Selected work

Representative Papers

Ecosystem-Driven Privacy Exposure in Mobile Gaming Apps: A Configuration-Aware Empirical Analysis

May 25, 2026

This study addresses the limitations of permission-based privacy risk assessments in accurately capturing the actual privacy exposure of mobile games within complex third-party SDK ecosystems. To overcome the constraints of traditional permission-centric paradigms, the authors propose a configuration-aware static analysis framework that systematically evaluates privacy risks by integrating manifest configurations, exported components, and the structural characteristics of SDK ecosystems. Through SDK categorization and statistical analyses—including Spearman correlation, Mann-Whitney U tests, and chi-squared tests—the study reveals that children-targeted games exhibit privacy exposure levels comparable to those of general-audience games. Furthermore, it demonstrates that advertising SDKs and more extensive, diverse SDK ecosystems significantly amplify privacy leakage risks, underscoring the dominant role of ecosystem-level design in shaping privacy exposure.

0 citationsRead paper

An Empirical Analysis of Google Play Data Safety Disclosures: A Consistency Study of Privacy Indicators in Mobile Gaming Apps

Mar 25, 2026

This study addresses the reliability of Google Play’s Data Safety disclosures, which rely on self-reports from developers and lack empirical validation against apps’ actual privacy practices. Focusing on mobile games, the authors conduct the first fine-grained empirical analysis by statically examining 41 APKs to extract privacy-relevant behaviors involving device identifiers, location data, and personal information. They develop a structured evaluation framework and apply chi-square tests to assess consistency between declared and observed behaviors. The findings reveal high consistency for device identifier disclosures (87.8%) but a substantial inconsistency rate for location data (56.1%). No significant differences emerge between children’s apps and general-audience apps, underscoring a systemic transparency deficit across the mobile application ecosystem.

0 citationsRead paper

Beyond Permissions: An Empirical Static Analysis of Privacy and Security Risks in Children-Oriented and General-Audience Mobile Apps for Gaming

Feb 11, 2026

This study addresses the underexplored privacy risks in child-directed mobile games, which, despite requesting fewer permissions, exhibit significant vulnerabilities in manifest configurations and third-party SDK integrations. The authors propose a three-stage static analysis framework to systematically compare Android applications targeted at children against those for general audiences, examining declared permissions, manifest settings (such as backup enablement, cleartext traffic allowance, and component exportation), and embedded SDKs. For the first time, they establish a privacy risk taxonomy tailored specifically to child-directed apps. Their findings reveal that these apps often perform as poorly—or worse—than general-audience apps in terms of configuration security and use of tracking SDKs, underscoring the critical role of architectural and configurational choices in privacy protection and offering actionable guidance for developers.

0 citationsRead paper

LATA: A Tool for LLM-Assisted Translation Annotation

Feb 11, 2026

This work addresses the challenge of accurately capturing deep semantic transformations in language pairs with substantial structural divergence, such as Arabic–English, where existing automated tools often fall short. The authors propose an interactive translation annotation method grounded in large language models (LLMs), which employs a templated prompt manager to generate controlled, JSON-formatted sentence segmentation and alignment outputs. By integrating a human-in-the-loop verification process, the approach innovatively combines the scalability of LLMs with the precision of expert annotation. A marginalia-style architecture enables fine-grained labeling of translation strategies, significantly enhancing the quality of parallel corpora for complex language pairs while maintaining annotation efficiency. This framework effectively bridges the gap between fully automated systems and high-quality manual annotation.

0 citationsRead paper

Explainable Deep Learning for Pediatric Pneumonia Detection in Chest X-Ray Images

Jan 14, 2026

This study addresses the challenge of achieving both high accuracy and clinical trustworthiness in the automatic detection of pediatric pneumonia from chest X-ray images. By integrating EfficientNet-B0 and DenseNet121 architectures with transfer learning and data augmentation strategies, the proposed approach achieves high classification performance on a publicly available pediatric X-ray dataset. To enhance interpretability and clinical relevance, Grad-CAM and LIME techniques are employed to visualize the model’s decision-making process, confirming that predictions are grounded in clinically meaningful pulmonary regions. Experimental results demonstrate that EfficientNet-B0 yields the best performance, attaining an accuracy of 84.6%, an F1-score of 0.8899, a Matthews correlation coefficient (MCC) of 0.6849, and a recall exceeding 0.99. These findings underscore the model’s strong diagnostic capability and transparency, offering a reliable AI-assisted solution for pediatric pneumonia diagnosis.

0 citationsRead paper

Recent publications

Latest Papers

Ecosystem-Driven Privacy Exposure in Mobile Gaming Apps: A Configuration-Aware Empirical Analysis

May 25, 2026

This study addresses the limitations of permission-based privacy risk assessments in accurately capturing the actual privacy exposure of mobile games within complex third-party SDK ecosystems. To overcome the constraints of traditional permission-centric paradigms, the authors propose a configuration-aware static analysis framework that systematically evaluates privacy risks by integrating manifest configurations, exported components, and the structural characteristics of SDK ecosystems. Through SDK categorization and statistical analyses—including Spearman correlation, Mann-Whitney U tests, and chi-squared tests—the study reveals that children-targeted games exhibit privacy exposure levels comparable to those of general-audience games. Furthermore, it demonstrates that advertising SDKs and more extensive, diverse SDK ecosystems significantly amplify privacy leakage risks, underscoring the dominant role of ecosystem-level design in shaping privacy exposure.

0 citationsRead paper

An Empirical Analysis of Google Play Data Safety Disclosures: A Consistency Study of Privacy Indicators in Mobile Gaming Apps

Mar 25, 2026

This study addresses the reliability of Google Play’s Data Safety disclosures, which rely on self-reports from developers and lack empirical validation against apps’ actual privacy practices. Focusing on mobile games, the authors conduct the first fine-grained empirical analysis by statically examining 41 APKs to extract privacy-relevant behaviors involving device identifiers, location data, and personal information. They develop a structured evaluation framework and apply chi-square tests to assess consistency between declared and observed behaviors. The findings reveal high consistency for device identifier disclosures (87.8%) but a substantial inconsistency rate for location data (56.1%). No significant differences emerge between children’s apps and general-audience apps, underscoring a systemic transparency deficit across the mobile application ecosystem.

0 citationsRead paper

Beyond Permissions: An Empirical Static Analysis of Privacy and Security Risks in Children-Oriented and General-Audience Mobile Apps for Gaming

Feb 11, 2026

This study addresses the underexplored privacy risks in child-directed mobile games, which, despite requesting fewer permissions, exhibit significant vulnerabilities in manifest configurations and third-party SDK integrations. The authors propose a three-stage static analysis framework to systematically compare Android applications targeted at children against those for general audiences, examining declared permissions, manifest settings (such as backup enablement, cleartext traffic allowance, and component exportation), and embedded SDKs. For the first time, they establish a privacy risk taxonomy tailored specifically to child-directed apps. Their findings reveal that these apps often perform as poorly—or worse—than general-audience apps in terms of configuration security and use of tracking SDKs, underscoring the critical role of architectural and configurational choices in privacy protection and offering actionable guidance for developers.

0 citationsRead paper

LATA: A Tool for LLM-Assisted Translation Annotation

Feb 11, 2026

This work addresses the challenge of accurately capturing deep semantic transformations in language pairs with substantial structural divergence, such as Arabic–English, where existing automated tools often fall short. The authors propose an interactive translation annotation method grounded in large language models (LLMs), which employs a templated prompt manager to generate controlled, JSON-formatted sentence segmentation and alignment outputs. By integrating a human-in-the-loop verification process, the approach innovatively combines the scalability of LLMs with the precision of expert annotation. A marginalia-style architecture enables fine-grained labeling of translation strategies, significantly enhancing the quality of parallel corpora for complex language pairs while maintaining annotation efficiency. This framework effectively bridges the gap between fully automated systems and high-quality manual annotation.

0 citationsRead paper

Explainable Deep Learning for Pediatric Pneumonia Detection in Chest X-Ray Images

Jan 14, 2026

This study addresses the challenge of achieving both high accuracy and clinical trustworthiness in the automatic detection of pediatric pneumonia from chest X-ray images. By integrating EfficientNet-B0 and DenseNet121 architectures with transfer learning and data augmentation strategies, the proposed approach achieves high classification performance on a publicly available pediatric X-ray dataset. To enhance interpretability and clinical relevance, Grad-CAM and LIME techniques are employed to visualize the model’s decision-making process, confirming that predictions are grounded in clinically meaningful pulmonary regions. Experimental results demonstrate that EfficientNet-B0 yields the best performance, attaining an accuracy of 84.6%, an F1-score of 0.8899, a Matthews correlation coefficient (MCC) of 0.6849, and a recall exceeding 0.99. These findings underscore the model’s strong diagnostic capability and transparency, offering a reliable AI-assisted solution for pediatric pneumonia diagnosis.

0 citationsRead paper