Smart city applications that request sensitive user information necessitate a comprehensive data privacy solution. Federated learning (FL), also known as privacy by design, is a new paradigm in machine learning (ML). However, FL models are susceptible to adversarial attacks, similar to other AI models. In this paper, we propose federated adversarial training (FAT) strategies to generate robust global models that are resistant to adversarial attacks. We apply two adversarial attack methods, projected gradient descent (PGD) and the fast gradient sign method (FGSM), to our air pollution dataset to generate adversarial samples. We then evaluate the effectiveness of our FAT strategies in defending against these attacks. Our experiments show that FGSM-based adversarial attacks have a negligible impact on the accuracy of global models, while PGD-based attacks are more effective. However, we also show that our FAT strategies can make global models robust enough to withstand even PGD-based attacks. For example, the accuracy of our FAT-PGD and FL-mixed-PGD models is 81.13% and 82.60%, respectively, compared to 91.34% for the baseline FL model. This represents a reduction in accuracy of 10%, but this could be potentially mitigated by using a more complex and larger model. Our results demonstrate that FAT can enhance the security and privacy of sustainable smart city applications. We also show that it is possible to train robust global models from modest datasets per client, which challenges the conventional wisdom that adversarial training requires massive datasets.
{"title":"Federated Adversarial Training Strategies for Achieving Privacy and Security in Sustainable Smart City Applications","authors":"Sapdo Utomo, Adarsh Rouniyar, Hsiu-Chun Hsu, Pao-Ann Hsiung","doi":"10.3390/fi15110371","DOIUrl":"https://doi.org/10.3390/fi15110371","url":null,"abstract":"Smart city applications that request sensitive user information necessitate a comprehensive data privacy solution. Federated learning (FL), also known as privacy by design, is a new paradigm in machine learning (ML). However, FL models are susceptible to adversarial attacks, similar to other AI models. In this paper, we propose federated adversarial training (FAT) strategies to generate robust global models that are resistant to adversarial attacks. We apply two adversarial attack methods, projected gradient descent (PGD) and the fast gradient sign method (FGSM), to our air pollution dataset to generate adversarial samples. We then evaluate the effectiveness of our FAT strategies in defending against these attacks. Our experiments show that FGSM-based adversarial attacks have a negligible impact on the accuracy of global models, while PGD-based attacks are more effective. However, we also show that our FAT strategies can make global models robust enough to withstand even PGD-based attacks. For example, the accuracy of our FAT-PGD and FL-mixed-PGD models is 81.13% and 82.60%, respectively, compared to 91.34% for the baseline FL model. This represents a reduction in accuracy of 10%, but this could be potentially mitigated by using a more complex and larger model. Our results demonstrate that FAT can enhance the security and privacy of sustainable smart city applications. We also show that it is possible to train robust global models from modest datasets per client, which challenges the conventional wisdom that adversarial training requires massive datasets.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"176 5","pages":""},"PeriodicalIF":3.4,"publicationDate":"2023-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139257197","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Edge AI, an interdisciplinary technology that enables distributed intelligence with edge devices, is quickly becoming a critical component in early health prediction. Edge AI encompasses data analytics and artificial intelligence (AI) using machine learning, deep learning, and federated learning models deployed and executed at the edge of the network, far from centralized data centers. AI enables the careful analysis of large datasets derived from multiple sources, including electronic health records, wearable devices, and demographic information, making it possible to identify intricate patterns and predict a person’s future health. Federated learning, a novel approach in AI, further enhances this prediction by enabling collaborative training of AI models on distributed edge devices while maintaining privacy. Using edge computing, data can be processed and analyzed locally, reducing latency and enabling instant decision making. This article reviews the role of Edge AI in early health prediction and highlights its potential to improve public health. Topics covered include the use of AI algorithms for early detection of chronic diseases such as diabetes and cancer and the use of edge computing in wearable devices to detect the spread of infectious diseases. In addition to discussing the challenges and limitations of Edge AI in early health prediction, this article emphasizes future research directions to address these concerns and the integration with existing healthcare systems and explore the full potential of these technologies in improving public health.
{"title":"Edge AI for Early Detection of Chronic Diseases and the Spread of Infectious Diseases: Opportunities, Challenges, and Future Directions","authors":"E. Badidi","doi":"10.3390/fi15110370","DOIUrl":"https://doi.org/10.3390/fi15110370","url":null,"abstract":"Edge AI, an interdisciplinary technology that enables distributed intelligence with edge devices, is quickly becoming a critical component in early health prediction. Edge AI encompasses data analytics and artificial intelligence (AI) using machine learning, deep learning, and federated learning models deployed and executed at the edge of the network, far from centralized data centers. AI enables the careful analysis of large datasets derived from multiple sources, including electronic health records, wearable devices, and demographic information, making it possible to identify intricate patterns and predict a person’s future health. Federated learning, a novel approach in AI, further enhances this prediction by enabling collaborative training of AI models on distributed edge devices while maintaining privacy. Using edge computing, data can be processed and analyzed locally, reducing latency and enabling instant decision making. This article reviews the role of Edge AI in early health prediction and highlights its potential to improve public health. Topics covered include the use of AI algorithms for early detection of chronic diseases such as diabetes and cancer and the use of edge computing in wearable devices to detect the spread of infectious diseases. In addition to discussing the challenges and limitations of Edge AI in early health prediction, this article emphasizes future research directions to address these concerns and the integration with existing healthcare systems and explore the full potential of these technologies in improving public health.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"53 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2023-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139262414","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the field of ocean data monitoring, collaborative control and path planning of unmanned aerial vehicles (UAVs) are essential for improving data collection efficiency and quality. In this study, we focus on how to utilize multiple UAVs to efficiently cover the target area in ocean data monitoring tasks. First, we propose a multiagent deep reinforcement learning (DRL)-based path-planning method for multiple UAVs to perform efficient coverage tasks in a target area in the field of ocean data monitoring. Additionally, the traditional Multi-Agent Twin Delayed Deep Deterministic policy gradient (MATD3) algorithm only considers the current state of the agents, leading to poor performance in path planning. To address this issue, we introduce an improved MATD3 algorithm with the integration of a stacked long short-term memory (S-LSTM) network to incorporate the historical interaction information and environmental changes among agents. Finally, the experimental results demonstrate that the proposed MATD3-Stacked_LSTM algorithm can effectively improve the efficiency and practicality of UAV path planning by achieving a high coverage rate of the target area and reducing the redundant coverage rate among UAVs compared with two other advanced DRL algorithms.
{"title":"Maximizing UAV Coverage in Maritime Wireless Networks: A Multiagent Reinforcement Learning Approach","authors":"Qianqian Wu, Qiang Liu, Zefan Wu, Jiye Zhang","doi":"10.3390/fi15110369","DOIUrl":"https://doi.org/10.3390/fi15110369","url":null,"abstract":"In the field of ocean data monitoring, collaborative control and path planning of unmanned aerial vehicles (UAVs) are essential for improving data collection efficiency and quality. In this study, we focus on how to utilize multiple UAVs to efficiently cover the target area in ocean data monitoring tasks. First, we propose a multiagent deep reinforcement learning (DRL)-based path-planning method for multiple UAVs to perform efficient coverage tasks in a target area in the field of ocean data monitoring. Additionally, the traditional Multi-Agent Twin Delayed Deep Deterministic policy gradient (MATD3) algorithm only considers the current state of the agents, leading to poor performance in path planning. To address this issue, we introduce an improved MATD3 algorithm with the integration of a stacked long short-term memory (S-LSTM) network to incorporate the historical interaction information and environmental changes among agents. Finally, the experimental results demonstrate that the proposed MATD3-Stacked_LSTM algorithm can effectively improve the efficiency and practicality of UAV path planning by achieving a high coverage rate of the target area and reducing the redundant coverage rate among UAVs compared with two other advanced DRL algorithms.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"10 4","pages":""},"PeriodicalIF":3.4,"publicationDate":"2023-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139269798","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Giacomo Gori, Lorenzo Rinieri, Amir Al Sadi, A. Melis, Franco Callegati, Marco Prandini
The correct and efficient measurement of security properties is key to the deployment of effective cyberspace protection strategies. In this work, we propose GRAPH4, which is a system that combines different security metrics to design an attack detection approach that leverages the advantages of modern network architectures. GRAPH4 makes use of attack graphs that are generated by the control plane to extract a view of the network components requiring monitoring, which is based on the specific attack that must be detected and on the knowledge of the complete network layout. It enables an efficient distribution of security metrics tasks between the control plane and the data plane. The attack graph is translated into network rules that are subsequently installed in programmable nodes in order to enable alerting and detecting network anomalies at a line rate. By leveraging data plane programmability and security metric scores, GRAPH4 enables timely responses to unforeseen conditions while optimizing resource allocation and enhancing proactive defense. This paper details the architecture of GRAPH4, and it provides an evaluation of the performance gains it can achieve.
{"title":"GRAPH4: A Security Monitoring Architecture Based on Data Plane Anomaly Detection Metrics Calculated over Attack Graphs","authors":"Giacomo Gori, Lorenzo Rinieri, Amir Al Sadi, A. Melis, Franco Callegati, Marco Prandini","doi":"10.3390/fi15110368","DOIUrl":"https://doi.org/10.3390/fi15110368","url":null,"abstract":"The correct and efficient measurement of security properties is key to the deployment of effective cyberspace protection strategies. In this work, we propose GRAPH4, which is a system that combines different security metrics to design an attack detection approach that leverages the advantages of modern network architectures. GRAPH4 makes use of attack graphs that are generated by the control plane to extract a view of the network components requiring monitoring, which is based on the specific attack that must be detected and on the knowledge of the complete network layout. It enables an efficient distribution of security metrics tasks between the control plane and the data plane. The attack graph is translated into network rules that are subsequently installed in programmable nodes in order to enable alerting and detecting network anomalies at a line rate. By leveraging data plane programmability and security metric scores, GRAPH4 enables timely responses to unforeseen conditions while optimizing resource allocation and enhancing proactive defense. This paper details the architecture of GRAPH4, and it provides an evaluation of the performance gains it can achieve.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"81 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2023-11-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139274725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Afzal Badshah, Ghani Ur Rehman, Haleem Farman, Anwar Ghani, Shahid Sultan, Muhammad Zubair, Moustafa M. Nasralla
The Internet of Things (IoT), cloud, and fog computing are now a reality and have become the vision of the smart world. Self-directed learning approaches, their tools, and smart spaces are transforming traditional institutions into smart institutions. This transition has a positive impact on learner engagement, motivation, attendance, and advanced learning outcomes. In developing countries, there are many barriers to quality education, such as inadequate implementation of standard operating procedures, lack of involvement from learners and parents, and lack of transparent performance measurement for both institutions and students. These issues need to be addressed to ensure further growth and improvement. This study explored the use of smart technologies (IoT, fog, and cloud computing) to address challenges in student learning and administrative tasks. A novel framework (a five-element smart institution framework) is proposed to connect administrators, teachers, parents, and students using smart technologies to improve attendance, pedagogy, and evaluation. The results showed significant increases in student attendance and homework progress, along with improvements in annual results, student discipline, and teacher/parent engagement.
{"title":"Transforming Educational Institutions: Harnessing the Power of Internet of Things, Cloud, and Fog Computing","authors":"Afzal Badshah, Ghani Ur Rehman, Haleem Farman, Anwar Ghani, Shahid Sultan, Muhammad Zubair, Moustafa M. Nasralla","doi":"10.3390/fi15110367","DOIUrl":"https://doi.org/10.3390/fi15110367","url":null,"abstract":"The Internet of Things (IoT), cloud, and fog computing are now a reality and have become the vision of the smart world. Self-directed learning approaches, their tools, and smart spaces are transforming traditional institutions into smart institutions. This transition has a positive impact on learner engagement, motivation, attendance, and advanced learning outcomes. In developing countries, there are many barriers to quality education, such as inadequate implementation of standard operating procedures, lack of involvement from learners and parents, and lack of transparent performance measurement for both institutions and students. These issues need to be addressed to ensure further growth and improvement. This study explored the use of smart technologies (IoT, fog, and cloud computing) to address challenges in student learning and administrative tasks. A novel framework (a five-element smart institution framework) is proposed to connect administrators, teachers, parents, and students using smart technologies to improve attendance, pedagogy, and evaluation. The results showed significant increases in student attendance and homework progress, along with improvements in annual results, student discipline, and teacher/parent engagement.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"57 8","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136347046","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Linear networks (sometimes called chain-type networks) occur frequently in Internet of Things (IoT) applications, where sensors or actuators are deployed along pipelines, roads, railways, mines, and international borders. LoRa, short for Long Range, is an increasingly important technology for the IoT with great potential for linear networking. Despite its potential, limited research has explored LoRa’s implementation in such networks. In this paper, we addressed two important issues related to LoRa linear networks. The first is contention, when multiple nodes attempt to access a shared channel. Although originally designed to deal with interference, LoRa’s technique of synchronisation with a transmission node permits a novel approach to contention, which we explored. The second issue revolves around routing, where linear networks permit simpler strategies, in contrast to the common routing complexities of mesh networks. We present gossip routing as a very lightweight approach to routing. All our evaluations were carried out using real equipment by developing real networks. We constructed networks of up to three hops in length and up to three nodes in width. We carried out experiments looking at contention and routing. We demonstrate using the novel approach that we could achieve up to 98% throughput. We compared its performance considering collocated scenarios that achieved 84% and 89% throughputby using relay widths of two and three at each hop, respectively. Lastly, we demonstrate the effectiveness of gossip routing by using various transmission probabilities. We noticed high performance up to 98% throughputat Tprob = 0.90 and Tprob = 0.80 by employing two and three active relay nodes, respectively. The experimental result showed that, at Tprob = 0.40, it achieved an average performance of 62.8% and 73.77% by using two and three active relay nodes, respectively. We concluded that LoRa is an excellent technology for Internet of Things applications where sensors and actuators are deployed in an approximately linear fashion.
{"title":"Design Considerations and Performance Evaluation of Gossip Routing in LoRa-Based Linear Networks","authors":"Rao Muzamal Liaqat, Philip Branch, Jason But","doi":"10.3390/fi15110366","DOIUrl":"https://doi.org/10.3390/fi15110366","url":null,"abstract":"Linear networks (sometimes called chain-type networks) occur frequently in Internet of Things (IoT) applications, where sensors or actuators are deployed along pipelines, roads, railways, mines, and international borders. LoRa, short for Long Range, is an increasingly important technology for the IoT with great potential for linear networking. Despite its potential, limited research has explored LoRa’s implementation in such networks. In this paper, we addressed two important issues related to LoRa linear networks. The first is contention, when multiple nodes attempt to access a shared channel. Although originally designed to deal with interference, LoRa’s technique of synchronisation with a transmission node permits a novel approach to contention, which we explored. The second issue revolves around routing, where linear networks permit simpler strategies, in contrast to the common routing complexities of mesh networks. We present gossip routing as a very lightweight approach to routing. All our evaluations were carried out using real equipment by developing real networks. We constructed networks of up to three hops in length and up to three nodes in width. We carried out experiments looking at contention and routing. We demonstrate using the novel approach that we could achieve up to 98% throughput. We compared its performance considering collocated scenarios that achieved 84% and 89% throughputby using relay widths of two and three at each hop, respectively. Lastly, we demonstrate the effectiveness of gossip routing by using various transmission probabilities. We noticed high performance up to 98% throughputat Tprob = 0.90 and Tprob = 0.80 by employing two and three active relay nodes, respectively. The experimental result showed that, at Tprob = 0.40, it achieved an average performance of 62.8% and 73.77% by using two and three active relay nodes, respectively. We concluded that LoRa is an excellent technology for Internet of Things applications where sensors and actuators are deployed in an approximately linear fashion.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"31 9","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135086366","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Nowadays, e-commerce websites have become part of people’s daily lives; therefore, it has become necessary to seek help in assessing and improving the usability of the services of e-commerce websites. Essentially, usability studies offer significant information about users’ assessment and perceptions of satisfaction, effectiveness, and efficiency of online services. This research investigated the usability of two e-commerce web-sites in Saudi Arabia and compared the effectiveness of different behavioral measurement techniques, such as heuristic evaluation, usability testing, and eye-tracking. In particular, this research selected the Extra and Jarir e-commerce websites in Saudi Arabia based on a combined approach of criteria and ranking. This research followed an experimental approach in which both qualitative and quantitative approaches were employed to collect and analyze the data. Each of the behavioral measurement techniques identified usability issues ranging from cosmetic to catastrophic issues. It is worth mentioning that the heuristic evaluation by experts provided both the majority of the issues and identified the most severe usability issues compared to the number of issues identified by both usability testing and eye-tracking combined. Usability testing provided fewer problems, most of which had already been identified by the experts. Eye-tracking provided critical information regarding the page design and element placements and revealed certain user behavior patterns that indicated certain usability problems. Overall, the research findings appeared useful to user experience (UX) and user interface (UI) designers to consider the provided recommendations to enhance the usability of e-commerce websites.
{"title":"Assessing Interactive Web-Based Systems Using Behavioral Measurement Techniques","authors":"Thanaa Saad AlSalem, Majed Aadi AlShamari","doi":"10.3390/fi15110365","DOIUrl":"https://doi.org/10.3390/fi15110365","url":null,"abstract":"Nowadays, e-commerce websites have become part of people’s daily lives; therefore, it has become necessary to seek help in assessing and improving the usability of the services of e-commerce websites. Essentially, usability studies offer significant information about users’ assessment and perceptions of satisfaction, effectiveness, and efficiency of online services. This research investigated the usability of two e-commerce web-sites in Saudi Arabia and compared the effectiveness of different behavioral measurement techniques, such as heuristic evaluation, usability testing, and eye-tracking. In particular, this research selected the Extra and Jarir e-commerce websites in Saudi Arabia based on a combined approach of criteria and ranking. This research followed an experimental approach in which both qualitative and quantitative approaches were employed to collect and analyze the data. Each of the behavioral measurement techniques identified usability issues ranging from cosmetic to catastrophic issues. It is worth mentioning that the heuristic evaluation by experts provided both the majority of the issues and identified the most severe usability issues compared to the number of issues identified by both usability testing and eye-tracking combined. Usability testing provided fewer problems, most of which had already been identified by the experts. Eye-tracking provided critical information regarding the page design and element placements and revealed certain user behavior patterns that indicated certain usability problems. Overall, the research findings appeared useful to user experience (UX) and user interface (UI) designers to consider the provided recommendations to enhance the usability of e-commerce websites.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"31 14","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135086361","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Ye Yuan, Wang Wang, Guangze Wen, Zikun Zheng, Zhemin Zhuang
Product reviews provide crucial information for both consumers and businesses, offering insights needed before purchasing a product or service. However, existing sentiment analysis methods, especially for Chinese language, struggle to effectively capture contextual information due to the complex semantics, multiple sentiment polarities, and long-term dependencies between words. In this paper, we propose a sentiment classification method based on the BiLSTM algorithm to address these challenges in natural language processing. Self-Attention-CNN BiLSTM (SAC-BiLSTM) leverages dual channels to extract features from both character-level embeddings and word-level embeddings. It combines BiLSTM and Self-Attention mechanisms for feature extraction and weight allocation, aiming to overcome the limitations in mining contextual information. Experiments were conducted on the onlineshopping10cats dataset, which is a standard corpus of e-commerce shopping reviews available in the ChineseNlpCorpus 2018. The experimental results demonstrate the effectiveness of our proposed algorithm, with Recall, Precision, and F1 scores reaching 0.9409, 0.9369, and 0.9404, respectively.
{"title":"Sentiment Analysis of Chinese Product Reviews Based on Fusion of DUAL-Channel BiLSTM and Self-Attention","authors":"Ye Yuan, Wang Wang, Guangze Wen, Zikun Zheng, Zhemin Zhuang","doi":"10.3390/fi15110364","DOIUrl":"https://doi.org/10.3390/fi15110364","url":null,"abstract":"Product reviews provide crucial information for both consumers and businesses, offering insights needed before purchasing a product or service. However, existing sentiment analysis methods, especially for Chinese language, struggle to effectively capture contextual information due to the complex semantics, multiple sentiment polarities, and long-term dependencies between words. In this paper, we propose a sentiment classification method based on the BiLSTM algorithm to address these challenges in natural language processing. Self-Attention-CNN BiLSTM (SAC-BiLSTM) leverages dual channels to extract features from both character-level embeddings and word-level embeddings. It combines BiLSTM and Self-Attention mechanisms for feature extraction and weight allocation, aiming to overcome the limitations in mining contextual information. Experiments were conducted on the onlineshopping10cats dataset, which is a standard corpus of e-commerce shopping reviews available in the ChineseNlpCorpus 2018. The experimental results demonstrate the effectiveness of our proposed algorithm, with Recall, Precision, and F1 scores reaching 0.9409, 0.9369, and 0.9404, respectively.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"121 11","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135137872","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this article, we investigate the potential of synthetic resumes as a means for the rapid generation of training data and their effectiveness in data augmentation, especially in categories marked by sparse samples. The widespread implementation of machine learning algorithms in natural language processing (NLP) has notably streamlined the resume classification process, delivering time and cost efficiencies for hiring organizations. However, the performance of these algorithms depends on the abundance of training data. While selecting the right model architecture is essential, it is also crucial to ensure the availability of a robust, well-curated dataset. For many categories in the job market, data sparsity remains a challenge. To deal with this challenge, we employed the OpenAI API to generate both structured and unstructured resumes tailored to specific criteria. These synthetically generated resumes were cleaned, preprocessed and then utilized to train two distinct models: a transformer model (BERT) and a feedforward neural network (FFNN) that incorporated Universal Sentence Encoder 4 (USE4) embeddings. While both models were evaluated on the multiclass classification task of resumes, when trained on an augmented dataset containing 60 percent real data (from Indeed website) and 40 percent synthetic data from ChatGPT, the transformer model presented exceptional accuracy. The FFNN, albeit predictably, achieved lower accuracy. These findings highlight the value of augmented real-world data with ChatGPT-generated synthetic resumes, especially in the context of limited training data. The suitability of the BERT model for such classification tasks further reinforces this narrative.
{"title":"Generating Synthetic Resume Data with Large Language Models for Enhanced Job Description Classification","authors":"Panagiotis Skondras, Panagiotis Zervas, Giannis Tzimas","doi":"10.3390/fi15110363","DOIUrl":"https://doi.org/10.3390/fi15110363","url":null,"abstract":"In this article, we investigate the potential of synthetic resumes as a means for the rapid generation of training data and their effectiveness in data augmentation, especially in categories marked by sparse samples. The widespread implementation of machine learning algorithms in natural language processing (NLP) has notably streamlined the resume classification process, delivering time and cost efficiencies for hiring organizations. However, the performance of these algorithms depends on the abundance of training data. While selecting the right model architecture is essential, it is also crucial to ensure the availability of a robust, well-curated dataset. For many categories in the job market, data sparsity remains a challenge. To deal with this challenge, we employed the OpenAI API to generate both structured and unstructured resumes tailored to specific criteria. These synthetically generated resumes were cleaned, preprocessed and then utilized to train two distinct models: a transformer model (BERT) and a feedforward neural network (FFNN) that incorporated Universal Sentence Encoder 4 (USE4) embeddings. While both models were evaluated on the multiclass classification task of resumes, when trained on an augmented dataset containing 60 percent real data (from Indeed website) and 40 percent synthetic data from ChatGPT, the transformer model presented exceptional accuracy. The FFNN, albeit predictably, achieved lower accuracy. These findings highlight the value of augmented real-world data with ChatGPT-generated synthetic resumes, especially in the context of limited training data. The suitability of the BERT model for such classification tasks further reinforces this narrative.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":" 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135241383","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Farouq E. Shaibu, Elizabeth N. Onwuka, Nathaniel Salawu, Stephen S. Oyewobi, Karim Djouani, Adnan M. Abu-Mahfouz
The rapid development of 5G communication networks has ushered in a new era of high-speed, low-latency wireless connectivity, as well as the enabling of transformative technologies. However, a crucial aspect of ensuring reliable communication is the accurate modeling of path loss, as it directly impacts signal coverage, interference, and overall network efficiency. This review paper critically assesses the performance of path loss models in mid-band and high-band frequencies and examines their effectiveness in addressing the challenges of 5G deployment. In this paper, we first present the summary of the background, highlighting the increasing demand for high-quality wireless connectivity and the unique characteristics of mid-band (1–6 GHz) and high-band (>6 GHz) frequencies in the 5G spectrum. The methodology comprehensively reviews some of the existing path loss models, considering both empirical and machine learning approaches. We analyze the strengths and weaknesses of these models, considering factors such as urban and suburban environments and indoor scenarios. The results highlight the significant advancements in path loss modeling for mid-band and high-band 5G channels. In terms of prediction accuracy and computing effectiveness, machine learning models performed better than empirical models in both mid-band and high-band frequency spectra. As a result, they might be suggested as an alternative yet promising approach to predicting path loss in these bands. We consider the results of this review to be promising, as they provide network operators and researchers with valuable insights into the state-of-the-art path loss models for mid-band and high-band 5G channels. Future work suggests tuning an ensemble machine learning model to enhance a stable empirical model with multiple parameters to develop a hybrid path loss model for the mid-band frequency spectrum.
{"title":"Performance of Path Loss Models over Mid-Band and High-Band Channels for 5G Communication Networks: A Review","authors":"Farouq E. Shaibu, Elizabeth N. Onwuka, Nathaniel Salawu, Stephen S. Oyewobi, Karim Djouani, Adnan M. Abu-Mahfouz","doi":"10.3390/fi15110362","DOIUrl":"https://doi.org/10.3390/fi15110362","url":null,"abstract":"The rapid development of 5G communication networks has ushered in a new era of high-speed, low-latency wireless connectivity, as well as the enabling of transformative technologies. However, a crucial aspect of ensuring reliable communication is the accurate modeling of path loss, as it directly impacts signal coverage, interference, and overall network efficiency. This review paper critically assesses the performance of path loss models in mid-band and high-band frequencies and examines their effectiveness in addressing the challenges of 5G deployment. In this paper, we first present the summary of the background, highlighting the increasing demand for high-quality wireless connectivity and the unique characteristics of mid-band (1–6 GHz) and high-band (>6 GHz) frequencies in the 5G spectrum. The methodology comprehensively reviews some of the existing path loss models, considering both empirical and machine learning approaches. We analyze the strengths and weaknesses of these models, considering factors such as urban and suburban environments and indoor scenarios. The results highlight the significant advancements in path loss modeling for mid-band and high-band 5G channels. In terms of prediction accuracy and computing effectiveness, machine learning models performed better than empirical models in both mid-band and high-band frequency spectra. As a result, they might be suggested as an alternative yet promising approach to predicting path loss in these bands. We consider the results of this review to be promising, as they provide network operators and researchers with valuable insights into the state-of-the-art path loss models for mid-band and high-band 5G channels. Future work suggests tuning an ensemble machine learning model to enhance a stable empirical model with multiple parameters to develop a hybrid path loss model for the mid-band frequency spectrum.","PeriodicalId":37982,"journal":{"name":"Future Internet","volume":"64 5","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135540128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：