The Hang Seng University of Hong Kong

Artificial intelligence (AI) is transforming the preservation and research of historical newspapers by providing powerful tools that overcome longstanding challenges in terms of digitization, analysis, and access. This study offers a comprehensive global analysis of AI-driven innovations—including advanced Optical Character Recognition (OCR), Large Language Models (LLMs) for post-correction, and Natural Language Processing (NLP) techniques—that significantly enhance text extraction, image restoration, metadata generation, and semantic enrichment. Through qualitative case studies and comparative examinations of projects worldwide, this research demonstrates how AI not only improves the accuracy and efficiency of preservation workflows but also enables novel forms of computational inquiry such as cross-lingual analysis, sentiment detection, and discourse tracking. This study further explores emerging ethical and practical challenges and outlines future directions like multimodal analysis and collaborative digital infrastructures. The findings underscore AI’s transformative role in unlocking historical newspaper archives for both scholarly and public use, thereby fostering a deeper understanding of cultural heritage and historical narratives on a global scale.

T cells play a vital role in the immune system by identifying and eliminating infected or tumor cells through the recognition of antigen peptides presented by MHC. The strength of this immune response is directly influenced by the affinity of the T-cell receptor (TCR) for antigen peptides. Recent advancements in high-throughput sequencing technology have enabled the generation of large amounts of paired sequence data comprising TCRs and their corresponding bound antigen peptides. Although various deep-learning methodologies have been proposed to predict TCR-peptide binding, several challenges remain unsolved, including the absence of negative data, the difficulty in modeling the noncontributory residues, and the lack of model validation strategies. To address these challenges, we have developed a probabilistic generative model, termed the Sequence Interaction Model (SIM). This model is designed to predict sequence interactions and has been applied to TCR-peptide binding. SIM is trained solely on positive data sets and offers several advantages: it elucidates the binding preference through an interaction score matrix, generates peptides that bind specifically to a collection of TCRs and distinguishes between noncontributory residues and those essential for binding. We validated our model on the simulated dataset, paired sequence data and experimental structural data.

Does financial development increase income inequality? This topic is theoretically and practically significant given the surge in income inequality across most economies in recent decades and the costs associated with bailing out financial institutions that were ‘too big to fail’ during the 2008 global financial crisis. As the literature has yet to reach a consensus, this study examines the impact of financial development on income inequality by introducing regime type as a conditional factor. Although democracies are expected to be redistributive, the redistributive mechanisms are nullified or even reversed under the perverse influence of capital. Democratic governments may also be less capable than autocracies of limiting the excesses of financial development. By focusing on the interactive effect between financial development and democracy, financial development is shown to increase inequality in democracies but decrease it in non-democratic regimes. Supplementary analysis shows that the development of financial institutions, particularly their depth, drives the results.

Supply chain disruptions affect shipping service reliability, operations efficiency and carbon emissions. In disruptions like Red Sea Crisis, shipping lines struggled to anticipate the effectiveness of rerouting decisions while minimising shipping costs and emissions. A novel transitional response regression model (TRRM) within a three-stage high-fit predictive framework is developed, providing decision supports on rerouting and evaluating its environmental impacts. The model applies in the Red Sea Crisis with a longitudinal analysis across Asia-Europe trade services with rerouting via Cape of Good Hope, identifying key predictors, including vessel speed changes and number of port changes, from over fifty predictors. It provides improvement on a three-target measure of industry service reliability, container throughput and carbon emission mitigation by up to 85 %, with recommended reroutes with a lower emission increase rate from 47 % to 21 %. The model enhances emergency rerouting decisions with predictive insights in balancing service reliability, operations efficiency and emissions under the geopolitical uncertainty.

Purpose Prior research on peer-to-peer (P2P) lending and electronic word-of-mouth (eWOM) has overlooked how eWOM valence ambiguity and the external environment influence the relationship between eWOM volume and platform performance. Based on signaling and information processing theories, this study differentiates various types of eWOM valence ambiguities and examines how these ambiguities, along with the external environment, affect the impact of eWOM volume on P2P lending platform performance. Design/methodology/approach Using data from Chinese P2P lending platforms (January 2018–July 2019), econometric analyses evaluate the connection between eWOM volume and platform performance as well as the moderating effects of valence ambiguity and external uncertainty. Findings The findings confirm that eWOM volume significantly influences platform performance. However, this positive effect diminishes with high inconsistencies between star ratings and textual eWOM or increased external uncertainty. Research limitations/implications This study’s scope is limited to Chinese P2P lending. Future research should explore diverse markets and user engagement through experimental designs. Practical implications This study offers practical strategies for P2P lending platforms in emerging markets, emphasizing the importance of managing inconsistencies between star ratings and textual reviews and addressing lenders’ concerns in uncertain signaling environments. Originality/value This study advances the understanding of P2P lending and eWOM dynamics by analyzing the interplay between eWOM volume and internal and external factors related to eWOM diagnosability. It highlights how inconsistencies in star ratings and textual eWOM, coupled with external uncertainty, exacerbate decision-making challenges for potential lenders and reduce the positive impact of eWOM volume.

Service migration is crucial in mobile edge computing (MEC) to ensure seamless service provision as users move. Although some migration schemes have been proposed, they fail to efficiently support the migration of microservices in a directed acyclic graph (DAG)-based service across different edge servers, resulting in high service latency. This paper focuses on the DAG-based service migration problem and proposes a collaborative microservice migration framework for MEC, aiming to minimize the service migration latency while efficiently distributing the migration workload across edge servers. We divide edge servers into clusters and formulate the DAG-based service migration problem as a two-stage optimization problem. In the first stage, a deep reinforcement learning-based service pre-migration algorithm is developed to identify the optimal cluster of edge servers for hosting the migrated service. In the second stage, a microservice migration algorithm is devised, utilizing topological sorting and network flow techniques to further determine the target edge server for each microservice. Our design addresses the inherent dependencies among microservices within a DAG task and adapts well to dynamic network environments. Experimental results on real-world datasets demonstrate that our approach significantly reduces service migration latency.

Abstract The current study investigated the contributions of background factors (i.e., gender, age, and parents’ education), learning strategies, behavioural engagement in different domains (i.e., reading, writing, numeracy, and ICT), and cognitive skills (i.e., literacy and numeracy) to problem solving in technology-rich environments among tertiary students across 24 countries. The sample for this study comprised 12,148 tertiary students (M age = 25.68 years, 55% female). The results indicated that younger students exhibited better performance in problem solving within technology-rich environments. Furthermore, students whose parents had higher levels of education were more likely to demonstrate higher levels of problem-solving proficiency. Students who employed more effective learning strategies tended to have higher levels of problem-solving abilities. Higher levels of literacy and numeracy performance were associated with better problem-solving performance. Moreover, the study revealed a positive relationship between parents' education at the country level and students' problem-solving performance. This study further explored the similarities and differences in predicting the study variables related to problem solving. The implications of the study's findings were discussed in relation to learning, teaching, and policymaking in higher education. These findings can inform educational practices and policies to enhance problem-solving skills among tertiary students, ultimately improving their preparedness for technology-rich environments.

The rapid growth of Internet of Things (IoT) networks has introduced challenges in securing large-scale, real-time environments against evolving cyber threats. This study evaluates scalable machine learning workflows implemented in PySpark for intrusion detection using the RT-IoT2022 dataset. We compare manual feature engineering with automated pipeline-based approaches across classifiers including Logistic Regression, Naïve Bayes, Decision Tree, and Random Forest. Leveraging PySpark's distributed processing and modular components—such as Pipeline, StringIndexer, VectorAssembler, and MinMaxS-caler—we assess how workflow design affects performance metrics (Accuracy, Precision, Recall, and F1 Score), execution time, and model interpretability. Our findings reveal trade-offs between modularity, transparency, and latency, highlighting the need to align workflow architecture with deployment goals. The results provide practical insights for designing explainable, scalable, and resource-aware intrusion detection systems for real-time IoT security.