Loading [Contrib]/a11y/accessibility-menu.js
Research article Special Issues

Multi-behavioral recommendation model based on dual neural networks and contrast learning

  • In order to capture the complex dependencies between users and items in a recommender system and to alleviate the smoothing problem caused by the aggregation of multi-layer neighborhood information, a multi-behavior recommendation model (DNCLR) based on dual neural networks and contrast learning is proposed. In this paper, the complex dependencies between behaviors are divided into feature correlation and temporal correlation. First, we set up a personalized behavior vector for users and use a graph-convolution network to learn the features of users and items under different behaviors, and we then combine the features of self-attention mechanism to learn the correlation between behaviors. The multi-behavior interaction sequence of the user is input into the recurrent neural network, and the temporal correlation between the behaviors is captured by combining the attention mechanism. The contrast learning is introduced based on the double neural network. In the graph convolution network layer, the distances between users and similar users and between users and their preference items are shortened, and the distance between users and their short-term preference is shortened in the circular neural network layer. Finally, the personalized behavior vector is integrated into the prediction layer to obtain more accurate user, behavior and item characteristics. Compared with the sub-optimal model, the HR@10 on Yelp, ML20M and Tmall real datasets are improved by 2.5%, 0.3% and 4%, respectively. The experimental results show that the proposed model can effectively improve the recommendation accuracy compared with the existing methods.

    Citation: Suqi Zhang, Wenfeng Wang, Ningning Li, Ningjing Zhang. Multi-behavioral recommendation model based on dual neural networks and contrast learning[J]. Mathematical Biosciences and Engineering, 2023, 20(11): 19209-19231. doi: 10.3934/mbe.2023849

    Related Papers:

    [1] Kenneth Li-minn Ang, Kah Phooi Seng . Biometrics-based Internet of Things and Big data design framework. Mathematical Biosciences and Engineering, 2021, 18(4): 4461-4476. doi: 10.3934/mbe.2021226
    [2] Jiang Zhao, Dan Wu . The risk assessment on the security of industrial internet infrastructure under intelligent convergence with the case of G.E.'s intellectual transformation. Mathematical Biosciences and Engineering, 2022, 19(3): 2896-2912. doi: 10.3934/mbe.2022133
    [3] Kittur Philemon Kibiwott , Yanan Zhao , Julius Kogo, Fengli Zhang . Verifiable fully outsourced attribute-based signcryption system for IoT eHealth big data in cloud computing. Mathematical Biosciences and Engineering, 2019, 16(5): 3561-3594. doi: 10.3934/mbe.2019178
    [4] Noman Zahid, Ali Hassan Sodhro, Usman Rauf Kamboh, Ahmed Alkhayyat, Lei Wang . AI-driven adaptive reliable and sustainable approach for internet of things enabled healthcare system. Mathematical Biosciences and Engineering, 2022, 19(4): 3953-3971. doi: 10.3934/mbe.2022182
    [5] Ridha Ouni, Kashif Saleem . Secure smart home architecture for ambient-assisted living using a multimedia Internet of Things based system in smart cities. Mathematical Biosciences and Engineering, 2024, 21(3): 3473-3497. doi: 10.3934/mbe.2024153
    [6] Hyeji Roh, Seulgi Shin, Jinseo Han, Sangsoon Lim . A deep learning-based medication behavior monitoring system. Mathematical Biosciences and Engineering, 2021, 18(2): 1513-1528. doi: 10.3934/mbe.2021078
    [7] Radi P. Romansky, Irina S. Noninska . Challenges of the digital age for privacy and personal data protection. Mathematical Biosciences and Engineering, 2020, 17(5): 5288-5303. doi: 10.3934/mbe.2020286
    [8] Dengzhi Liu, Zhimin Li, Chen Wang, Yongjun Ren . Enabling secure mutual authentication and storage checking in cloud-assisted IoT. Mathematical Biosciences and Engineering, 2022, 19(11): 11034-11046. doi: 10.3934/mbe.2022514
    [9] Xiangshuai Duan, Naiping Song, Fu Mo . An edge intelligence-enhanced quantitative assessment model for implicit working gain under mobile internet of things. Mathematical Biosciences and Engineering, 2023, 20(4): 7548-7564. doi: 10.3934/mbe.2023326
    [10] Yongjun Ren, Yan Leng, Yaping Cheng, Jin Wang . Secure data storage based on blockchain and coding in edge computing. Mathematical Biosciences and Engineering, 2019, 16(4): 1874-1892. doi: 10.3934/mbe.2019091
  • In order to capture the complex dependencies between users and items in a recommender system and to alleviate the smoothing problem caused by the aggregation of multi-layer neighborhood information, a multi-behavior recommendation model (DNCLR) based on dual neural networks and contrast learning is proposed. In this paper, the complex dependencies between behaviors are divided into feature correlation and temporal correlation. First, we set up a personalized behavior vector for users and use a graph-convolution network to learn the features of users and items under different behaviors, and we then combine the features of self-attention mechanism to learn the correlation between behaviors. The multi-behavior interaction sequence of the user is input into the recurrent neural network, and the temporal correlation between the behaviors is captured by combining the attention mechanism. The contrast learning is introduced based on the double neural network. In the graph convolution network layer, the distances between users and similar users and between users and their preference items are shortened, and the distance between users and their short-term preference is shortened in the circular neural network layer. Finally, the personalized behavior vector is integrated into the prediction layer to obtain more accurate user, behavior and item characteristics. Compared with the sub-optimal model, the HR@10 on Yelp, ML20M and Tmall real datasets are improved by 2.5%, 0.3% and 4%, respectively. The experimental results show that the proposed model can effectively improve the recommendation accuracy compared with the existing methods.



    With the dramatic progress of sensor, radio, frequency identification and positioning technologies, wearable devices have become a popular research topic in recent years [1]. Information and communication technologies are also applied to these devices in the medical industry to facilitate interactions between patients, hospitals and medical personnel, which enables the intelligence and automation of modern healthcare systems. The Internet of Medical Things (IoMT) is developing steadily, with wearables having various sensors and computational capacities built in to continuously watch and monitor the patient's physiological data (blood pressure, heartbeat, breathing, etc.). The IoMT also transfers the collected clinical data to the remote cloud centers for analysis through technologies like Bluetooth, WiFi and ZigBee [2,3]. These data are most often applied for clinical care and disease diagnosis with the help of physicians. Useful information mined from the clinical database is used for patient condition analysis and treatment planning. This has led to a decrease in healthcare expenditures. Particularly amid the COVID-19 pandemic, telemedicine options were widely adopted due to distance constraints. Medical practitioners were required to utilize IoMT technologies to remotely monitor or perform procedures on patients [4]. The IoMT technology has now become an important tool for the health and well-being of all people.

    The IoMT has been used in various aspects of healthcare, including identification, remote monitoring and device monitoring. Data on patients' vital signs can be collected by using wireless sensor networks, which are made up of different wireless medical sensor types such as pressure sensors, biosensors and implanted sensors. In contemporary surgical suites, emergency departments and intensive care units, sensors are frequently used to monitor and show patient status. In addition, sensor-centered wearable devices address temporal and spatial challenges, offering real-time health monitoring for individuals, and reducing the treatment cost of patients. The worldwide market of the IoMT will reach USD 135.87 billion by 2025 from USD 24 billion in 2016 with an annual rate of 16.9 percent [5], as shown in Figure 1.

    Figure 1.  The worldwide market size of the IoMT [5].

    This paper uniquely presents a comprehensive overview of the IoMT background and delves into the importance of big data, cloud computing and artificial intelligence (AI) as major developing technologies for human health monitoring in daily life. These technologies are seen as crucial instruments for overcoming several issues, such as cost-effectiveness, security, privacy, accuracy and power usage. Moreover, we shed light on ethical challenges and prospects. Unlike previous IoMT surveys that typically focus on specific aspects, our work delivers a holistic view of the architecture, technologies, applications, and challenges, aiming to be a vital guide for new IoMT researchers and practitioners. We have searched the related studies in the Web of Science by using the keywords of Healthcare IoT, Internet of Medical Things, Healthcare 4.0 and Wearable devices for IoT; the growth trend is shown in Figure 2.

    Figure 2.  The growth trend of the IoMT.

    The paper is organized into several sections. Section 2 provides an architecture for IoMTs with three layers. Section 3 presents an introduction to cloud computing, big data, and AI technologies in the healthcare system. Various applications in different fields based on IoMTs are reviewed in Section 4. Section 5 presents an overview of future trends. In Section 6, difficulties with developing an IoMT for health monitoring and other potentially expansive applications are highlighted. We put an end to our effort in Section 7. Table 1 lists the acronyms that appear in the paper and their corresponding full names.

    Table 1.  Summary of acronyms.
    Acronyms Full forms
    IoMT Internet of Medical Things
    AI Artificial intelligence
    CPS Cyber physical system
    ML Machine learning
    IBM International Business Machines Corporation

     | Show Table
    DownLoad: CSV

    Figure 3 describes the general architecture of an IoMT, which encompasses the cloud layer, intermediate layer and device layer [6]. The patient's physiological data are given to a cloud database or caregiver so that they can monitor the patient's health problems remotely from anywhere and at any time.

    Figure 3.  Architecture of the IoMT.

    In this layer, there are various medical devices, implantable devices, portable devices, nurses, doctors, medicines, patients, ambulances, etc. They use radio frequency identification, the Electronic Product Code and different communication protocols during communication with the gateway and are mainly responsible for transmitting key medical information such as physiological signals.

    The sensing layer uses sensors to sense and collect environmental information, and it consists of two main sublayers, i.e., the data acquisition sublayer and the data access sublayer. The task of collecting data is performed by the data acquisition sublayer by using various types portable medical equipment and signal acquisition devices. A diverse range of portable medical equipment and signal acquisition devices are widely used to complete the task of data collecting. This process involves a variety of signal acquisition technologies, including radio frequency identification technology and universal packet radio service technology. In addition, the process involves a variety of sensors, such as DNA sensors, physiological signal sensors and chemical sensors. Finally, the collected data are delivered to the cyber physical system (CPS) nodes through the network. It is noteworthy that within the realm of the IoT, the nodes encompass three distinct types: passive CPS, active CPS and Internet CPS. Different sorts of nodes are automatically chosen during this procedure based on various items and requirements. The data access sublayer employs ZigBee, Wi-Fi, Bluetooth and other short-distance data transmission technologies to facilitate the integration of data collected from the data collection sublayer into the intermediate layer. The selection of the primary access method should be based on the veritable environmental attributes of the IoMT and the requirements of the various entities involved. However, in the device layer, challenges abound that are related to connectivity, notably in regions with limited or inconsistent network coverage. Bandwidth restrictions can lead to data transmission bottlenecks. Further, the limited processing capability can hinder real-time analytics, and battery life remains a constant concern, impacting device uptime and reliability. Emerging sensor advancements are currently expanding IoMT capabilities. Novel flexible and stretchable sensors allow better integration into wearables and implants [7]. Sensors leveraging nanotechnology and microfluidics offer higher sensitivity and accuracy [8,9]. Advances in low-power sensor electronics and energy harvesting help to address battery life constraints.

    The second layer is also known as a gateway or intermediate layer; it should handle most of the processing and be implemented by using fog computing technologies. To enable communication with physical objects in the medical field, various heterogeneous communication protocols are used. In addition, the implementation of a lightweight distributed access control system with keyword search is also a task of this layer. This system integrates data access control with keyword search functionality. It ensures the privacy and security of data while allowing users to quickly search for specific information by using keywords. In the IoMT domain, where data privacy is paramount and healthcare professionals require rapid access to specific medical information, this feature is highly valuable.

    The intermediate layer comprises two main sublayers: the network transport layer and the service layer. In the network transport layer, data information acquired from the sensing layer can be transmitted accurately and reliably in real time through mobile communication networks or private networks, making it the backbone of the IoMT. IoMT technology was created to develop integration adaptation technology, which is a technology applicable to hospital networks and pre-existing networks. The service layer's function is integration, which encompasses both the fusion of disparate networks and the fusion of data. To offer medical services to doctors and patients, the platform is built and interfaces are provided to the cloud layer based on this service to develop relevant third-party applications. There are tangible resource constraints concerning computation, storage and power. These constraints significantly limit the feasibility of implementing intricate security solutions, which are paramount for safeguarding critical data and ensuring seamless operations.

    The gathered information will be processed in an intermediary layer before being sent to the main servers. To manage the medical procedure effectively, large volumes of data will need to be saved and handled on these servers, employing various programs. Big data analytics and high-performance computing constitute an important part of enabling IoMT applications. At the same time, this layer accomplishes tasks related to security protocols. Therefore, cloud computing is an important component of the IoMT for the healthcare industry.

    The cloud layer consists of two parts: medical information applications and medical information decision applications. The main task of medical information applications is management, including patient data management, medical equipment management, outpatient and inpatient information management, etc. The main task of medical information decision applications is analysis, including disease cause analysis, patient data analysis, treatment plan analysis, medication rule analysis, etc. The cloud layer is predominantly challenged by interoperability inconsistencies, resulting in data silos that hinder seamless integration. Moreover, the ability to scale computational resources becomes increasingly complex, especially when managing and processing massive datasets derived from various heterogeneous sources.

    Nowadays, cloud computing, big data and AI technologies can be considered as a unified whole, and they play an irreplaceable and important role in the whole high-tech system [10]. Therefore, in the wake of developments in science and technology, the integration of these technologies will become increasingly important. Cloud computing can quickly collect sensor data scattered all over the world [11]. Subsequently, all of the collected data are transmitted to edge nodes through high-speed networks and then to the computing centers, greatly enhancing the timeliness of data processing and transmission efficiency. Coupled with big data processing technologies and advanced AI algorithms, data can be managed and found more efficiently, allowing users to quickly get the data information they needed and make informed decisions quickly [12]. In addition, cloud computing can easily integrate powerful AI and big data processing capabilities, which is equivalent to adding a "brain" to the IoMT system, making the devices in the IoMT system more "intelligent" [13].

    With the increasing complexity and volume of data generated in the healthcare field, the heterogeneity of healthcare data is increasing, which brings about huge challenges in the areas of accessing data, processing information and maintaining systems [14,15]. It is essential that information management technologies are applied to the data management and service processes of healthcare organizations [16,17,18]. Cloud computing is a key technology for solving such problems [19]. It is a kind of distributed computing, also called a network "cloud", where the massive data calculation and processing procedures are broken down into countless small procedures [20]. After that, these procedures are processed and analyzed by a system of multiple servers, and the results are returned to the users. Cloud computing efficiently manages large-scale data computation and processing by employing distributed computing frameworks, which parallelize tasks across multiple nodes. It enhances scalability through elasticity, automatically adjusting resources based on demand. Data partitioning enhances parallelism by breaking down datasets across distributed databases, while load balancers distribute computational tasks to prevent server overloads. Additionally, serverless models, in-memory processing and optimized cloud-based machine learning (ML) services streamline and expedite data processing, ensuring that vast datasets are handled efficiently and swiftly in the cloud environment. Depending on the scope of use, cloud computing includes three categories of clouds: private, public and hybrid [21]. Cloud computing has many benefits over conventional web application architectures, including high reliability, resource sharing and cost-effectiveness [22]. To achieve unified management and scheduling, a pool of resources based on virtualization technology is formed by various computing resources [23,24]. The resources in these clouds are available and shared indefinitely, even if the users are not aware of the internal operating environment of the cloud [23,25]. They are capable of selecting the appropriate resources according to their needs and have access to unlimited resources whenever and wherever. In the process of storing data, cloud computing technology has a certain classification function, which combines data redundancy and security technology for the classification and storage of related data information [26]. The integrity of data information and the security of client data are ensured by its ability to lower the risk of data loss and leakage. Cloud computing avoids the harsh requirement of excessive performance of a single server and greatly enhances the benefit rate of resources due to it being capable of accumulating and integrating various low-end computing resources. In addition, it makes it possible to achieve elastic scheduling [27]. In other words, when the resources rented by users are idle, the cloud platform is able to reuse this part of the idle resources, which fundamentally determines the use of cloud computing to support IoMT applications, which will have a high cost-performance ratio.

    Major cloud providers have significantly influenced the healthcare field. According to estimates from Synergy Research Group, in the second quarter of 2023, Amazon Web Services held a 32% market share in the worldwide cloud infrastructure market, followed by Microsoft Azure at 22% and Google Cloud Platform at 11% [28]. These giants have launched specific healthcare solutions, optimizing data processing and patient monitoring and emphasizing the practical integration of cloud computing in the sector. Their competitive edge is rooted in tailored services that cater to the unique challenges of healthcare. A better possibility for the switch from conventional hospitals to digital ones is presented by the advent and development of cloud computing technology, which can lay the groundwork for the archiving, analysis and dissemination of medical data.

    The burgeoning progress of hospital information technology has resulted in an escalating surge of data within healthcare institutions. Patients generate a substantial amount of health data in their daily lives through medical devices, encompassing metrics such as heart rate, blood pressure, oxygen saturation, blood gas analysis, etc [29,30]. As a result, a huge space is needed to store this valuable and large volume of data, which are used to analyze and then assign treatment plans [31]. Medical data have become the new force of big data, and the amount is growing. For example, the sports bracelet, the networked electronic blood pressure meter, the weight scale and telemetry cardiac monitoring in the hospital, which is able to generate electronic prescriptions, electronic medical records, test data, human medical images (ultrasound, digital radiography, computed tomography and magnetic resonance imaging) and reports and then aggregate them into big data.

    Big data technology involves the scientific organization and analysis of massive data to reveal the underlying patterns concealed within the data. It is supposed to include two modules [32]. In the first module, various structural types are classified or clustered and the massive data are divided into several types for easy processing, and then the amount of computation is reduced. In the second module, the data are analyzed by using predictive models and recommendation models to find out useful and valued information, explore the development pattern of things and help users to make decisions. The predictive models commonly used include time series models, decision trees, Bayesian classification models, etc. The recommendation algorithms commonly used mainly include content-based, collaborative filtering, association rule-based and utility-based recommendation algorithms.

    With the extensive utilization of computers and the Internet, the concept of stored data has witnessed a considerable surge compared to previous times. The significance of data information is demonstrated by employing big data technology to extract pertinent insights through scientific and technical methodologies. If comprehensive data are used to analyze patient information, it can better enhance the accuracy and precision of healthcare assessments and even predict disease progression and propose targeted countermeasures early. Hidden, previously unknown and potentially useful information and knowledge are extracted through highly automated analysis and inductive reasoning of various clinical data. Based on these hidden data, early warnings and diagnoses can be made for patients. Medical data can also be applied to other areas of medicine, where big data platforms can be used to predict diseases and analyze drug side effects [33]. Hospitals use medical data systems to upload all diagnostic records of patients to the cloud, which can be used as an important basis for studying the familial and regional distribution of the onset of certain diseases. It allows people to pay attention to environmental protection and thus change their surroundings to reduce the probability of developing diseases, while also encouraging more people to join together to protect the environment. Another role of medical data is to analyze drugs for side effects [34]. The analysis of the side effects of drugs is useful for improving patient outcomes and clinical guidance, as well as exerting a substantial influence on medical expenses and medication use.

    AI is an emerging technological science that is capable of incorporating human thoughts, simulating human thinking and extending intelligent behavior with theories, methods, technologies and application systems [35,36].

    It usually includes three categories, which are classified as super AI, strong AI and weak AI. The most popular AI technologies at the moment are ML, computer vision and natural language processing [37,38,39]. Computer vision is used in the medical field to read, recognize and diagnose various types of medical images, such as computed tomography images, X-ray images and ultrasound images [40,41]. Meanwhile, natural language processing can transform large, redundant and unstructured medical data into valuable structured data. In addition, in the medical field, manual diagnosis suffers from low efficiency and high error, and to solve this problem, ML techniques are applied to predict and diagnose diseases. Recent studies have demonstrated the integration of AI and ML algorithms in IoMT systems for a variety of healthcare applications [42]. In the ML process, an extensive corpus of medical data necessitates analysis to obtain sophisticated models to guide doctors' decisions [43,44]. This technology is able to simulate the cognitive process of the human brain when making decisions to provide intelligent medical diagnosis, reducing the errors of manual diagnosis and greatly improving the effectiveness of diagnosis and care for patients. In the diagnosis and detection of diseases, convolutional neural networks, recurrent neural networks, random forests and support vector machines have been employed to classify skin conditions, predict cardiovascular events and detect epileptic seizures by using data from wearables, remote monitoring devices and medical imaging [45,46,47,48,49]. The ML models can analyze complex physiological signs and imaging patterns to provide timely and accurate diagnosis. For treatment and drug recommendations, ensemble methods combining multiple ML models have shown success in learning from electronic health records and patient symptoms to predict responses to hepatitis C drugs and suggest personalized medication plans [50]. The AI agents can continuously optimize the medication by taking into account individual patient characteristics. Moreover, federated learning, where models are trained locally on patient devices without sharing raw data, has been introduced to IoMT systems to enhance data privacy and security during remote healthcare services [51,52]. The decentralized training approach ensures that sensitive personal information is kept confidential while the population-based model improves over time.

    In summary, AI is transforming IoMT-based healthcare from reactive to proactive. The applications range from anomaly detection and chronic disease management to drug recommendation and medical follow-ups [53,54]. With the increasing adoption of IoMT devices and the expanding access to health data, smart healthcare systems powered by AI will become indispensable tools to provide predictive, preventive and precise care.

    With the rapid development of IoMT technologies, they have been applied in various healthcare scenarios to improve diagnosis, monitoring and treatment. In this section, we provide an overview of promising applications of the IoMT across different domains. These examples demonstrate the wide range of capabilities and benefits that interconnected medical devices can offer patients, providers and healthcare systems when combined with enabling technologies. Table 2 lists some of the applications and their pros and cons. The specific descriptions of the applications are as follows.

    Table 2.  Summary of pros and cons of the applications.
    Applications Refs Pros Cons
    Smartwatch for COVID-19 Detection [55] Early, real-time, non-invasive, low-cost COVID-19 detection using smartwatch Small sample size, biased cohort, higher false alarms and requirement for follow-up testing
    [56] Low-cost, contactless, real-time WiFi-based monitoring Limited coverage and evaluation, privacy concerns, interference issues, need for medical integration
    [57] Real-time, integrated, predictive, remote monitoring Sensor accuracy dependency, IoT investment, data quality dependency, privacy concerns, limited clinical replacement, workflow changes
    Monitoring for Maternity [58] Integrated IoT system, emergency alert, advanced feature extraction, effective classification Sensor dependency, limited dataset, complex deep learning model
    [59] Remote real-time monitoring, early detection, AI analysis, automated alerts Privacy concerns, costs, AI reliability, limited testing, patient compliance
    [60] IoT maternal healthcare platform, wearable sensors, AI integration, remote monitoring, user acceptance insights Lack of technical details, limited evaluation, data accuracy challenges
    Monitoring for Sports Athletics [62] Real-time monitoring, wireless, fog computing, scalable, low energy, high accuracy Cost, user compliance, individual calibration, security concerns, limited validation
    Monitoring for Rehabilitation [67] Wearable IoT monitoring, real-time feedback, edge processing, network modeling Limited clinical evaluation, lack of security discussion, chest belt sensor, dependency on smartphones, unclear durability and costs
    [68] Osmotic computing integration, gamified rehabilitation, closed-loop model, flexibility and scalability New paradigm, narrow use case, qualitative evaluation, lack of security focus, limited comparisons
    [70] Augmented reality-based therapy, IoMT integration, personalized rehabilitation Insufficient data, lack of standards, privacy concerns, deployment challenges, unproven effectiveness

     | Show Table
    DownLoad: CSV

    Different physiological parameters are collected by various wearable and non-wearable devices for COVID-19 detection through the IoT. For example, Mishra et al. [55] developed a method for real-time health monitoring by using the physiological parameters detected by wearable devices; the associations between symptom type and severity, heart rate signals, and the impact of infection on activity and sleep were also examined. Li et al. [56] proposed a Wi-COVID framework by using at-home WiFi signals, which is a non-invasive and wearable-free method for screening the respiration rates of COVID-19 patients. A system based on the IoTs was applied to identify suspected coronavirus instances through the collection of real-time symptom data [57], which encompasses five modules, i.e., Symptom Data Acquisition and Uploading (utilizing wearable sensors), Quarantine/Isolation Center, Data Analysis Center, Health Physicians and Cloud Infrastructure. Ensuring the accuracy of COVID-19 symptom detection using consumer wearables necessitates rigorous clinical validation of these devices.

    The market for pregnancy-related wearable technology has expanded quickly. These devices include fetal monitors and multi-functional health screening instruments, which can assist in managing and monitoring maternal health indicators like blood pressure, blood sugar and fetal heart rate. Recent studies have proposed IoT and AI systems for smart maternal healthcare to reduce maternal mortality rates. Venkatasubramanian [58] developed an IoT platform by applying wearable sensors, fog computing and cloud analytics with deep learning models for the continuous monitoring, emergency alerts and predictive analytics of maternal and fetal health. Vemuri et al. [59] presented an IoT and AI system leveraging wearable sensors and ML to track maternal vitals, detect critical patterns and provide real-time alerts. Li et al. [60] proposed an intelligent IoMT platform for maternal medical services based on wearable devices and cloud computing, and the architecture is shown in Figure 4. Specifically, pregnant women wear smart devices to track vital signs like heart rate, blood pressure, etc. These readings are transmitted wirelessly to a gateway and then stored and analyzed in the cloud. Ultimately, through cloud computing, various applications can access this data, providing real-time health monitoring and advice for both doctors and expectant mothers. Pregnant women play a vital role in learning about their unborn children since they are the fetus's mother. In-home self-monitoring emerges as a vital component of prenatal health care. Bringing a change to the conventional paradigm of physiological signal gathering and testing, wearable devices enable the reliable management of maternal vital sign data in a real-time dynamic manner. Pregnant women can utilize wearable technology to assess things like fetal blood sugar, heart rate, blood pressure, blood oxygenation, pulse, lipids and the electrocardiogram. It is imperative to address the comfort, wearability and uninterrupted power supply challenges in wearable maternal sensors, while also considering the potential risks associated with sensor radiation exposure to the fetus. The intelligent maternity platform based on the IoMT and centered on pregnant women can greatly reduce the workload of medical and nursing staff, increase productivity, make it easier for pregnant women to get healthcare and raise the standard of obstetric care.

    Figure 4.  Three layers of maternity platform.

    Wireless technologies, body sensors and fitness trackers in the exercise space exert a substantial influence on life efficiency and the reliability of health systems [61]. Wearable devices are of increasing interest in evaluating physiological considerations, advancing health and enhancing exercise compliance among diverse groups of people ranging from patients to expert athletes. In [62], an IoT-based exercise health monitoring system was created by using a fog-assisted computing efficient wearable sensor network; also, wearable devices for monitoring the heart rate in a continuous and real-time manner, exercise rhythm and respiratory rate were analyzed. Sensor data are uploaded by the system to the Ethernet module of the IoMT system to determine the athlete's physical status, and then the user is given access to the data over the Internet. The basic flow is shown in Figure 5. It is important to note that fog computing services were used to categorize a patient's health status as protected or at risk particularly, this was achieved by using incoming health data from every part of the entire model processed via the queue first come first served approach, while maintaining permissions to use healthcare data saved in distributed fog and cloud environments. This reduces the amount of data transferred in the cloud, thus minimizing the cost of computing resources. Finally, experiments have demonstrated the high accuracy of the system in terms of predicting athlete abnormalities. The performance of the sensor has certain advantages when compared to other peer methods.

    Figure 5.  Fog assisted computational efficient wearable sensor network.

    Traditional smart wearable devices are mainly utilized to calculate the number of steps, the corresponding body data and the corresponding body indicators of athletes or sports enthusiasts. The energy consumption calculated by these metrics is inaccurate, which means that the error is relatively large. Based on this, a wearable device was designed by using relevant sensors and IoT technologies to accurately calculate and monitor the energy consumption in [63,64]. The pedometry algorithm, which calculates the correlation between human energy and movement consumption and provides the calculation structure to the smart device, is the corresponding core algorithm. In terms of data acquisition, sensor devices are used to provide real-time monitoring of the human movement state. Simple transmission methods are used to reduce algorithm complexity and algorithm consumption, as well as to reduce storage space usage. Hardware sensors are usually fixed to different body parts of the athlete. For example, to collect signals from arm movements fix the sensor on the arm, while for posture signals from torso movements fix the sensor on the waist. In comparison with traditional smart devices, we found that the system improved accuracy in terms of exercise energy expenditure. The cost issue can also be resolved further by the downsizing and intelligent use of wearable smart gadgets. Future trends include the refinement of biometric sensor technology, the development of advanced ML algorithms for personalized exercise recommendations and a heightened focus on data privacy and security with the common goal of optimizing personalized exercise and health management.

    The IoMT has shown great potential to transform rehabilitation practices through continuous remote monitoring, real-time feedback and intelligent data analytics. Recent studies have utilized IoMT devices such as wearable sensors, in combination with mobile applications and AI algorithms, to develop innovative solutions for rehabilitation [65,66]. The IoMT has demonstrated tremendous potential in the field of cardiac rehabilitation. IoMT systems can remotely monitor patients' vital signs during physical activities, such as the electrocardiogram, respiratory rate and activity levels, enabling the remote monitoring of cardiac rehabilitation. Relevant mathematical models have been developed to ensure the real-time transmission of crucial data and feedback information, even under unstable network conditions [67]. This enhances the accessibility and personalization of cardiac rehabilitation programs. In physical rehabilitation, IoMT devices like motion sensors, electromyography sensors and virtual reality games are integrated to track the progress of functional recovery and boost patients' engagement in treatment [68]. Interactive exercises and biofeedback are provided to motivate patients. ML techniques can also be applied to customize rehabilitation plans based on sensor data. For mental rehabilitation, IoMT wearable devices inspire the use of augmented reality and virtual reality in exposure therapy, making them applicable for treating phobias and post-traumatic stress disorder [69,70]. Multiple physiological signals are captured to infer emotional states and suggest suitable environments to induce positive physiological responses. This data-driven approach enhances the quantification and personalization of digital therapy. With increasing clinical evidence and interdisciplinary collaboration, the IoMT is poised to improve patient outcomes by transforming rehabilitation practices. In the future, the IoMT in the field of rehabilitation will show key trends such as personalized rehabilitation programs, remote monitoring, virtual reality technology, ML prediction and intelligent rehabilitation aids.

    While the IoMT is already being deployed in various healthcare settings, there remain abundant opportunities for future growth and innovation. Ongoing technological advancements will further expand the possibilities of smart, connected medical systems. Here we highlight several emerging trends that are shaping the next frontiers for the IoMT.

    With the exponential growth of IoT device connections over the past few years, the Internet has become more open and complex, and information security and privacy issues have become one of the key factors in the development of the Internet [71,72]. Blockchain technology has become an effective technology to protect the information security of IoT devices due to its outstanding features, including its data tampering prevention and decentralization [73]. Satoshi Nakamoto's white paper was publicized in 2008; after that, blockchain can solve the problems of scalability, reliability, and privacy of the IoT as a distributed system. The blockchain uses a decentralized consensus method to guarantee the accuracy of the data. Proof of work, proof of stake, practical byzantine fault tolerance and delegated proof of stake are the four main consensus mechanisms in existing blockchain systems [74,75]. In addition, some blockchain systems use consensus mechanisms such as proof of bandwidth, proof of elapsed time and proof of authority [76]. The five categories of blockchain propagation mechanisms serve as the foundation for building consensus and trust in blockchain, including advertisement-based propagation, sendheaders propagation, unsolicited push propagation, relay network propagation, and push/advertisement hybrid propagation [77,78].

    Given the huge potential of blockchain in the field of cybersecurity, blockchain technology has been applied to a wide range of health IoT [79]. By leveraging the nature of the blockchain distributed ledger, the data flow of blockchain can be brought into play under the premise of safeguarding data security and privacy, and the data barriers between hospitals, institutions and enterprises can be bridged, thus enabling the reuse of medical data. Blockchain technology can also be applied to applications such as drug traceability and medical consultation [80,81]. With the ability of blockchain to store and analyze data, a series of medical information in the process of medical treatment, such as medical aid acts and plans, as well as the physical condition and behavior of patients in the process of treatment, can be recorded on the chain so that, once there is a need to check and verify medical disputes, the responsibility and behavior of each role at each stage can be clarified through the traceability mechanism to protect the rights and interests of all parties, thus achieving the purpose of alleviating the problem of medical disputes. Medical technology companies such as Philips, Change Healthcare, PokitDok, Patientory, Hashed Healthcare, and Doc.ai are actively exploring the medical application of blockchain technology. Philips has indicated that healthcare systems may benefit from blockchain technology and they are currently exploring the use of blockchain to build trust and accountability in the medical research ecosystem. Change Healthcare in the USA has launched the first enterprise-level medical blockchain solution, aiming to create transparent claims scenarios. PokitDok announced a partnership with Intel to deliver a healthcare blockchain solution called DokChain to meet patients' desire to access and control their medical data, thus enabling improved connectivity in the healthcare system. Hashed Healthcare has partnered with USA state governments on blockchain healthcare pilots. Patientory has also launched healthcare smart contracts and token payments. Information technology companies such as Google, International Business Machines Corporation (IBM) and Intel are also actively researching the applications of blockchain technology in healthcare. In early 2017, the USA Food and Drug Administration partnered with IBM Waston Health to investigate the use of blockchain technology to share health data in order to improve public health. DeepMindHealth, Google's AI health-tech subsidiary, also announced the use of blockchain to allow hospitals and patients to track personal health data in real time. Blockchain technology companies Gem, Factom, Bit Health, BlockVerify, DNA.Bits, Chronicled, SimplyVital Health and Bitfury are also actively exploring the medical applications of blockchain technology. In addition, South Korea's MediBloc has built a blockchain-based personal medical information platform to provide a new way of managing medical information for doctors and patients. Robomed Network, a decentralized medical network, has also created smart contracts to reduce medical expenses. Currently, blockchain technology is also applied extensively in finance, public services, digital copyright, insurance, public welfare, justice and other fields [82]. Table 3 lists the applications of blockchain technology in the health IoT.

    Table 3.  Applications of blockchain technology in the field of health Internet of things.
    Types Fields Applications
    Enterprise Medical Field Philips/Change Healthcare/PokitDok/Patientory/Hashed Healthcare/Doc.ai
    IT Field IBM/Google/Intel
    Blockchain Technology Field Gem/Factom/Bit Health/Chronicled/BlockVerify/DNA.Bits/ SimplyVital Health/Bitfury
    Platform MediBloc/Robomed Network/DokChain

     | Show Table
    DownLoad: CSV

    The explosion of IoT devices and increasing computing power have generated unprecedented amounts of data. As 5G networks continue to facilitate an increasing number of connected mobile devices and data volumes continue to grow, traditional infrastructure and the enormous amounts of data created by linked devices can no longer be handled by cloud computing.

    Figure 6 shows edge computing as an architecture of distributed computing, principally designed to curtail response latency and bandwidth consumption by strategically situating data storage and processing nearer to the demand source [83,84]. Edge computing as a concept is related to cloud computing, which processes all data by uploading it to a cloud data center or server with centralized computing resources for processing [84,85,86]. Any request to access the information should be processed in the cloud. Therefore, cloud computing disadvantages become progressively conspicuous amid the data deluge characteristic of the IoT era. First, cloud computing falls short of accommodating the surge in voluminous data processing. Second, it is incapable of fulfilling the requirements for instantaneous data processing. According to the conventional cloud computing concept, IoT data collected by the terminal has to be transmitted to the cloud computing center first, and then return the result after computing through clusters, which inevitably has a long response time; several emerging application scenarios including intelligent mines and unmanned driving have extremely high requirements for response time, and it is not realistic to rely on cloud computing. Unlike cloud computing based on a centralized computing environment, edge computing conducts data processing and analysis proximate to the point of data acquisition, circumventing the necessity for data transfer across the network to the cloud or data center, thereby reducing response time and increasing bandwidth availability. Because of its ability to process data and faster response time, edge computing holds good promise in the IoT space, especially in the health IoT.

    Figure 6.  Edge computing paradigm.

    In telemedicine, portable IoT healthcare devices developed by using edge computing technology can minimize latency and increase processing speed to collect, store, generate and analyze patient data, which physicians can use to assess a patient's health status in real time [87,88]. IoT medical devices are able to bolster the ambit of current networks by interfacing with edge data centers, which can give medical staff access to important patient data. In the field of emergency medical services, since most of the current emergency medical services are deployed in the cloud, the devices are vulnerable to and limited by the mobile environment and extreme weather before the emergency patient arrives at the hospital or during the transfer between two hospitals; however, with edge computing technology, one can establish a two-way real time communication channel between the ambulance and the hospital, enabling real-time natural language and image processing to improve timeliness and efficiency. Medical staff can view patients' electronic medical records in real-time to ensure the integrity of diagnosis and treatment information, and, at the same time, it can ensure the security of hospital operation information, basic patient information, condition information and many other types of private data [89]. This is made possible by using edge computing technology to realize mobile electronic medical records.

    Edge computing has been iterated in recent years, and the algorithm, hardware computing power and precisely labeled data on the device side have all made a qualitative leap. Through the use of edge computing technology, the health IoT field can realize high-quality and user-friendly medical products and services for global medical institutions, as well as help doctors to improve the level and efficiency of treatment to the greatest extent.

    One of the most significant technological advancements for the future of healthcare is the digital twin. As a result of utilizing digital twin technology, a variety of new medical testing and scanning instruments and wearable devices have the potential to help the healthcare industry deliver treatment options to patients more quickly and cost effectively. Table 4 lists the relevant applications of digital twins in the IoMT.

    Table 4.  Applications of digital twin.
    Refs Goals Methods
    [90] Create a digital twin of the liver to study disease and drug toxicity Developed a computational model of liver physiology by using ordinary differential equations
    [91] Aim to computationally treat digital twins with drugs to find optimal personalized therapies Proposed constructing digital twins of patients by integrating multi-omics, clinical and environmental data by using network analysis
    [92] Develop a remote surgical simulation system based on digital twinning and the IoT to support remote medical services Implemented deep learning for clinical data analysis, digital twinning for virtual patient modeling and augmented reality for remote surgical guidance
    [93] Design a digital twinning architecture for elderly safety monitoring and demonstrate real-time monitoring and anomaly detection feasibility IoT sensors for data collection, edge computing for preliminary analysis and blockchain for data storage and sharing
    [94] Enable virtual medical services via cancer digital twins and demonstrate reliability in monitoring and diagnosis within the metaverse context Used ML to create real-time cancer digital twins
    [95] Develop an efficient and privacy-aware AI system for smart healthcare Proposed a digital twin-assisted quantum federated learning technique for 5G-based intelligent medical diagnosis

     | Show Table
    DownLoad: CSV

    Healthcare personnel can initially access patient conditions by building a digital twin model of the facility. Moreover, digital twin technology can predict possible future diseases of patients and prevent patient emergencies, enabling a significant reduction in healthcare emergencies. In the field of drug clinical trial design, by using digital twin technology, researchers can build digital twins of experimental and control groups instead of using some human volunteers with the help of data from completed trials, relieving the pressure of patient recruitment and addressing the issues of high cost, significant time consumption and suboptimal efficiency of clinical trials. The employment of digital twin technology has made personalized medicine a major direction in future medical practice. Digital twin technology models an individual's genetic makeup, physical characteristics and lifestyle to digitize the human body; it creates a fully functional replica of the body's internal systems and organs, thereby enhancing medical treatments and patient care options. In the field of medical systems, by combining the digital twin of medical apparatuses (surgical beds, monitors, therapeutic instruments, etc.) with those of medical auxiliary equipment (human exoskeletons, wheelchairs, cardiac stents, etc.), the digital twin is poised to cultivate a groundbreaking platform and pioneering experimental pathway for personalized health management and healthcare services [96].

    With the advancement of technology and the global spread of COVID-19 in recent years, human health monitoring devices have come to be regarded as a constant tool in daily life [97]. People can know their health status anytime and anywhere through these devices. However, most of these products are contact measurement devices, which may cause discomfort to the user, pose a risk of disease transmission due to contact, etc. As the technology associated with the IoT continues to evolve, contactless health monitoring devices are becoming increasingly widely used, as shown in Figure 7.

    Figure 7.  Applications of contactless health monitoring devices.

    Leveraging the capabilities of the IoTs coupled with AI technology, the contactless heart rate monitoring technology can be used to capture the human face under natural light conditions through the use of ordinary cameras, and to separate, extract and analyze the operation of weak periodic signals to accurately analyze the basic physiological indicators of the detection target, thus realizing the monitoring of human body functions. Sleep quality has a serious impact on human health. Long-term sleep disorders will lead to disorders of human physiological functions [98]. Therefore, monitoring sleep patterns is vitally important for human health. With the use of IoT and AI technology, the Israeli company EarlySense has launched a contactless, continuous patient monitoring solution. Its self-developed contactless sleep monitoring device can help clinicians to detect the deterioration of the patient's condition early by monitoring the patient's heart rate, respiration and exercise data to avoid the occurrence of adverse events such as the need for first aid, intensive care unit transfer and bedsores. Within the domain of elderly care monitoring, through the use of IoT and AI technology, contactless elderly care monitoring devices can automatically alert caregivers in the event of an emergency, track the user's daily activities and record health changes, increasing the security of the elderly alone at home and enhancing the well-being of the elderly in care facilities, while also optimizing caregiver efficacy and minimizing operating expenditures for these institutions. Due to the impact of the COVID-19 pandemic, temperature monitoring has become a constant concern in people's daily lives. Traditional contact temperature measurement in public places is not only time-intensive, it also poses risks of spreading the infection. Through the use of IoT technology, contactless temperature monitoring can measure human body temperature anytime and anywhere, which is easy to operate and can enhance the efficiency of economic and social development to a certain extent; also, it has been widely used.

    Modern IoT and AI technologies have substantially influenced medical care, and the ongoing evolution of IoT technology will allow non-contact health monitoring devices to play an increasingly important role in daily life [99].

    The integration of 5G and the anticipated 6G networks into the IoMT landscape has catalyzed transformative changes in healthcare [100]. With the proliferation of medical sensors, wearables and other smart medical devices, there is an exponential growth of data being generated at the network edge [101]. This drives the need for high-bandwidth, low-latency networks to support real-time monitoring, diagnostics and treatment applications.

    5G, with its gigabit data rates, ultra-reliable low latency communications and support for massive machine-type communications, provides an ideal platform for the IoMT. Applications enabled by 5G include remote surgery, telemedicine, remote patient monitoring and medical data analytics using cloud computing [102]. However, issues like security, privacy, network slicing and edge computing integration need to be addressed. 6G will take the IoMT to the next level by providing holographic communications for an immersive medical experience, 3D ubiquitous coverage via aerial and satellite networks and extreme capacity and intelligence using AI [103,104]. Potential 6G healthcare use cases involve tactile Internet, bio-nano communications, augmented/virtual reality and decentralized IoMT networks. While offering new capabilities, 6G also faces challenges in terms of managing the network complexity, ensuring security and providing the desired quality of service.

    In conclusion, 5G and 6G will be game-changers for the IoMT, enabling personalized healthcare through wearables, remote diagnostics, AI-driven analytics and next-generation interfaces [105]. However, realizing the full potential requires extensive research and development efforts to address the emerging networking and security challenges. Stringent attention to security and privacy concerns is imperative [106]. By developing advanced security measures, adhering to regulatory standards and embracing innovative encryption technologies, the IoMT can fully harness the potential of 5G and 6G networks. As these technologies continue to mature, they hold the promise of reshaping healthcare delivery, ensuring patient privacy and ultimately improving healthcare outcomes on a global scale.

    While the IoMT is booming, there are many challenges. The summary of challenges is presented in Table 5. Cost-effectiveness has emerged as a key challenge limiting the mass adoption of IoMT systems, and it needs to be given sufficient attention. Although technological advancements have considerably reduced hardware prices, it would be foolish to undervalue the infrastructure and setup expenses of the IoMT. The sensors used in the IoMT to gather patient data, the network used to transmit that data through various nodes, the fog computing technology used to process and store the data, the numerous connected medical devices and the cloud technology used to store, process and analyze the data all have high maintenance and upgrade costs. The increase in cost relative to conventional medical systems comes mainly from the integration of technology. Therefore, for the development of IoMT devices, lower costs for maintaining sensors and the initial setup will be taken into account, along with increased implementation on a more routine basis.

    Table 5.  Summary of challenges in the IoMT.
    Challenges Description Solutions
    Cost-Effectiveness Infrastructure, setup and maintenance expenses and high costs from technology integration Focus on lower maintenance costs and regular implementation
    Security and privacy Vulnerability to malicious attacks, networks and sensors can be compromised [107,108,109] Develop proper authentication and verification protocols, encryption algorithms and blockchain technology [110,111,112,113]
    Accuracy Data inconsistencies, sensor inaccuracies, interference, overwhelming data volumes and integration difficulties due to varied communication protocols Focus on standardization of data formats, advanced analytics, rigorous device testing, secure connectivity, regular calibration, professional training, user feedback mechanisms, interoperability platforms and effective data filtering
    Power Consumption High power consumption and battery-powered devices need frequent replacements [114,115] Optimize data creation, reduce energy consumption in processing and transmission and design sustainable electricity-generating devices

     | Show Table
    DownLoad: CSV

    As sensitive patient health information proliferates and data architectures become more intricate, security is increasing as a major and ongoing challenge for the IoMT. Through the integration of the IoMT framework with technology, wearable devices show great promise in terms of adaptability to their users, particularly because of the data stored in the cloud that are capable of being processed and applied for further research, as well as the ability to screen user's progress. By 2025, it is anticipated that the healthcare sector will produce the most data of any other sector. These data transmitted through the IoT, as well as the Internet system itself, are vulnerable to malicious attacks. Sensors are the basis for data collection and data communication in the IoT. However, networks can be easily hacked and sensors can be compromised. The evolution of IoMT technology has concurrently engendered concerns about user privacy and data security [107,108,109]. Alasmari and Anwar [110] suggested that IoT devices often have operating principles that automate data collection and security vulnerabilities that ignore authentication and that the loss of medical data can have severe effects on people as well as the healthcare system. To alleviate this problem, Agrawal and Sharma [111] proposed that proper authentication and verification protocols and the deployment of appropriate encryption algorithms should be developed. User-controlled single sign-on telemedicine systems based on smartcards were presented in [112], which was able to protect user privacy and improve data security. Alsamhi and Lee [113] used blockchain technology to address security and privacy concerns and avoid exposing personal data.

    Indeed, while blockchain serves as a cornerstone for data integrity and trustworthiness, it is not the sole guardian. Techniques such as the Advanced Encryption Standard, which encrypts data in fixed-size blocks, and elliptic curve cryptography, which offers efficient public-key encryption, are paramount in ensuring robust data protection. Additionally, the incorporation of physical unclonable functions, which generate unique cryptographic keys based on device-specific physical characteristics, further enhances security, especially against hardware-focused attacks [116,117].

    When it comes to providing healthcare, the IoMT is particularly effective at evaluating, interpreting and using various forms of medical records for clinical decision-making. The accuracy of the data acquired by the sensors is a key factor that affects healthcare decisions. Inaccurate information has the ability to mislead and hurt patients. It is crucial to make sure that the tool's sharpness and accuracy hold steady through repeated applications. This guarantees that caregivers have a thorough understanding of each patient's health situation, and it allows for quicker, more precise therapy. IoT sensors and actuators collect a large amount of healthcare information, but this information is not always structured. These data closely resemble big data, and their characteristics include diversity, volume and a high frequency of data generation. Now, to enable insightful analysis and informed decisions, it becomes imperative to methodically structure and process this vast healthcare information. However, IoT devices cannot perform complex calculations in the field because of their limited data processing capabilities. Cloud technology could be one of the key technologies to address this issue [118]. It serves as a convenient, powerful and cost-effective technology to address numerous data challenges, including data integration and aggregation. It offers an on-demand usage approach and infinite computing power.

    Another obstacle that prevents the widespread implementation of IoMT devices is power consumption. Researchers have created new solutions to lower the energy consumption of various wireless gadgets [114,115]. Efforts to optimize data creation and minimize energy consumption in processing and transmission remain imperative. In addition, most IoMT devices are battery-powered, and the sensors require frequent battery replacement and short battery life, making long-term real-time patient monitoring impossible. Designing a medical device that can generate electricity sustainably should be the present priority, which could also help to alleviate the global energy crisis.

    In this paper, we provide an overview of IoMT architecture, key enabling technologies, applications across healthcare domains and challenges that must be addressed. A strength is the broad coverage of IoMT application domains. However, technical challenges like security, privacy and accuracy need deeper investigation. While the IoMT holds the potential to transform healthcare through remote monitoring, early diagnosis and reduced costs, interdisciplinary collaboration is required to realize this potential. Key areas for future investigation include the development of robust data security mechanisms for the IoMT, improved energy efficiency of wearable sensors and studies examining the clinical efficacy and ethical considerations of IoMT technologies. As IoMT continues to mature, we expect solutions to the outlined challenges to emerge, paving the way for more smart, connected and efficient healthcare systems.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    This work was funded by the National Natural Science Foundation of China (No. 62171114).

    The authors declare there is no conflict of interest.



    [1] G. J. Zhang, F. Z. Zhang, Z. H. Zhang, Y. Xiang, N. J. Yuan, X. Xie, et al., DRN: A deep reinforcement learning framework for news recommendation, in Proceedings of the 2018 World Wide Web Conference, (2018), 167–176. https://doi.org/10.1145/3178876.3185994
    [2] Q. M. Diao, M. H. Qiu, C. Y. Wu, A. J. Smola, J Jiang, C. Wang, Jointly modeling aspects, ratings and sentiments for movie recommendation, in Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, (2014), 193–202. https://doi.org/10.1145/2623330.2623758
    [3] G. R. Zhou, X. Q. Zhu, C. R. Song, Y. Fan, H. Zhu, X. Ma, et al., Deep interest network for click-through rate prediction, in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, (2018), 1059–1068. https://doi.org/10.1145/3219819.32198234.
    [4] R. Ma, Q. Zhang, J. Wang, L. Z. Cui, X. J. Huang, Mention recommendation for multimodal microblog with cross-attention memory network, in Proceedings of the 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, (2018), 195–204.
    [5] B. Jin, C. Cao, X. He, D. P. Jin, Y. Li, Multi-behavior recommendation with graph convolutional networks, in Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, (2020), 659–668. https://doi.org/10.1145/3397271.3401072
    [6] C. Chen, W. Ma, M. Zhang, Z. W. Wang, X. Q. He, C. Y. Wang, et al. Graph heterogeneous multi-relational recommendation, in Proceedings of the AAAI Conference on Artificial Intelligence, 35 (2021), 3958–3966. https://doi.org/10.1609/aaai.v35i5.16515
    [7] Y. Ni, D. Ou, S. Liu, X. Li, W. W. Ou, A. X. Zeng, et al., Perceive your users in depth: Learning universal user representations from multiple e-commerce tasks, in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, (2018), 596–605.
    [8] L. Guo, L. Hua, R. Jia, B. Q. Zhao, X. B. Wang, B. Cui, Buying or browsing? Predicting real-time purchasing intent using attention-based deep network with multiple behavior, in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, (2019), 1984–1992. https://doi.org/10.1145/3292500.3330670
    [9] S. Hochreiter, J. Schmidhuber, Long short-term memory, Neural Comput., 9 (1997), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735 doi: 10.1162/neco.1997.9.8.1735
    [10] Y. Tang, Y. Huang, Z. Wu, H. L. Meng; M. X. Xu; L. H. Cai, et al., Question detection from acoustic features using recurrent neural network with gated recurrent unit, in 2016 IEEE International Conference on Acoustics, Speech and Signal Processing, (2016), 6125–6129. https://doi.org/10.1109/ICASSP.2016.7472854
    [11] L. Xia, C. Huang, Y. Xu, P. Dai, X. Y. Zhang, H. S. Yang, et al., Knowledge-enhanced hierarchical graph transformer network for multi-behavior recommendation, in Proceedings of the AAAI Conference on Artificial Intelligence, 35 (2021), 4486–4493. https://doi.org/10.1609/aaai.v35i5.16576
    [12] Z. Wang, J. Zhang, J. Feng, Z. Chen, Knowledge graph embedding by translating on hyperplanes, in Proceedings of the AAAI conference on artificial intelligence, 28 (2014). https://doi.org/10.1609/aaai.v28i1.8870
    [13] L. Xia, Y Xu, C Huang, P. Dai, L. f. Bo, Graph meta network for multi-behavior recommendation, in Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval, (2021), 757–766. https://doi.org/10.1145/3404835.3462972
    [14] S. Gu, X. Wang, C. Shi, D. Xiao, Self-supervised Graph Neural Networks for Multi-behavior Recommendation, in International Joint Conference on Artificial Intelligence, 2022 (2022).
    [15] W. Wei, C. Huang, L. Xia, Y. Xu, J. S. Zhao, D. W. Yin, Contrastive meta learning with behavior multiplicity for recommendation, in Proceedings of the fifteenth ACM international conference on web search and data mining, (2022), 1120–1128.
    [16] Y. Wu, R. Xie, Y. Zhu, X. Ao, X. Chen, X. Zhang, et al., Multi-view multi-behavior contrastive learning in recommendation, in Database Systems for Advanced Applications: 27th International Conference, (2022), 166–182. https://doi.org/10.1007/978-3-031-00126-0_11
    [17] A. Da'u, N. Salim, Recommendation system based on deep learning methods: A systematic review and new directions, Artif. Intell. Rev., 53 (2020), 2709–2748. https://doi.org/10.1007/s10462-019-09744-1 doi: 10.1007/s10462-019-09744-1
    [18] J. Wei, J. He, K. Chen, Y. Zhou, Z. Y. Tang, et al., Collaborative filtering and deep learning based recommendation system for cold start items, Expert Syst. Appl., 69 (2017), 29–39. https://doi.org/10.1016/j.eswa.2016.09.040 doi: 10.1016/j.eswa.2016.09.040
    [19] M. Fu, H. Qu, Z. Yi, L. Lu, Y. S. Liu, et al., A novel deep learning-based collaborative filtering model for recommendation system, IEEE Transact. Cybern., 49 (2018), 1084–1096. https://doi.org/10.1109/TCYB.2018.2795041 doi: 10.1109/TCYB.2018.2795041
    [20] F. Scarselli, M. Gori, A. C. Tsoi, M. Hagenbuchner, G. Monfardini, et al., The graph neural network model, IEEE Transact. Neural Networks, 20 (2008), 61–80. https://doi.org/10.1109/TNN.2008.2005605 doi: 10.1109/TNN.2008.2005605
    [21] C. Zhang, D. Song, C. Huang, A. Swami, N. V. Chawla, et al., Heterogeneous graph neural network, in Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, (2019), 793–803. https://doi.org/10.1145/3292500.3330961
    [22] J Zhou, G Cui, S. Hu, Z. Y. Zhang, C. Yang, Z. Y. Liu, et al., Graph neural networks: A review of methods and applications, AI Open, 1 (2020), 57–81. https://doi.org/10.1016/j.aiopen.2021.01.001 doi: 10.1016/j.aiopen.2021.01.001
    [23] W. Zaremba, I. Sutskever, O. Vinyals, Recurrent neural network regularization, arXiv preprint. 2014 (2014).
  • This article has been cited by:

    1. Kulsoom Saima Bughio, David Michael Cook, Syed Afaq Ali Shah, 2024, Chapter 26, 978-981-97-5936-1, 314, 10.1007/978-981-97-5937-8_26
    2. Kulsoom S. Bughio, David M. Cook, Syed Afaq A. Shah, Developing a Novel Ontology for Cybersecurity in Internet of Medical Things-Enabled Remote Patient Monitoring, 2024, 24, 1424-8220, 2804, 10.3390/s24092804
    3. Shorouk E. El-deep, Amr A. Abohany, Karam M. Sallam, Amr A. Abd El-Mageed, A comprehensive survey on impact of applying various technologies on the internet of medical things, 2025, 58, 1573-7462, 10.1007/s10462-024-11063-z
    4. Jiachen Hou, Jing Li, 2025, 9780443303005, 83, 10.1016/B978-0-443-30300-5.00007-5
    5. Qinyu Liu, Chengshu Lu, Qizhuang Lv, Lanjie Lei, Emerging point-of-care testing technology for the detection of animal pathogenic microorganisms, 2025, 512, 13858947, 162548, 10.1016/j.cej.2025.162548
    6. Aliyu Sani Bashiru, H. Badamasi, A. Bashar, OVERVIEW OF IOT ON REMOTE PATIENT FOR MONITORING AND SMART MEDICAL DEVICES, 2025, 9, 2616-1370, 262, 10.33003/fjs-2025-0904-3595
    7. Shatha A. Baker, Kifaa Hadi Thanoon, Marwah Najm Abed, Muataz Almansor, Taha Y. Abdulqader, Omar I. Alsaif, 2025, A Secure and Privacy-Centric Blockchain Platform for Cloud-Based Healthcare Data Management, 979-8-3503-5546-8, 1, 10.1109/ITIKD63574.2025.11004752
  • Reader Comments
  • © 2023 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1909) PDF downloads(55) Cited by(0)

Figures and Tables

Figures(12)  /  Tables(5)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog