Web Information Systems Engineering – WISE 2024

This paper addresses a problem of constructing a Learning-to-Rank (LtR) model tailored to a target domain without using any domain-specific queries and relevance judgements. Our proposed method, PR-Rank, incorporates domain features, which are represented in a real-valued vector and can be estimated by domain experts, for adapting LtR models. The key component in our method is a parameter regression model that learns to regress the optimal parameters of the LtR model from the domain features. This eliminates the need for access to users’ queries and relevance judgements in a target domain, which is often unavailable in new and emerging services. In our experiments, we compared the performance of the proposed method against a domain-agnostic method, using publicly available LtR datasets including OHSUMED, MQ2007/2008, TREC Web track, and MSLR. The results showed that our method could outperform the baseline model trained on a large amount of data without considering domain differences.

Learned index is a novel index structure and changed the way we treat the traditional field of DBMS index. It views index as models and uses a learning-based approach to fit the distribution of stored data. The models input the key and output the predicted location of the target keys. To achieve higher query throughput, we propose WELGOR. We train the linear regression model with priority of the keys. To improve the mapping ability of the model, we use a hybrid model which adds the design of a simple linear model to better indexing keys. Besides, we also optimize the space allocation for gap design in node while achieving comparable throughput. Experiments show that WELGOR achieves 23% to 93% improvement in throughput compared with state-of-art methods.

Empathy in psychology arises from interactive processes between affection and cognition. Previous research on empathetic dialogue systems primarily focused on integrating emotional context with semantics or leveraging external knowledge to enhance cognitive understanding. However, these approaches have not effectively integrated external knowledge with contextual emotional information. Drawing inspiration from the psychological concept of associative empathy, our work introduces the TAKE model to trace associative empathy keywords for generating empathetic responses based on graph attention. Initially, we construct associative empathetic representations based on three types of empathetic keywords, followed by employing a graph attention mechanism to track associative empathetic traits. Finally, we integrate these representations to generate empathetic responses. Our TAKE model demonstrates superior comprehensive performance through automated machine evaluations and human assessments, highlighting its effectiveness in generating more empathetic responses.

Collaboration tools contain many intents in workplace conversation, and identifying these intents is important to increase workplace productivity. However, labelling these intents for a large collection of conversations is expensive and data availability for new intents is quite limited. The pre-trained models show outstanding results in text classification tasks and large language models have recently produced accurate models with few samples. We explored few-shot learning methods and active learning strategies for this problem. Our proposed method, “SetFit with AL” is a combination of Sentence Transformer Fine-tuning (SetFit) and active learning for intent identification. This method fine-tunes a sentence-transformer model to develop accurate models. The intent classification evaluation dataset was used to evaluate this method. The results show that our proposed method outperforms state-of-the-art large language model GPT-3.5 and is comparable to GPT-4. This method also can utilize user feedback to adapt to new data and develop personalized models. Thus, the contribution of this paper is that fine-grained intents are identified using minimal data and the model is adaptable based on user feedback.

Machine learning classifiers typically rely on the assumption of balanced training datasets, with sufficient examples per class to facilitate effective model learning. However, this assumption often fails to hold. Consider a common scenario where the positive class has only a few labelled instances compared to thousands in the negative class. This class imbalance, coupled with limited labelled data, poses a significant challenge for machine learning algorithms, especially in the ever-growing data landscape. This challenge is further amplified when dealing with short text datasets, as these inherently provide less information for computational models to leverage. While techniques like data sampling and fine-tuning pre-trained language models exist to address these limitations, our analysis reveals their inconsistencies in achieving reliable performance. We propose a novel model that leverages contrastive learning within a two-stage approach to overcome these challenges. Our proposed framework involves unsupervised Fine-Tuning of a language model to learn representation on short text followed by fine-tuning on a few labels integrated with GPT-generated text using a novel contrastive learning algorithm designed to effectively model short texts and handle class imbalance simultaneously. Our experimental results demonstrate that the proposed method significantly outperforms established baseline models.

Information spread within social networks is a complex process with broad implications. Predicting information diffusion is crucial for understanding information spread within social networks. However, previous research has primarily focused on the homogeneity characteristics of internal cascades, such as temporal and social relationships, neglecting the impact of external information propagation. Additionally, conventional methods of feature integration simply merge cascade and user embeddings, which may introduce excessive redundant information and result in the loss of valuable contextual information critical for accurate predictions. To address these limitations, we present a novel model, the Equivariant Diffusion-based Sequential Hypergraph Neural Network with Co-Attention Fusion (EDSHNN-CAF). Within its cascade feature learning module, the model proposes hypergraphs with equivariant diffusion operators to incorporate external cascade influences alongside internal features. This approach effectively captures complex high-order interconnections and accurately reflects the dynamics of information diffusion. In the feature fusion and prediction module, a co-attention mechanism is designed to seamlessly integrate cascade and user embeddings, revealing their complex interdependencies and significantly enhancing predictive capabilities. Experimental results on four real datasets showcase the promising performance of EDSHNN-CAF in predicting information diffusion, outperforming existing state-of-the-art information diffusion prediction models.

In a hyperconnected environment, medical institutions are particularly concerned with data privacy when sharing and transmitting sensitive patient information due to the risk of data breaches, where malicious actors could intercept sensitive information. A collaborative learning framework, including transfer, federated, and incremental learning, can generate efficient, secure, and scalable models while requiring less computation, maintaining patient data privacy, and ensuring an up-to-date model. This study aims to address the detection of COVID-19 using chest X-ray images through a proposed collaborative learning framework called CL3. Initially, transfer learning is employed, leveraging knowledge from a pre-trained model as the starting global model. Local models from different medical institutes are then integrated, and a new global model is constructed to adapt to any data drift observed in the local models. Additionally, incremental learning is considered, allowing continuous adaptation to new medical data without forgetting previously learned information. Experimental results demonstrate that the CL3 framework achieved a global accuracy of 89.99% when using Xception with a batch size of 16 after being trained for six federated communication rounds.

In query optimization, the precision of selectivity estimates for query predicates is foundational for selecting efficient execution plans. Spatial selectivity estimation, which assesses the count of relevant objects meeting specific spatial criteria, poses a significant challenge due to its multi-dimensional and complex nature. Moreover, it is essential that these estimation methods be both fast and minimize memory usage. In this paper, we leverage optimizer feedback to tackle the challenging task of estimating selectivity for multi-dimensional spatial predicates. We redefine spatial selectivity estimation as a regression problem and investigate the application of three types of machine learning (ML) models: neural networks, tree-based models, and instance-based models to address this challenge. We compare these ML approaches against baseline methods that rely on Minimum Bounding Rectangles (MBRs), encompassing both RTree-based and histogram-based estimations. Through extensive empirical evaluations using a real dataset, our study guides the choice of ML models in accordance with the data collected by the optimizer.

In this work, we study adversarial training in the presence of incorrectly labeled data. Specifically, the predictive performance of an adversarially trained Machine Learning (ML) model trained on clean data and when the labels of training data and adversarial examples contain erroneous labels. Such erroneous labels may arise organically from a flawed labeling process or maliciously akin to a poisoning attacker.We extensively investigate the effect of incorrect labels on model accuracy and robustness with variations to 1) when incorrect labels are applied to the adversarial training process, 2) the extent of data impacted by incorrect labels (a poisoning rate), 3) the consistency of the incorrect labels either applied randomly or with a constant mapping, 4) the model architecture used for classification, and 5) an ablation study on varying training settings of pretraining, adversarial initialization, and adversarial training strength. We further observe generalization of such behaviors over multiple datasets.An input label change to an incorrect one may occur before the model is trained in the training dataset, or during the adversarial sample curation, where annotators make mistakes labeling the sourced adversarial example. Interestingly our results indicate that this flawed adversarial training process may counter-intuitively function as data augmentation, yielding improved outcomes for the adversarial robustness of the model.

The rise of artificial intelligence and data science across industries underscores the pressing need for effective management and governance of machine learning (ML) models. Traditional approaches to ML models management often involve disparate storage systems and lack standardized methodologies for versioning, audit, and re-use. Inspired by data lake concepts, this paper develops the concept of ML Model Lake as a centralized management framework for datasets, codes, and models within organizations environments. We provide an in-depth exploration of the Model Lake concept, delineating its architectural foundations, key components, operational benefits, and practical challenges. We discuss the transformative potential of adopting a Model Lake approach, such as enhanced model lifecycle management, discovery, audit, and reusability. Furthermore, we illustrate a real-world application of Model Lake and its transformative impact on data, code and model management practices.

Multiple Traveling Salesmen Problem with Pivot Cities (PCMTSP) extends the Multiple Traveling Salesmen Problem (MTSP) by introducing pivot cities, which are allowed to be visited by multiple traveling salesmen. The difficulty of this problem lies in the repeatable visits of pivot cities and the efficient construction of legal solutions. Besides, the number of pivot cities and the visiting times of the pivot cities directly enlarge the solution space exponentially. Though a few studies have made attempts to solve this problem, there is no common benchmark to fairly evaluate the performance of the proposed methods in these studies. To fill this gap, this paper constructs a PCMTSP generator and then establishes a benchmark test suite consisting of PCMTSP instances with three different scales, namely small-scale, medium-scale, and large-scale, and three different visit types, namely intensive, sparse, and normal. Finally, this paper adapts five classical ant colony optimization (ACO) algorithms to solve the constructed PCMTSP instances. Experimental results demonstrate that the five ACOs are capable of solving PCMTSP. Hopefully, with this benchmark set, the research on PCMTSP can be boosted. In particular, the constructed benchmark set can be downloaded from https://gitee.com/bzy1999/pcmtsp .

Large Language Models (LLMs) have demonstrated remarkable efficiency in tackling various tasks based on human instructions, but studies reveal that they often struggle with tasks requiring reasoning, such as math or physics. This limitation raises questions about whether LLMs truly comprehend embedded knowledge or merely learn to replicate the token distribution without a true understanding of the content. In this paper, we delve into this problem and aim to enhance the reasoning capabilities of LLMs. First, we investigate if the model has genuine reasoning capabilities by visualizing the text generation process at the attention and representation level. Then, we formulate the reasoning process of LLMs into a causal framework, which provides a formal explanation of the problems observed in the visualization. Finally, building upon this causal framework, we propose Deconfounded Causal Adaptation (DCA), a novel parameter-efficient fine-tuning (PEFT) method to enhance the model’s reasoning capabilities by encouraging the model to extract the general problem-solving skills and apply these skills to different questions. Experiments show that our method outperforms the baseline consistently across multiple benchmarks, and with only 1.2M tunable parameters, we achieve better or comparable results to other fine-tuning methods. This demonstrates the effectiveness and efficiency of our method in improving the overall accuracy and reliability of LLMs.

As the significance of understanding the cause-and-effect relationships among variables increases in the development of modern systems and algorithms, learning causality from observational data has become a preferred and efficient approach over conducting randomized control trials. However, purely observational data could be insufficient to reconstruct the true causal graph. Consequently, many researchers tried to utilise some form of prior knowledge to improve causal discovery process. In this context, the impressive capabilities of large language models (LLMs) have emerged as a promising alternative to the costly acquisition of prior expert knowledge. In this work, we further explore the potential of using LLMs to enhance causal discovery approaches, particularly focusing on score-based methods, and we propose a general framework to utilise the capacity of not only one but multiple LLMs to augment the discovery process.

Uncensored LLMs represent a category of language models free from ethical constraints, thus prone to misuse for various malicious purposes like generating malware. However, their capabilities fall short compared to commercially available LLMs, which are censored and unsuitable for such nefarious activities. Previously, researchers could bypass censorship in LLMs to generate malicious content using Jail Breaks. However, over time and with the introduction of new security measures, such exploits have become increasingly rare. In this research, we propose a novel technique in which we combine censored and uncensored LLMs for the generation of Ransomware. The uncensored LLM will generate the initial malware, which will then be refined by the censored LLM to create a final, functional Ransomware. We have tested the developed Ransomware in the latest version of Windows OS and found it suitable for exploitation purposes. Additionally, with minor efforts, the Ransomware can be updated using LLM for code obfuscation and unnecessary functionality addition for bypassing antivirus and antimalware solutions.

With the rise in sedentary lifestyles and burdening work routines, mental health problems have been growing exponentially in recent years. While there are many online therapy agents, most of them lack human-like cognitive capabilities. The objective of this study is to develop and analyze a framework for delivering and assessing Cognitive Behavioural Therapy (CBT), utilizing the sophisticated attributes of state-of-the-art large language models (LLM). This paper presents our three key contributions: (A) Implementation and evaluation of the efficacy of utilizing LLMs, such as Llama2, GPT-3.5, and GPT-4, on CBT data. (B) Curation of real-world CBT conversations, which were gathered and annotated with the help of professionals in the mental health domain. (C) A novel approach for evaluating the performance of AI-based CBT agents or chatbots. Our technique leverages widely used assessment scales in the fields of cognitive behavioral therapy (CBT), natural language processing (NLP), and computer vision. To improve the quality of CBT conversation creation in LLMs, we use a preference-based learning method that bears resemblance to reinforcement learning with human feedback (RLHF). By incorporating the novel evaluation scale alongside three widely used metrics-BLEU, PPL, and Distinct - we were able to establish that the proposed model outperforms state-of-the-art LLMs. For instance, a BLEU score of 0.1739 was achieved compared to GPT-4’s 0.1633.

Most available data is unstructured, making it challenging to access valuable information. Automatically building Knowledge Graphs (KGs) is crucial for structuring data and making it accessible, allowing users to search for information effectively. KGs also facilitate insights, inference, and reasoning. Traditional NLP methods, such as named entity recognition and relation extraction, are key in information retrieval but face limitations, including predefined entity types and the need for supervised learning. Current research leverages large language models’ capabilities, such as zero- or few-shot learning. However, unresolved and semantically duplicated entities and relations still pose challenges, leading to inconsistent graphs and requiring extensive post-processing. Additionally, most approaches are topic-dependent. In this paper, we propose iText2KG (The code and the dataset are available at https://github.com/AuvaLab/itext2kg ), a method for incremental, topic-independent KG construction without post-processing. This plug-and-play, zero-shot method is applicable across a wide range of KG construction scenarios and comprises four modules: Documents Distiller, Incremental Entities Extractor, Incremental Relations Extractor, and Graph Integrator. Our method demonstrates superior performance compared to baseline methods across three scenarios: converting scientific papers to graphs, websites to graphs, and CVs to graphs.

The proliferation of online scams has become a pressing concern in the digital age, exacerbated by the rise of Artificial Intelligence. Malicious actors now employ sophisticated techniques to create convincing fraudulent schemes, targeting vulnerable individuals through personalized approaches on social media. This paper addresses the challenges of scam website detection by leveraging the capabilities of Large Language Models (LLMs). While other papers have focused on fine-tuning LLMs, our research investigates if readily available LLMs can be directly applied to scam website detection. This paper explores text-based and screenshot-based methods, utilizing five prominent LLMs to analyze website content. The findings indicate that existing LLMs are effective in identifying scam websites and providing rapid, expert responses for assessing website legitimacy. A novel categorization of criteria is proposed based on the LLMs’ decision-making processes. By comparing these models’ performances, this paper aims to develop a more efficient and accessible solution for identifying fraudulent websites. This work contributes to enhancing cybersecurity measures, potentially reducing online scams and increasing user trust in digital interactions.

Despite the recent broad adoption of Large Language Models (LLMs) across various domains, their potential for enriching information systems in extracting and exploring Linked Data (LD) and Resource Description Framework (RDF) triplestores has not been extensively explored. This paper examines the integration of LLMs within existing systems, emphasising the enhancement of conversational user interfaces (UIs) and their capabilities for data extraction by producing more accurate SPARQL queries without the requirement for model retraining. Typically, conversational UI models necessitate retraining with the introduction of new datasets or updates, limiting their functionality as general-purpose extraction tools. Our approach addresses this limitation by incorporating LLMs into the conversational UI workflow, significantly enhancing their ability to comprehend and process user queries effectively. By leveraging the advanced natural language understanding capabilities of LLMs, our method improves RDF entity extraction within web systems employing conventional chatbots. This integration facilitates a more nuanced and context-aware interaction model, critical for handling the complex query patterns often encountered in RDF datasets and Linked Open Data (LOD) endpoints. The evaluation of this methodology shows a marked enhancement in system expressivity and the accuracy of responses to user queries, indicating a promising direction for future research in this area. This investigation not only underscores the versatility of LLMs in enhancing existing information systems but also sets the stage for further explorations into their potential applications within more specialised domains of web information systems.

In this paper, we introduce BioLinkerAI, a neuro-symbolic approach tailored for biomedical entity linking. Traditional domain-specific entity linking approaches necessitate substantial labeled-training datasets and present challenges when adapting to each new dataset or setting. BioLinkerAI integrates symbolic methodologies with sub-symbolic models to mitigate the constraints of limited training data availability. The symbolic component utilizes a rule-based entity extraction mechanism, underpinned by an extensive set of linguistic and domain-specific rules. Concurrently, the sub-symbolic component employs a Large Language Model (LLM) to achieve precise candidate disambiguation. This mechanism enhances entity linking accuracy, especially when a single entity in a knowledge base such as UMLS aligns with multiple terms, leveraging the contextual intelligence encapsulated within the LLM’s embeddings. Empirical evaluations conducted on a range of biomedical benchmarks demonstrate the superior performance of BioLinkerAI. Notably, it surpasses existing benchmarks in entity linking accuracy (e.g., achieving an improvement from 65.4 (best baseline) to 78.5 (our model) on unseen data accuracy, representing the most stringent evaluation paradigm). Additionally, BioLinkerAI consistently performs adeptly on both structured sentences and individual keywords. To facilitate broader utilization, the source code, datasets, and a public API are available ( https://github.com/SDM-TIB/BioLinkerAI ).

Food recommendation systems help consumers make sustainable and nutritionally complete choices, promoting healthy eating habits and addressing the growing interest in food sustainability and waste reduction. Large Language Models (LLMs), such as ChatGPT, are increasingly used for food recommendations due to their natural language processing capabilities. However, providing personalised and contextually relevant suggestions remains challenging because of the lack of a robust conceptualisation of healthy and sustainable food aligned with users’ dietary and lifestyle preferences. Ontologies can address this by offering a structured and semantically rich framework for organising information. In this paper, we propose a modular ontology to enhance the contextual knowledge of LLMs, enabling them to deliver personalised, contextually relevant food recommendations. The ontology’s modules are based on competency questions derived from a research project focused on sustainable and healthy food recommendations. To evaluate the effectiveness of this approach, we conducted experiments where ChatGPT-4 answered these competency questions with and without ontology integration. The answers were then assessed in a user study. Preliminary experimental results indicate significant improvements in the quality and relevance of recommendations when the ontology is employed.

Recent developments in large language models (LLMs) have led to significant improvements in intelligent dialogue systems’ ability to handle complex inquiries. However, current LLMs still exhibit limitations in specialized domain knowledge, particularly in technical fields such as agriculture. To address this problem, we propose ShizishanGPT, an intelligent question answering system for agriculture based on the Retrieval Augmented Generation (RAG) framework and agent architecture. ShizishanGPT consists of five key modules: including a generic GPT-4 based module for answering general questions; a search engine module that compensates for the problem that the large language model’s own knowledge cannot be updated in a timely manner; an agricultural knowledge graph module for providing domain facts; a retrieval module which uses RAG to supplement domain knowledge; and an agricultural agent module, which invokes specialized models for crop phenotype prediction, gene expression analysis, and so on. We evaluated the ShizishanGPT using a dataset containing 100 agricultural questions specially designed for this study. The experimental results show that the tool significantly outperforms general LLMs as it provides more accurate and detailed answers due to its modular design and integration of different domain knowledge sources. Our source code, dataset, and model weights are publicly available at https://github.com/Zaiwen/CropGPT .

The potential for artificial intelligence (AI) in analyzing medical images is vast and promises significant future advancements. It brings opportunities for community and remote-area hospitals to be equipped with professional capabilities once exclusive to top-tier medical institutions. However, applying in-lab AI methods to real-world applications of medical imaging is challenging due to the complexity of gathering training datasets as well as the need for intricate systems and specialized devices. In this paper, we demonstrate how the web platform could benefit the application of AI methods in medical imaging based on the lightweight design, cross-platform portability, streamlined distribution and deployment of the web. Specifically, we design and implement a web-based assistant for predicting spontaneous preterm births via ultrasound images. During the development phase, we leverage crowdsourcing on the web to annotate ultrasound images and gather domain-specific features to train the AI model for predicting spontaneous preterm birth. During the deployment phase, we employ WebAR to present AI-assisted diagnostic insights for physicians. Evaluation results show that our system achieves an AUC of 0.769, nearing the diagnostic proficiency of top-tier physicians. Besides, our WebAR system exhibits only 527.2–1754.2 ms latency, enabling effective assisted diagnosis.

Accurate bathymetry using remotely sensed data is essential for various ocean-related fields such as marine resource exploration, environmental protection and offshore development. Traditional bathymetric techniques often face limitations in high-risk areas, whereas satellite-based methods offer advantages such as low cost and extensive coverage. This work aims to integrate the complementary strengths of ICESat-2 and Sentinel-2 satellites. We propose a novel dual-distance noise reduction algorithm to extract bathymetric information from ICESat-2 data, which is then integrated with Sentinel-2 optical imagery using a U-Net deep learning model. This approach enables precise inference of near-shore bathymetric distributions. Experimental results demonstrate the efficacy of the dual-distance noise reduction algorithm in accurately identifying photon signal points, achieving an average $$R^2$$ R 2 of 0.906 and an RMSE of 0.778 m in bathymetric estimation. The study provides a robust scientific basis for active-passive fusion bathymetry inversion strategies in different scenarios.

The growing capabilities of AI raise questions about their trustworthiness in healthcare, particularly due to opaque decision-making and limited data availability. This paper proposes a novel approach to address these challenges, introducing a Bayesian Monte Carlo Dropout model with kernel modelling. Our model is designed to enhance reliability on small medical datasets, a crucial barrier to the wider adoption of AI in healthcare. This model leverages existing language models for improved effectiveness and seamlessly integrates with current workflows. We demonstrate significant improvements in reliability, even with limited data, offering a promising step towards building trust in AI-driven medical predictions and unlocking its potential to improve patient care.

Individuals with visual impairments face significant challenges navigating environments, especially with tasks such as object identification and traversing unfamiliar spaces. Often, their needs are inadequately addressed, leading to applications that do not meet their specific requirements. Traditional object detection models frequently lack this demographic's accuracy, speed, and efficiency. However, recent Internet of Things (IoT) advancements offer promising solutions, providing real-time guidance and alerts about potential hazards through IoT-enabled navigation apps and smart city infrastructure. This paper presents an extension of our MoSIoT framework, incorporating the YOLOv8 convolutional neural network for precise object detection and a specialized decision layer to improve environmental understanding. Additionally, advanced distance measurement techniques are incorporated to provide crucial information on object proximity. Our model demonstrates increased efficiency and adaptability across diverse environments using transfer learning and robust regularization techniques. Systematic evaluation indicates significant improvements in object detection accuracy, measured by mean Average Precision at 50% Intersection over Union (mAP50) from 0.44411 to 0.51809 and mAP50-95 from 0.24936 to 0.29586 for visually impaired individuals, ensuring reliable real-time feedback for safe navigation. These enhancements significantly improve the MoSIoT framework, thereby greatly enhancing accessibility, safety, independence, and mobility for users with visual impairments.

Consumer behavior and habits play a crucial role in household energy consumption patterns. Influencing user behaviors towards sustainable electricity consumption practices consists an open challenge. To address this issue, the Internet of Behaviors (IoB) has emerged as a new paradigm that combines real-time data coming from Internet of Things (IoT) devices with information gathered from behavioral science and data analytics to influence people’s behavior. In the energy sector, IoB systems can build highly personalized models that allow smart home devices to encourage users to adopt more sustainable energy behaviours. This paper proposes a hybrid forecasting approach combining Long Short-Term Memory (LSTM) with a Transformer model to accurately predict electricity consumption in individual households. The proposed approach is integrated into a new IoB system designed to provide personalized and timely alerts that encourage energy-efficient practices, reducing thus costs and energy waste. The performance of this approach are compared with several baseline models using a real dataset related to household electricity consumption. The results show that the hybrid approach achieved lower Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) across various electricity patterns, demonstrating a better ability to anticipate future energy demands. The improved forecasting accuracy enables the IoB system to generate more precise and timely alerts, potentially leading to more effective user behaviors influence and significant energy savings.

This study investigates the application of Machine Learning to enhance customer service efficiency within the Holy Makkah Municipality, focusing on the Tasahalat service. By analyzing historical data, we developed a model designed to automatically classify customer inquiries and deliver prompt, automated responses. Several Machine Learning algorithms, including Support Vector Machines, Naive Bayes, and Random Forest, were employed in the process. The research emphasizes the superior performance of the Random Forest algorithm, which achieved an impressive accuracy of 93% and a robust F1 score. By utilizing Natural Language Processing techniques to handle Arabic-language data, the model substantially improves response times and service quality, leading to enhanced customer satisfaction and more efficient municipal operations. This research underscores the potential of integrating advanced Machine Learning technologies into public services, setting the stage for future innovations in the sector.

Attitudes towards artificial intelligence (AI) are influenced by individual intentions to use it and concerns about its implications, which can vary across different age groups and genders, highlighting the need for more nuanced design and communication strategies. This study explores how gender and age influence attitudes towards AI by examining intentions and concerns from a cross-cultural perspective. Using a sample of 562 participants from the UK (281) and the Arab Gulf Cooperation Council (GCC) (281), the research investigates demographic and cultural differences in AI use intentions and concerns. The study finds that gender and age significantly influence AI acceptance in the UK, whereas these factors have a less pronounced impact in the Arab context. The results highlight that women generally express more ethical and privacy concerns about AI than men, and older adults show more apprehension towards AI acceptance than younger individuals. By showing that cultural nuances play a role in shaping these attitudes, we also show the need for tailored strategies to address demographic-specific concerns to reduce fear towards AI and, at the same time, avoid over-acceptance. The combined influence of age and gender can enhance the effectiveness of AI strategies, emphasizing the importance of considering personal and cultural factors in design and policy-making, e.g., in aiding trust calibration and informed adoption of AI.

The manufacturing of semiconductor wafers is a complex process that is prone to defects. This paper introduces DefectClassifierX, an automated defects pattern classification system that uses a convolutional neural network model based on the ResNet-152 architecture. The main objective of the approach is to improve defect classification in semiconductor manufacturing, which precisely classifies wafer defect patterns. To validate the approach taken, several experiments were performed using a dataset named “WM-300K+ wafer map [single and mixed]”. For single and mixed, the dataset contained 36 different defect patterns. The performance measures extracted from the developed model demonstrated truly outstanding accuracy with precision, recall, and f1 score of 0.9. These results reflect an exceptional average classification accuracy of 97.74% in both single and mixed defect types and outperforms previous studies in wafer defect pattern classification. It is expected to offer a giant leap in increasing the efficiency and effectiveness of wafer defect analysis in semiconductor manufacturing.

Recommending safe routes has become a fundamental necessity due to the increasing number of accidents. This heavily relies on how to evaluate the road severity. Existing recommendation systems are based on user feedback, either positive or negative. However, this type of evaluation overlooks many aspects of road severity like accident history and volunteered geographic information on road conditions. To fill this gap, we elaborate a comprehensive and predictive road risk analysis, relying on objective and subjective data. To recommend safe roads, we propose an algorithm called Safe-Path based on accurate and reliable risk values. To validate our approach, we conduct some experiments to benchmark various machine and deep learning models.

Visualizing the correlations between structured medical data in the form of Electronic health records (EHRs) is of major importance for effective and efficient medical data analysis and decision-making. This work describes an unsupervised semi-structured and feature-based tool for dynamic EHR data visualization called “mirrored dendrograms”. It accepts as input semi-structured EHRs, and allows the user to select the target features to be visualized and mapped against each other, and their relative weights on the visualization process. It then invokes a hierarchical clustering process to cluster the data following the user-chosen features, and produces a dendrogram structure for each combination of target features. The dendrograms are mirrored against each other by mapping their nodes using the transportation optimization problem, allowing the user to dynamically zoom-in and out of the mapping at different granularity levels. We have evaluated our solution using a sample dataset of 114 EHRs of patients who suffer from migraine disorder. A group of 20 testers participated in the evaluations to assess the tool compared with existing solutions. Results showcase the tool’s performance.

Stroke is a life-threatening medical condition caused by an inadequate blood supply to the brain. According to the World Health Organization (WHO), stroke is a leading cause of death and disability worldwide. After a stroke, the affected brain areas fail to function normally, making early detection of warning signs crucial for effective treatment and reducing disease severity. Various Machine Learning (ML) and Deep Learning (DL) models have been developed to predict stroke occurrence. This research highlights the effectiveness of Federated Learning (FL), a decentralized training approach that bolsters privacy while preserving model performance. Our models outperform traditional ML and DL methods, achieving an accuracy of 98%. Evaluations using metrics such as accuracy, precision, recall, and F1 score confirm the robustness and generalizability of our approach.