Item Metadata only Modeling the Interplay of Ghrelin, Leptin, and Exercise on Energy Balance: A Systems Dynamics Approach(2025-01-01) Chudtong M.; Nagy I.; De Gaetano A.; Chudtong M.; Mahidol UniversityWe detail a possible mathematical model describing the dynamics of key energy substrates, regulatory hormones, appetite, and fat mass incorporating the effect of exercise. The model describes physiological responses to three mixed meals (breakfast, lunch, dinner), represented by difference-ofexponential influx functions. The size of these meal pulses is modulated by appetite, which is dynamically regulated by ghrelin and leptin concentrations. The system is formulated as set of coupled nonlinear algebraic and ordinary differential equations (ODEs), including state variables for appetite, gut glucose and lipid contents, glycemia, lipidemia, insulinemia, plasma ghrelin and leptin concentrations, weight and fat mass, as well as the effect of exercise.Item Metadata only Agentic Stage-Based LLM Framework for Multi-Turn Mental Health Support Conversations in Thai(2025-01-01) Changpun R.; Khoprasertthaworn N.; Jongpipatchai P.; Petcharat T.; Rungsimontuchat K.; Suttiwan P.; Nupairoj N.; Hemrungrojn S.; Tuicomepee A.; Achakulvisut T.; Vateekul P.; Changpun R.; Mahidol UniversityWhile mental health support needs are growing in Thailand, access to professional support remains limited. LLMbased chatbots offer a scalable solution, yet most existing systems are restricted to single-turn, solution-oriented responses. We propose an Agentic Stage-Based LLM Framework for Multi- Turn Mental Health Support Conversations in Thai, structuring conversations around five core counseling stages: rapport building, problem identification, goal setting, working, and termination. Drawing from Person-Centered Therapy (PCT) and Acceptance and Commitment Therapy (ACT), our framework incorporates two key components: an approach-selection agent that selects appropriate counseling approaches and a monitoring agent that manages stage transitions in multi-turn conversations. We evaluated the proposed framework against three single-agent baselines using LLM-simulated users and compared positive user reaction rates as judged by LLMs. Our framework achieved a 79.01% positive user reaction rate, outperforming single agents with standard, AugESC, and COOPER-CoT prompts by 8.91%, 11.68%, and 17.02%, respectively. Ablation studies validated the necessity of each agentic module in the proposed framework, with results surpassing single-agent baselines without stage-based architecture by 96.15% and 88.46% in Process and Working evaluations, respectively. A/B testing with real users and evaluation by 3 counseling practitioners demonstrated significant improvements in seven of eight mental health support evaluation metrics, highlighting the potential to deliver LLM-based mental health support in Thai.Item Metadata only ThonburianTTS: Enhancing Neural Flow Matching Models for Authentic Thai Text-to-Speech(2025-01-01) Aung T.; Sriwirote P.; Thavornmongkol T.; Pipatsrisawat K.; Achakulvisut T.; Aung Z.H.; Aung T.; Mahidol UniversityWe introduce ThonburianTTS, a finetuned Thai text-to-speech (TTS) system based on the E2-TTS and F5-TTS architectures, designed to improve pronunciation accuracy, alignment robustness, and zero-shot speaker adaptation for the Thai language. Our models are trained on both Thai script and International Phonetic Alphabet (IPA) transcriptions to evaluate the impact of phonetic input on synthesis quality. We evaluate performance using objective metrics, including Word Error Rate (WER), Syllable Error Rate (SylER), Character Error Rate (CER), Speech Naturalness (MOSNet), Speaker Similarity (SIM-O) and Synthesis Speed (RTF). Our best model, F5-TTS trained on Thai script, achieves a WER of 25.72 %, SylER of 14.17 %, CER of 8.70 %, a MOSNet score of 3.9451, and a SIM-O score of 88.30 %. While IPA-based models yield comparable or higher scores in naturalness and speaker similarity, they underperform in accuracy-related metrics such as WER, SylER, and CER. We also show that increasing the Number of Function Evaluations (NFE) leads to improved model accuracy. ThonburianTTS outperforms strong baselines such as MMS-TTS and PyThaiTTS in both intelligibility and speaker similarity, highlighting the effectiveness of flow matching-based architectures for high-quality TTS in tonal, low-resource languages like Thai. The code and pretrained models are available at https://github.com/biodatlab/thonburian-tts.Item Metadata only Optimizing BERT for Sentiment Classification of Amazon Product Reviews: A Study on Class Imbalance and Misclassification Analysis(2025-01-01) Korsakpaisarn O.; Noraset T.; Lapamnuaypol J.; Jin'no K.; Korsakpaisarn O.; Mahidol UniversityThis paper presents a custom BERT-based sentiment classification pipeline for Amazon product reviews. Reviews are grouped into three sentiment classes - bad (1-2 stars), normal (3 stars), and good (4-5 stars). Two practical challenges are addressed: the semantic ambiguity of the neutral ('normal') class and skewed class distributions that bias prediction toward majority categories. In particular, the underrepresentation of neutral reviews often yields models that favor the dominant classes, limiting effectiveness in real applications. To mitigate these issues, stratified sampling and class-weighted cross-entropy are applied while fine-tuning bert-base-uncased. A moderate emphasis on the normal class increases its recall and F1 without significantly lowering overall accuracy, according to multi-run trials across four weight settings.Item Metadata only Through an AI's Looking Glass: Discovering Dental Sexual Dimorphism with Explainable AI(2025-01-01) Hirunchavarod N.; Sributsayakarn N.; Pornprasertsuk-Damrongsri S.; Jirarattanasopha V.; Intharah T.; Hirunchavarod N.; Mahidol UniversityWe propose a framework that extends explainable AI (XAI) from per-instance interpretation to dataset-wide knowledge discovery in dental morphology. A deep convolutional neural network was trained on 5,132 panoramic radiographs from 2,778 individuals to predict sex. Using OPG-SHAP, a domain-specific XAI method, we identified influential oral parts and validated them with statistical analysis. The upper canine was the most influential region, with females showing a significantly higher width-to-height ratio (0.391) than males (0.347), aligning with existing literature. Additionally, the upper third molar emerged as a novel sexually dimorphic feature, with males showing a higher ratio (1.064) than females (1.036). Both differences were statistically significant (p<0.001). Our results demonstrate how interpretable AI can rediscover known anatomical patterns and reveal new insights, enabling clinically meaningful knowledge extraction from neural networks.Item Metadata only YOLO-Based Sequential Image Approach for Improved Smoke Detection in Early-Stage Wildfires(2025-01-01) Thamrongweingpung S.; Usanavasin S.; Khowyabud N.; Pipatanagovit P.; Kamonsuphathawat P.; Ngamsukhonratana N.; Limpacharoenkul R.; Galajit K.; Karnjana J.; Thamrongweingpung S.; Mahidol UniversityWildfires in northern Thailand, particularly during the dry season from November to May, are a major source of hazardous air pollution and sharp increases in PM2.5 levels. Despite no-burning policies, conventional monitoring methods such as lookout towers remain inadequate due to limited visibility, staffing shortages, and delayed response times. Existing smoke detection models mostly rely on single-frame analysis, which often misclassifies fog or clouds as smoke and fails to detect faint or obscured smoke plumes. This reduces the reliability of earlywarning systems in real-world deployment. The objective of this study is to evaluate whether time-series analysis can improve detection performance over single-frame baselines. We fine-tuned a YOLOv8n model on the FireSpot dataset and designed a sequential evaluation protocol that aggregates predictions across short image sequences, converting per-frame outputs into eventlevel classifications. This temporal consistency filters out transient false positives while preserving true smoke evidence. Experimental results show that per-image classification achieved a precision of 0.9780, a recall of 0.8230, an F1-score of 0.8940, and a balanced accuracy of 0.9020. The sequential framework improved performance with perfect precision of 1.0000, a recall of 0.8970, an F1-score of 0.9460, and balanced accuracy of 0.9490. These results confirm that temporal modeling significantly enhances robustness and reliability, providing a practical pathway for early-warning deployment in wildfire-prone regions.Item Metadata only Chai-Calibrated Hybrid Assessment for IELTS Speaking with Human-Referenced Validation(2025-01-01) Polasa P.; Laoaree S.; Thanadunpremdet T.; Rodjananant N.; Kritsuthikul N.; Polasa P.; Mahidol UniversityWe present CHAI, a rubric-aligned framework for IELTS Speaking that combines an accent-aware ASR backbone with self-supervised speech representations to deliver criterion level feedback. CHAI adopts a dual-agent design: a low-latency Coach for live turn-taking (Whisper-TH large) and a read-only Judge for independent scoring (Whisper-base). Evidence integrates pronunciation similarity from HuBERT-style embeddings with alignment/timing cues, prosody, and transcript-derived indicators to estimate bands for Fluency & Coherence, Lexical Resource, and Grammatical Range & Accuracy. Two certified IELTS examiners Expert A and B) and a small crowd panel (Crowd mean) serve as human references in a classroom-style evaluation with Thai EFL learners across three role-play scenarios (restaurant, airport, job interview). Agreement is reported on the band scale using mean absolute error (MAE) as the primary metric, with latency tracked for usability. The hybrid fusion with a lightweight human prior yields the lowest overall MAE (0.410), outperforming single-model baselines and individually considered human references (Expert A: 0.430; Expert B: 0.451; Crowd mean: 0.512); per-criterion MAE likewise favors the hybrid (F/C 0.409, LR 0.402, GRA 0.418). Latency supports near-real-time classroom feedback for ∼ 10 s turns. Despite a focus on a Thaicentric corpus and sensitivity to ASR timing and fairness, results indicate that model-human hybridization is a practical pathway to consistent, scalable IELTS-aligned feedback.Item Metadata only Automated Monitoring of Motorcycle Helmet Usage for Road Safety(2025-01-01) Rodthanong P.; Ploypicha P.; Pleangnoi S.; Sivaraksa M.; Rodthanong P.; Mahidol UniversityMotorcycle safety is a major concern, with helmet usage being a critical factor in reducing severe injuries during accidents. Continuous monitoring of helmet compliance in hightraffic areas is challenging due to the limitations of manual observation. This paper presents an automated helmet detection and monitoring system that combines the YOLOv8 object detection model with BoT-SORT tracking to process CCTV footage in real time. Two detection models are employed: one for motorcycles and pedestrians within a Region of Interest (ROI), and another for helmets and non-helmeted heads within a counting zone. BoTSORT ensures consistent object identities across frames, reducing overcounting. The system integrates with Google Sheets and Drive APIs for automated logging and reporting of compliance data. Experimental evaluation demonstrates high precision and recall, with challenges primarily arising from lighting variations and object occlusion. Overall, the proposed system provides a reliable framework for large-scale helmet compliance monitoring and road safety enforcement.Item Metadata only Integration of Cross-Enterprise Document Sharing (XDS.b) with Blockchain Technology(2025-01-01) Julled P.; Khurat A.; Mongkolwat P.; Julled P.; Mahidol UniversityHealthcare information sharing and interoperability are vital for ensuring healthcare quality and safety, especially for patients receiving services from multiple providers. Integrating the Healthcare Enterprise (IHE) has developed the Cross-Enterprise Document Sharing-b (XDS.b) profile to facilitate seamless exchange of health documents. However, ensuring security, including data integrity, availability, and privacy, remains a significant challenge. Presently, no endorsed security implementations exist specifically for XDS.b. To address these cybersecurity concerns and enhance health information sharing, a novel method utilizing blockchain technology has been implemented demonstrated and freely available. This method guarantees the integrity and availability of health information, enabling decentralized document sharing while effectively addressing cybersecurity issues through the inherent security features of blockchain technology.Item Metadata only Comparative Analysis of Data Imputation Methods on F1 Performance Across Multiple Classification Algorithms(2025-01-01) Tangworakitthaworn P.; Fujita K.; Wiphaalongkot N.; Tangworakitthaworn P.; Mahidol UniversityThe significant issue of developing the machine learning is quality and completeness of the datasets. Therefore, the suitable datasets should not have the missing values because these can lead to reducing the predictive accuracy and introducing the bias. This research project aims to evaluate the comparative analysis of the data imputation methods on F1 performance across multiple classification algorithms, which are Logistic Regression, Random Forest, and Linear Support Vector Machine (SVM). Moreover, the imputation applied on this project are divided into 5 modes which are Mode1: imputed by AI without data description, and this mode will impute the missing data by random imputation, Mode2: imputed by AI with data description, and this mode will impute the missing data by Model-based (iterative) imputation, Mode3: imputed by mean algorithm, Mode4: imputed by KNN algorithm, and Mode5: imputed by median algorithm. The datasets used for the comparative analysis cover the different size of missing data, ranging from 50,000 to 200,000 missing entries. As a result, the research findings revealed that the data imputation method using Mode2 (AI with Data Description) was the most effective for high percentages of missing data, while the data imputation method using Mode1 (AI without Data Description) was the least effective.Item Metadata only Assessing the NGINX Server's Configuration Security Based on CIS Benchmarks(2025-01-01) Lekcharuthas G.; Khurat A.; Choetkiertikul M.; Ragkhitwetsagul C.; Lekcharuthas G.; Mahidol UniversityWebsites and applications commonly rely on web server software such as NGINX to handle server-side tasks. Administrators often copy configuration files of these servers from online sources (e.g., GitHub) and adapt them, but these files can be misconfigured and introduce security vulnerabilities. This paper presents an automated tool that assesses NGINX configuration files against the CIS Benchmark for NGINX by the Center for Internet Security (CIS). We categorized benchmark recommendations applicable to configuration files, implemented the tool, and evaluated it on 23 popular NGINXbased GitHub repositories. On average, only about 4.01% of scannable recommendations were implemented; configurations for logging and encryption were absent from defaults. These findings raise concerns for developers adopting such files without thorough review. Our evaluation shows that the tool can be used to identify insecure or missing configurations in online-sourced configurations and promotes best practices of having secure configurations for a stronger security posture.Item Metadata only Apply Aspect-Based Sentimental Analysis on Course Evaluation(2025-01-01) Kraisangka J.; Noraset T.; Kertkeidkachorn N.; Kraisangka J.; Mahidol UniversityCourse evaluations provide valuable insights into teaching effectiveness; however, analyzing open-ended feedback is challenging due to its qualitative nature and the scale of the responses. This study employs Aspect-Based Sentiment Analysis (ABSA) to analyze student evaluations from an international undergraduate program, with a focus on the role of data augmentation. We compare three methods: Back-Translation, Paraphrasing, and Generative AI, under transfer and non-transfer learning using BART-Large-CNN and LoRA Llama3.2-3B-Instruct. Models are evaluated with 5-fold cross-validation on both original and augmented datasets. Results show that Back-Translation yields the most consistent improvements for BART-Large-CNN, raising accuracy and F1 by approximately 2%. For Llama, Generative AI performs best in the non-transfer setting, while Back-Translation is more effective with transfer learning. These findings highlight the value of data augmentation in enhancing ABSA for educational feedback and guide on applying NLP to large-scale course evaluation.Item Metadata only Bridging FLEx and Computational Linguistics: A Web-Based System for Bidirectional Conversion and FAIR-Compliant Interoperability(2025-01-01) Limpisiri T.; Leenoi D.; Panutatpinyo R.; Chongkolrattanapond P.; Limpisiri T.; Mahidol UniversityFieldWorks Language Explorer (FLEx) is widely used for documenting under-resourced languages, but its XML based formats (SFM, LIFT) hinder integration with computational workflows that rely on tabular data. This paper introduces a web-based system for near-lossless, bidirectional conversion between FLEx formats and CSV/Excel. The tool provides configurable field mapping, validation diagnostics, and support for media references, enabling linguists to edit data in familiar tabular form and reimport it into FLEx with minimal technical overhead. Evaluation across five Thai lexical resources (110k+ entries) demonstrates consistent structural fidelity, efficient conversion, and successful preservation of media references. A user study with ten participants highlights strong usability and reliability, though challenges remain in handling complex LIFT structures, restricted audio integration, and large-scale corpora. By lowering barriers for non-specialist users and aligning with FAIR data principles, this work establishes a practical pathway for integrating FLEx resources into broader NLP pipelines.Item Metadata only Human-mediated dispersal of Geniotrigona thoracica (Apidae: Meliponini) colonies promotes high genetic diversity and reduces population structuring in managed populations(2025-12-10) Duangphakdee O.; Jeratthitikul E.; Poolprasert P.; Pongkitsittiporn R.; Inson C.; Rattanawannee A.; Duangphakdee O.; Mahidol UniversityThe stingless bee Geniotrigona thoracica is a key managed pollinator in Southeast Asia, valued for its honey, propolis, and colony trade. In Thailand, frequent humanmediated movement of colonies raises concerns about its effects on genetic diversity and population structure.Weanalysed variation in mitochondrial (COI and 16S rRNA) and nuclear (five microsatellite loci) markers from 70 colonies sampled across 17 meliponaries in seven southern provinces. Microsatellite data revealed high genetic diversity and low nuclear differentiation (K D1; Fst D0.0024-0.1219; all P >0.05), with extensive gene flow (Nm D 3.60-207.83) among provinces. In contrast, mitochondrial markers indicated moderate-to-high differentiation (Fst D0.619), consistent with mitonuclear discordance arising from sex-biased. Managed colonies exhibited elevated heterozygosity and allelic richness, likely reflecting admixture from colony exchange, while unique haplotypes in certain provinces suggest introductions from external sources. Significant inbreeding was detected only in Yala, possibly linked to habitat loss and reduced effective population size. Our findings indicate that current meliponicultural practices maintain high genetic diversity in G. thoracica despite mitochondrial structuring, but increasing colony movement between genetically distinct populations may risk erosion of local adaptations, underscoring the need for genetic screening prior to translocation.Item Metadata only Thai Tune: Enhancing Communication for Deaf through Technological Integration(2025-01-01) Deeprom R.; Aiumbhornsin K.; Sihabut B.; Noraset T.; Liamruk P.; Deeprom R.; Mahidol UniversityDeaf individuals in Thailand face communication barriers, particularly in mastering written Thai, due to a lack of tailored tools for vocabulary and grammar improvement. This project addresses these challenges by developing a mobile app that enhances Thai language proficiency, facilitating communication between Deaf and hearing individuals. Utilizing NLP and intuitive design, the app features synonym suggestions, sentence restructuring, and mood adjustments to improve clarity. Motivated by the goal of social inclusion, Thai Tune aims to bridge the communication gap, empowering Deaf individuals and fostering inclusivity in social and professional contexts.