Existing deep reinforcement learning (DRL) agents suffer from severe reward instability during late-stage exploration, particularly when encountering novel states in complex continuous environments. A variety of existing studies focus on improving an agent's reward exploration. However, they ignore the instability problem that arises when the agent faces new states in the later stages of exploration. This paper proposes a novel companion learning network (CLN) based on the idea that the guidance can accelerate human learning efficiency and reduce the risk of making mistakes. The CLN integrates a short-term partner network to intensively learn localized environmental patterns, offering adaptive action guidance for recent states. Simultaneously, a global Qnetwork dynamically incorporates the partner's decaying guidance signals, balancing autonomous exploration with error mitigation. As training progresses, the partner's influence gradually diminishes, allowing the Qnetwork to solidify robust policies without persistent dependence. Extensive experiments on four OpenAI Gym environments demonstrate that the CLN can significantly improve the exploration stability in most tested scenarios, achieving up to 49% reduction in late-stage reward standard deviation compared to baseline DRL methods.

Accurately predicting traffic conditions is vital for smart city development, yet it remains challenging due to the intricate spatio-temporal dependencies in road networks. Existing works often propose intra-mixing deep learning-based prediction models for individual nodes and share parameters among them or spatial intermixing deep learning-based models for traffic predictions. However, these approaches may neglect essential principles of information exchange in traffic flow or capture useless or even erroneous spatio-temporal dependencies. To address these limitations, we propose a Mixing Spatio-Temporal neural network (MiST) for enhancing traffic predictions. In MiST, we propose (i) a temporal encoder that embeds the traffic data along with periodic features, (ii) a spatial encoder that embeds the positional information in graph and hypergraph spectral domains, as well as spatial node identities, and (iii) a mixing spatio-temporal encoder that merges the diverse features provided by the temporal and spatial encoders. Our empirical evaluations on real-world traffic prediction tasks, including flow and speed predictions, validate the superiority of MiST, underscoring its innovative contribution to traffic prediction methodologies.

Recommendations are important web services in the era of information explosion. Particularly, group recommendations aim to suggest new items to groups such that the members of groups are likely interested in. However, existing works still suffer from sparsity and cold-start issues (e.g., cold-start groups or items) for groups with few interactions on items. Most of them model the preferences or features of entities (i.e., users, items and groups) from heterogeneous interactions (i.e., user-item, group-item and user-group interactions) between two distinct types of entities, while ignoring the homogeneous interactions (i.e., user-user, item-item and group-group interactions) between entities of one type. To this end, we propose a new model, called H3Rec, which learns the representations of entities by developing two graph embedding layers based on an interaction graph of all entities. Specifically, the two graph embedding layers make full use of the hidden information in the H igher-order H eterogeneous and H omogeneous interactions of the graph. Therefore, H3Rec can alleviate the sparsity and cold-start issues and improve the performance of group recommendations. The experimental results on two real world datasets in different domains show the superiority of H3Rec in group recommendations, especially for cold-start groups and items.

Medical ultrasound technology has garnered significant attention in recent years, with Ultrasound-guided regional anesthesia (UGRA) and carpal tunnel diagnosis (CTS) being two notable examples. Instance segmentation, based on deep learning approaches, is a promising choice to support the analysis of ultrasound data. However, many instance segmentation models cannot achieve the requirement of ultrasound technology e.g. real-time. Moreover, fully supervised instance segmentation models require large numbers of images and corresponding mask annotations for training, which can be time-consuming and labor-intensive in the case of medical ultrasound data. This paper proposes a novel weakly supervised framework, CoarseInst, to achieve real-time instance segmentation of ultrasound images with only box annotations. CoarseInst not only improves the network structure, but also proposes a two-stage "coarse-to-fine" training strategy. Specifically, median nerves are used as the target application for UGRA and CTS. CoarseInst consists of two stages, with pseudo mask labels generated in the coarse mask generation stage for self-training. An object enhancement block is incorporated to mitigate the performance loss caused by parameter reduction in this stage. Additionally, we introduce a pair of loss functions, the amplification loss, and the deflation loss, that work together to generate the masks. A center area mask searching algorithm is also proposed to generate labels for the deflation loss. In the self-training stage, a novel self-feature similarity loss is designed to generate more precise masks. Experimental results on a practical ultrasound dataset demonstrate that CoarseInst could achieve better performance than some state-of-the-art fully supervised works.

Most existing deep learning approaches, such as instance segmentation, are for natural images only. However, due to the unique characteristics of medical ultrasound images, they may not be suitable for ultrasound image diagnosis. In this study, we introduce Boundmask, an instance segmentation framework specially designed for medical ultrasound median nerve images. In Boundmark, firstly, we propose the nesting attention module (NAM), which combines spatial and channel attention to enhance the feature information so that we can still get rich feature information even with a simple backbone. Secondly, we design a boundary-guided segmentation mechanism (BGSM) that considers the object’s unique traits and border information while segmenting. The experiments conducted using clinical data demonstrate that Boundmask has a high practical value. The results show that it achieves 54.2 AP on the ultrasound median nerve image dataset and outperforms most existing instance segmentation models.

Numerous transformer-based medical image segmentation methods have been proposed and achieved good segmentation results. However, it is still a challenge to train and deploy transformer networks to mobile medical devices due to a large number of model parameters. To resolve the training and model parameter problems, in this paper, we propose a Transformer-based network for Medical Image Segmentation using Knowledge Distillation named MISTKD. The MISTKD consists of a teacher network and a student network. It achieves comparable performance to state-of-the-art transformer works using fewer parameters by employing the teacher network to train the student network. The training can be implemented by extracting the sequence in the teacher and student encoder networks during the training procedure. The losses between sequences are further calculated, thus the student network can learn from the teacher network. The experimental results on Synapse show that the proposed work achieves competitive performance using only one-eighth parameters.

The problem of traffic speed predictions is still very challenging due to the complex and dynamic urban traffic conditions. Many existing works have implied the importance of integrating spatial correlations into models to explore nonlinear spatio-temporal dependencies and make traffic predictions in near future. However, some of the works only consider pairwise correlations and cannot model the hidden information among multiple nodes well, and the others only consider hyper-correlations (that can be shared by more than two nodes) and discount the role of the pairwise ones for propagating spatial dependencies. Therefore, we propose a Spatio-Temporal neural nEtwork based on both Pairwise and Hyper-correlations (STEPH) for traffic speed predictions. It is distinguished primarily by incorporating both types of spatial correlations into temporal information and designing new hybrid spatio-temporal blocks in neural networks to effectively overcome the challenge. Experiments on two real-world traffic datasets demonstrate the effectiveness of the proposed model, and show its superiority performance compared to other state-of-the-art baselines.

The flexible job shop scheduling (FJSS) is important in real-world factories due to the wide applicability. FJSS schedules the operations of jobs to be executed by specific machines at the appropriate time slots based on two decision steps, namely, the job sequencing (i.e., the sequence of jobs executed on a machine) and the job routing (i.e., the route of a job to a machine). Most current studies utilize either deep reinforcement learning (DRL) or multi-agent reinforcement learning (MARL) for FJSS with a large search space. However, these studies suffer from two major limitations: no integration between DRL and MARL, and independent agents without cooperation. To this end, we propose a new model for FJSS, called DeepMAG based on Deep reinforcement learning with Multi -Agent Graphs. DeepMAG has two key contributions. (1) Integration between DRL and MARL. DeepMAG integrates DRL with MARL by associating a different agent to each machine and job. Each agent exploits DRL to find the best action on the job sequencing and routing. After a job-associated agent chooses the best machine, the job becomes a job candidate for the machine to proceed to its next operation, while a machine-associated agent selects the next job from its job candidate set to be processed. (2) Cooperative agents. A multi-agent graph is built based on the operation relationships among machines and jobs. An agent cooperates with its neighboring agents to take one cooperative action. Finally, we conduct experiments to evaluate the performance of DeepMAG and experimental results show that it outperforms the state-of-the-art techniques.(c) 2022 Elsevier B.V. All rights reserved.

Background COVID-19–related health inequalities were reported in some studies, showing the failure in public health and communication. Studies investigating the contexts and causes of these inequalities pointed to the contribution of communication inequality or poor health literacy and information access to engagement with health care services. However, no study exclusively dealt with health inequalities induced by the use of social media during COVID-19. Objective This review aimed to identify and summarize COVID-19–related health inequalities induced by the use of social media and the associated contributing factors and to characterize the relationship between the use of social media and health disparities during the COVID-19 pandemic. Methods A systematic review was conducted on this topic in light of the protocol of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 statement. Keyword searches were performed to collect papers relevant to this topic in multiple databases: PubMed (which includes MEDLINE [Ovid] and other subdatabases), ProQuest (which includes APA PsycINFO, Biological Science Collection, and others), ACM Digital Library, and Web of Science, without any year restriction. Of the 670 retrieved publications, 10 were initially selected based on the predefined selection criteria. These 10 articles were then subjected to quality analysis before being analyzed in the final synthesis and discussion. Results Of the 10 articles, 1 was further removed for not meeting the quality assessment criteria. Finally, 9 articles were found to be eligible and selected for this review. We derived the characteristics of these studies in terms of publication years, journals, study locations, locations of study participants, study design, sample size, participant characteristics, and potential risk of bias, and the main results of these studies in terms of the types of social media, social media use–induced health inequalities, associated factors, and proposed resolutions. On the basis of the thematic synthesis of these extracted data, we derived 4 analytic themes, namely health information inaccessibility–induced health inequalities and proposed resolutions, misinformation-induced health inequalities and proposed resolutions, disproportionate attention to COVID-19 information and proposed resolutions, and higher odds of social media–induced psychological distress and proposed resolutions. Conclusions This paper was the first systematic review on this topic. Our findings highlighted the great value of studying the COVID-19–related health knowledge gap, the digital technology–induced unequal distribution of health information, and the resulting health inequalities, thereby providing empirical evidence for understanding the relationship between social media use and health inequalities in the context of COVID-19 and suggesting practical solutions to such disparities. Researchers, social media, health practitioners, and policy makers can draw on these findings to promote health equality while minimizing social media use–induced health inequalities.

In this paper, we study how to denoise medical ultrasound images and improve the performance of instance segmentation using deep learning technology. Since medical ultrasound images usually contain a lot of noises, we first propose a novel unsupervised learning approach called Dual Image (DI) for denoising of medical ultrasound images. DI consists of three main features. Firstly, unlike many existing supervised denoising methods, it does not need clean medical ultrasound images for denoising. Instead, it uses Computed Tomography (CT) images and the noise patches extracted from medical ultrasound images for denoising. Secondly, to effectively select noise patches from medical ultrasound images, a patch selection algorithm based on entropy is formulated. Thirdly, to minimize structure variation of denoised medical ultrasound images, a new reconstruction block is designed for combining the structural information from the structural enhancement block. After denoising, since medical ultrasound images are usually poor in features, to further improve the instance segmentation performance, we extend SOLOv2 to Segmenting on Ultrasound Image (SOUI) by proposing the Double Feature Pyramid Network (D-FPN) and mask fusion branch to strengthen the communication and fusion of different feature layers. Extensive experiments have been performed to study the performance of DI and SOUI using practical medical ultrasound images. We demonstrate that DI can greatly improve the quality of medical ultrasound images and minimize structure variation of denoised medical ultrasound images. SOUI gives 53.7% AP(Average Precision) on practical medical ultrasound images, and outperforms most the state-of-the-art instance segmentation methods including SOLOv2. Code is available at: https://github.com/ztt0821/SOUI.

Although log-structured merge trees (LSM-trees) are commonly adopted in many NoSQLs as they can significantly improve the write performance in updating a database, most of the proposed LSM-trees are concentrated on storing a single version of data. On the other hand, in many Internet of Things (IoT) applications, it is important to maintain the old versions of data in addition to the latest version. In this article, we introduce our design and implementation of an enhancement of LevelDB to multiversion LevelDB (called MVLevelDB) with the purpose to efficiently support temporal queries on multiversion data in IoT applications. Based on the temporal consistency, we formulated the log-structured multiversion tree (LSMV-tree) to be implemented into MVLevelDB. In LSMV-tree, each data version is associated with two time-stamps to define its validity interval, and both the data versions and the components are time-sorted to improve the efficiency in searching data in processing temporal queries. To handle the problem of multicomponents data versions, we designed the data version duplication (DvD) method in which a data version will be duplicated in the next component if it is valid while its component is being flushed from the main memory to disk storage. Extensive experiments using a benchmark program have been performed to investigate the performance of MVLevelDB as compared with LevelDB both in writing and reading data.

Background Mental disorders (MDs) impose heavy burdens on health care (HC) systems and affect a growing number of people worldwide. The use of mobile health (mHealth) apps empowered by artificial intelligence (AI) is increasingly being resorted to as a possible solution. Objective This study adopted a topic modeling (TM) approach to investigate the public trust in AI apps in mental health care (MHC) by identifying the dominant topics and themes in user reviews of the 8 most relevant mental health (MH) apps with the largest numbers of reviewers. Methods We searched Google Play for the top MH apps with the largest numbers of reviewers, from which we selected the most relevant apps. Subsequently, we extracted data from user reviews posted from January 1, 2020, to April 2, 2022. After cleaning the extracted data using the Python text processing tool spaCy, we ascertained the optimal number of topics, drawing on the coherence scores and used latent Dirichlet allocation (LDA) TM to generate the most salient topics and related terms. We then classified the ascertained topics into different theme categories by plotting them onto a 2D plane via multidimensional scaling using the pyLDAvis visualization tool. Finally, we analyzed these topics and themes qualitatively to better understand the status of public trust in AI apps in MHC. Results From the top 20 MH apps with the largest numbers of reviewers retrieved, we chose the 8 (40%) most relevant apps: (1) Wysa: Anxiety Therapy Chatbot; (2) Youper Therapy; (3) MindDoc: Your Companion; (4) TalkLife for Anxiety, Depression & Stress; (5) 7 Cups: Online Therapy for Mental Health & Anxiety; (6) BetterHelp-Therapy; (7) Sanvello; and (8) InnerHour. These apps provided 14.2% (n=559), 11.0% (n=431), 13.7% (n=538), 8.8% (n=356), 14.1% (n=554), 11.9% (n=468), 9.2% (n=362), and 16.9% (n=663) of the collected 3931 reviews, respectively. The 4 dominant topics were topic 4 (cheering people up; n=1069, 27%), topic 3 (calming people down; n=1029, 26%), topic 2 (helping figure out the inner world; n=963, 25%), and topic 1 (being an alternative or complement to a therapist; n=870, 22%). Based on topic coherence and intertopic distance, topics 3 and 4 were combined into theme 3 (dispelling negative emotions), while topics 2 and 1 remained 2 separate themes: theme 2 (helping figure out the inner world) and theme 1 (being an alternative or complement to a therapist), respectively. These themes and topics, though involving some dissenting voices, reflected an overall high status of trust in AI apps. Conclusions This is the first study to investigate the public trust in AI apps in MHC from the perspective of user reviews using the TM technique. The automatic text analysis and complementary manual interpretation of the collected data allowed us to discover the dominant topics hidden in a data set and categorize these topics into different themes to reveal an overall high degree of public trust. The dissenting voices from users, though only a few, can serve as indicators for health providers and app developers to jointly improve these apps, which will ultimately facilitate the treatment of prevalent MDs and alleviate the overburdened HC systems worldwide.

Background Medication nonadherence represents a major burden on national health systems. According to the World Health Organization, increasing medication adherence may have a greater impact on public health than any improvement in specific medical treatments. More research is needed to better predict populations at risk of medication nonadherence. Objective To develop clinically informative, easy-to-interpret machine learning classifiers to predict people with psychiatric disorders at risk of medication nonadherence based on the syntactic and structural features of written posts on health forums. Methods All data were collected from posts between 2016 and 2021 on mental health forum, administered by Together 4 Change, a long-running not-for-profit organisation based in Oxford, UK. The original social media data were annotated using the Tool for the Automatic Analysis of Syntactic Sophistication and Complexity (TAASSC) system. Through applying multiple feature optimisation techniques, we developed a best-performing model using relevance vector machine (RVM) for the probabilistic prediction of medication nonadherence among online mental health forum discussants. Results The best-performing RVM model reached a mean AUC of 0.762, accuracy of 0.763, sensitivity of 0.779, and specificity of 0.742 on the testing dataset. It outperformed competing classifiers with more complex feature sets with statistically significant improvement in sensitivity and specificity, after adjusting the alpha levels with Benjamini–Hochberg correction procedure. Discussion. We used the forest plot of multiple logistic regression to explore the association between written post features in the best-performing RVM model and the binary outcome of medication adherence among online post contributors with psychiatric disorders. We found that increased quantities of 3 syntactic complexity features were negatively associated with psychiatric medication adherence: “dobj_stdev” (standard deviation of dependents per direct object of nonpronouns) (OR, 1.486, 95% CI, 1.202–1.838, P < 0.001), “cl_av_deps” (dependents per clause) (OR, 1.597, 95% CI, 1.202–2.122, P, 0.001), and “VP_T” (verb phrases per T-unit) (OR, 2.23, 95% CI, 1.211–4.104, P, 0.010). Finally, we illustrated the clinical use of the classifier with Bayes' monograph which gives the posterior odds and their 95% CI of positive (nonadherence) versus negative (adherence) cases as predicted by the best-performing classifier. The odds ratio of the posterior probability of positive cases was 3.9, which means that around 10 in every 13 psychiatric patients with a positive result as predicted by our model were following their medication regime. The odds ratio of the posterior probability of true negative cases was 0.4, meaning that around 10 in every 14 psychiatric patients with a negative test result after screening by our classifier were not adhering to their medications. Conclusion Psychiatric medication nonadherence is a large and increasing burden on national health systems. Using Bayesian machine learning techniques and publicly accessible online health forum data, our study illustrates the viability of developing cost-effective, informative decision aids to support the monitoring and prediction of patients at risk of medication nonadherence.

Background: Due to its convenience, wide availability, low usage cost, neural machine translation (NMT) has increasing applications in diverse clinical settings and web-based self-diagnosis of diseases. Given the developing nature of NMT tools, this can pose safety risks to multicultural communities with limited bilingual skills, low education, and low health literacy. Research is needed to scrutinise the reliability, credibility, usability of automatically translated patient health information. Objective: We aimed to develop high-performing Bayesian machine learning classifiers to assist clinical professionals and healthcare workers in assessing the quality and usability of NMT on depressive disorders. The tool did not require any prior knowledge from frontline health and medical professionals of the target language used by patients. Methods: We used Relevance Vector Machine (RVM) to increase generalisability and clinical interpretability of classifiers. It is a typical sparse Bayesian classifier less prone to overfitting with small training datasets. We optimised RVM by leveraging automatic recursive feature elimination and expert feature refinement from the perspective of health linguistics. We evaluated the diagnostic utility of the Bayesian classifier under different probability cut-offs in terms of sensitivity, specificity, positive and negative likelihood ratios against clinical thresholds for diagnostic tests. Finally, we illustrated interpretation of RVM tool in clinic using Bayes' nomogram. Results: After automatic and expert-based feature optimisation, the best-performing RVM classifier (RVM_DUFS12) gained the highest AUC (0.8872) among 52 competing models with distinct optimised, normalised features sets. It also had statistically higher sensitivity and specificity compared to other models. We evaluated the diagnostic utility of the best-performing model using Bayes' nomogram: it had a positive likelihood ratio (LR+) of 4.62 (95% C.I.: 2.53, 8.43), and the associated posterior probability (odds) was 83% (5.0) (95% C.I.: 73%, 90%), meaning that approximately 10 in 12 English texts with positive test are likely to contain information that would cause clinically significant conceptual errors if translated by Google; it had a negative likelihood ratio (LR-) of 0.18 (95% C.I.: 0.10,0.35) and associated posterior probability (odds) was 16% (0.2) (95% C.I: 10%, 27%), meaning that about 10 in 12 English texts with negative test can be safely translated using Google.

Background From Ebola, Zika, to the latest COVID-19 pandemic, outbreaks of highly infectious diseases continue to reveal severe consequences of social and health inequalities. People from low socioeconomic and educational backgrounds as well as low health literacy tend to be affected by the uncertainty, complexity, volatility, and progressiveness of public health crises and emergencies. A key lesson that governments have taken from the ongoing coronavirus pandemic is the importance of developing and disseminating highly accessible, actionable, inclusive, coherent public health advice, which represent a critical tool to help people with diverse cultural, educational backgrounds and varying abilities to effectively implement health policies at the grassroots level. Objective We aimed to translate the best practices of accessible, inclusive public health advice (purposefully designed for people with low socioeconomic and educational background, health literacy levels, limited English proficiency, and cognitive/functional impairments) on COVID-19 from health authorities in English-speaking multicultural countries (USA, Australia, and UK) to adaptive tools for the evaluation of the accessibility of public health advice in other languages. Methods We developed an optimised Bayesian classifier to produce probabilistic prediction of the accessibility of official health advice among vulnerable people including migrants and foreigners living in China. We developed an adaptive statistical formula for the rapid evaluation of the accessibility of health advice among vulnerable people in China. Results Our study provides needed research tools to fill in a persistent gap in Chinese public health research on accessible, inclusive communication of infectious diseases' prevention and management. For the probabilistic prediction, using the optimised Bayesian machine learning classifier (GNB), the largest positive likelihood ratio (LR+) 16.685 (95% confidence interval: 4.35, 64.04) was identified when the probability threshold was set at 0.2 (sensitivity: 0.98; specificity: 0.94). Conclusion Effective communication of health risks through accessible, inclusive, actionable public advice represents a powerful tool to reduce health inequalities amidst health crises and emergencies. Our study translated the best-practice public health advice developed during the pandemic into intuitive machine learning classifiers for health authorities to develop evidence-based guidelines of accessible health advice. In addition, we developed adaptive statistical tools for frontline health professionals to assess accessibility of public health advice for people from non-English speaking backgrounds.

Linguistic intelligence and the ability to converse with human are important and indispensable parts of humanoid robots. One of the most challenging tasks in knowledge-grounded task-oriented dialog systems (KTDS) is the knowledge selection task, which aims to find the proper knowledge snippets to respond to user dialog requests. In this paper, we first propose domain adapted-BERT (DA-BERT) which employs pre-trained bidirectional encoder representations from transformers (BERT) with domain adaptive training and dynamic masking probability for knowledge selection in KTDS. Domain adaptive training can minimize the domain gap between the general text data that BERT is pre-trained on and the dialog–knowledge joint data while dynamic masking probability enhances the training in an easy-to-hard manner. After knowledge selection, the next task in KTDS is knowledge-grounded generation. To improve the performance in knowledge-grounded generation, we propose GPT-PR to employ post-ranking on the generator’s outputs. Post-ranking eliminates the possibility of generating hallucination response by a large portion during the sampling-based decoding process and thus can improve the quality of the generated response. Experimental results on the benchmark dataset show that our proposed pre-training and post-ranking methods, DA-BERT and GPT-PR, respectively, outperform the state-of-the-art models with large margins across all the evaluation metrics. Moreover, in the experiments, we also analyze the bad cases of DA-BERT and GPT-PR and do visualizations to facilitate further research in this direction.

Neural machine translation technologies are having increasing applications in clinical and healthcare settings. In multicultural countries, automatic translation tools provide critical support to medical and health professionals in their interaction and exchange of health messages with migrant patients with limited or non-English proficiency. While research has mainly explored the usability and limitations of state-of-the-art machine translation tools in the detection and diagnosis of physical diseases and conditions, there is a persistent lack of evidence-based studies on the applicability of machine translation tools in the delivery of mental healthcare services for vulnerable populations. Our study developed Bayesian machine learning algorithms using relevance vector machine to support frontline health workers and medical professionals to make better informed decisions between risks and convenience of using online translation tools when delivering mental healthcare services to Spanish-speaking minority populations living in English-speaking countries. Major strengths of the machine learning classifier that we developed include scalability, interpretability, and adaptability of the classifier for diverse mental healthcare settings. In this paper, we report on the process of the Bayesian machine learning classifier development through automatic feature optimisation and the interpretation of the classifier-enabled assessment of the suitability of original English mental health information for automatic online translation. We elaborate on the interpretation of the assessment results in clinical settings using statistical tools such as positive likelihood ratios and negative likelihood ratios.

One of the most challenging tasks in Knowledge-grounded Task-oriented Dialog Systems (KTDS) is the knowledge selection task, which aims to find the proper knowledge snippets to handle user requests. This paper proposes DA-BERT to employ pre-trained BERT with domain adaptive training and newly proposed dynamic masking probability to deal with knowledge selection in KTDS. Domain adaptive training minimizes the domain gap between the general text data BERT is pre-trained on and the dialog-knowledge joint data; and dynamic masking probability enhances the training in an easy-to-hard mode. Experimental results on the benchmark dataset show that our proposed training method outperforms the state-of-the-art models with large margins across all the evaluation metrics. Moreover, we analyze the bad case of our method and recognize several typical errors in the bad case set to facilitate further research in this direction.

The sequential pattern in the human movement is one of the most important aspects for location recommendations in geosocial networks. Existing location recommenders have to access users' raw check-in data to mine their sequential patterns that raises serious location privacy breaches. In this paper, we propose a new Privacy-preserving LOcation REcommendation framework (PLORE) to address this privacy challenge. First, we employ the nnth-order additive Markov chain to exploit users' sequential patterns for location recommendations. Further, we contrive the probabilistic differential privacy mechanism to reach a good trade-off between high recommendation accuracy and strict location privacy protection. Finally, we conduct extensive experiments to evaluate the performance of PLORE using three large-scale real-world data sets. Extensive experimental results show that PLORE provides efficient and highly accurate location recommendations, and guarantees strict privacy protection for user check-in data in geosocial networks.

Objective: To determine the linguistic/textual features of English health educational materials for predicting the probabilistic distribution of critical conceptual mistakes in neural machine translations (Google Translate: English to Chinese) of public-oriented online health resources on infectious diseases and viruses. Methods: We collected 200 English source texts on infectious diseases and their human translations to Chinese from HON. Net certified health education websites. Human translations were compared with machine translations (Google Translate) by native Chinese speakers to identify critical conceptual mistakes. To overcome overfitting issues of machine learning with small, high-dimensional datasets, Bayesian machine learning classifiers (relevance vector machine, RVM) was trained (70% and 30% train/test data split; 5-fold cross-validation) on English source texts classified as linked or not with machine translation outputs containing critical conceptual mistakes, to identify possible source text features causing clinically significant machine translation errors. We compared the performance of RVM with the combined features through separate optimization (CFSO: 21), to RVM trained on the original combined features (OCF: 135) (20 structural; 115 semantic features), combined features through joint optimization (CFJO: 48); optimized structural features (OTF: 5), and optimized semantic features (OSF: 16). In addition, RVM (CFSO) was compared to classifiers using individual standard (currently available) parameters to measure English complexity (Flesch Reading Ease FRE; Gunning Fog Index - GFI; SMOG Readability Index-SMOG). Results: The AUC, sensitivity, specificity and accuracy of RVM MLCs trained on different features sets were: CFSO (AUC: 0.685; sensitivity: 0.73, specificity: 0.63; accuracy: 0.68); OCF (AUC: 0.7; sensitivity: 0.42, specificity: 0.8; accuracy: 0.625); CFJO (AUC: 0.690; sensitivity: 0.54, specificity: 0.73; accuracy: 0.64); OTF (AUC: 0.587; sensitivity: 0.58, specificity: 0.53; accuracy: 0.55); OSF (AUC: 0.679; sensitivity: 0.58, specificity: 0.67; accuracy: 0.625). The best-performing model was RVM trained on the combined features through separate optimisation (CFSO) (16% of the original combined features). RVM (CFSO) outperformed binary classifiers (BCs) using standard English readability tests. The accuracy, sensitivity, specificity of the three BCs were FRE (accuracy 0.457; sensitivity 0.903, specificity 0.011); GFI (accuracy 0.5735; sensitivity 0.685, specificity 0.462); SMOG (accuracy 0.568; sensitivity 0.674, specificity 0.462). Conclusion: Our study found that machine-generated Chinese medical translation errors were not caused by difficult medical jargon or a lack of readability of source language information. It was certain English structures (passive voices; sentences starting with conjunctions), semantic polysemy (different meanings of a word when used in common versus specialized domains) which tend to cause critical conceptual mistakes in neural machine translation systems (English to Chinese) of health education information on infectious diseases.