Recommendation Systems

The purpose of recommender systems (RS) is to facilitate user collaboration and communication on the platform. Nevertheless, there is limited knowledge regarding the extent of this relationship and the techniques by which RS could promote persistent user engagement with the platform. In order to fill this void, the present study investigates the role of RS in transforming users' short-term engagement with the RS into long-lasting involvement with the platform. We present a theoretical framework by reviewing relevant literature in the domains of RS and user engagement to probe these issues. We provide open challenges in this field along with metrics in the present study.

Recommendation systems play an invisible but powerful role in our daily lives, helping us discover music, movies, courses, and even products we might enjoy. Yet two stubborn challenges remain. The first is the cold-start problem, where new users receive generic suggestions because the system knows nothing about them. The second is the lack of explicit feedback, since most people rarely bother to leave ratings or reviews. This conceptual paper explores a potential solution: adding a moodbased conversational chatbot as a feedback layer for recommendation systems. Instead of asking users to rate items with stars or numbers, the chatbot engages in simple, natural dialogue: "How are you feeling today?" or "What are you in the mood for?" This conversational design serves two purposes. It helps reduce cold-start limitations by gathering useful mood input right away, and it encourages richer, more sincere feedback without forcing users into mechanical rating tasks. The result is a system that feels more personal, more engaging, and more aligned with the way people actually make choices. By blending mood signals with existing recommendation algorithms, such systems could open new possibilities in entertainment, e-commerce, education, and even healthcare.

Search is a prominent channel for discovering products on an e-commerce platform. Ranking products retrieved from search becomes crucial to address customer's need and optimize for business metrics. While learning to Rank (LETOR) models have been extensively studied and have demonstrated efficacy in the context of web search; it is a relatively new research area to be explored in the e-commerce. In this paper, we present a framework for building LETOR model for an e-commerce platform. We analyze user queries and propose a mechanism to segment queries between broad and narrow based on user's intent. We discuss different types of features - query, product and query-product and discuss challenges in using them. We show that sparsity in product features can be tackled through a denoising auto-encoder while skip-gram based word embeddings help solve the query-product sparsity issues. We also present various target metrics that can be employed for evaluating search results and com...

This article explores the transformative role of libraries in the digital era, with a focus on how AIMAN COLLEGE OF ARTS AND SCIENCES FOR WOMEN is leveraging smart technologies to enhance user experiences. The integration of digital resources, RFID technology, AI-powered recommendations, and sustainable practices is reshaping the library into a hub of innovation and learning. While addressing challenges like data security and initial setup costs, the article highlights the library's commitment to providing a tech-savvy, eco-friendly, and community-centric environment. The insights presented demonstrate the institution's role in empowering students and setting benchmarks for future-ready libraries.

Online trading is essential for day-today life in the rapid growth of internet and e-commerce. Customers are intended through the online advertisements and necessity to take decision making for the product to purchase among similar ones. Collaborative Filtering Techniques are used in Recommendation System to forecasting the rating of product for the customer's right choice. In this paper, Samsung Mobile Review Rating (SMRR) dataset are considered for study. The BFGS algorithm runs for different latent factors with bias values and shows that RMSE converged in fewer iterations and hence forecasting the unknown ratings.

User-based and item-based collaborative filtering techniques are among most explored strategies of making products’ recommendations to Users on online shopping platforms. However, a notable weakness of the collaborative filtering techniques is the cold start problem. Which include cold user problem, cold item problem and cold system problem – i.e., the failure of collaborative filtering to make recommendation of products to a new user, failure of an item to be recommended, or combination of the two respectively.  Literature investigation has shown that cold user problem could be effectively addressed using technique of personalized questionnaire. Unfortunately, where the products’ database is too large (as in Amazon.com), results obtained from personalized questionnaire technique could contain some user preference uncertainties. This paper presents technique of improving personalized questionnaire with uncertainty reduction technique. In addition, the paper presents classification o...

Recommender systems have garnered significant attention from researchers due to their potential for delivering personalized recommendations in light of the vast amount of information available online. These systems have found applications in various domains, including financial services, movies, and research articles. Their implementation in the tourism industry is particularly promising. Travelers often face the daunting task of selecting the right tourist attractions from a plethora of options, which can consume considerable time and energy. By leveraging personalized recommendation technologies, it is possible to provide highly accurate travel suggestions tailored to individual preferences. This study proposes the development of a customized recommendation system (RS) aimed at assisting travelers in the Qassim region of the Kingdom of Saudi Arabia. By using this region as a case study, the proposed RS consists of two main modules: a user registration and login module and a recommendation technique and tag module. The system will capture users' interests and prompt them to select from various options, subsequently presenting them with tailored recommendations based on their preferences. This approach aims to enhance the travel experience by offering relevant suggestions that align with the interests of each traveler.

Travel planning has transitioned from traditional guidebooks and travel agents to online
booking platforms and, more recently, to AI-powered intelligent assistants. This survey
explores WanderSync, a Generative AI-based travel planning platform that leverages LLMs
(Large Language Models) and real-time travel APIs to deliver personalized itineraries. The
study compares WanderSync with existing travel platforms such as TripIt, Google Travel, and
Kayak, identifying the strengths, weaknesses, and gaps. The paper concludes with insights into
the future of AI-driven travel planning, emphasizing personalization, collaboration, and
visualization

Collaborative filtering systems rely heavily on matrix factorization techniques, which often face scalability issues when handling large datasets. This paper presents an efficient parallel algorithm that leverages distributed computing to perform largescale matrix factorization. Experimental results show that our algorithm significantly reduces computation time while maintaining high accuracy. The approach has practical implications for recommendation systems, particularly in ecommerce and social media platforms.

This research demonstrates that personality traits, specifically those measured by the Big Five model, can significantly improve recommendation systems when combined with an item-based K-Nearest Neighbors (KNN) model. The proposed Hybrid KNN approach successfully addresses the cold-start problem by leveraging psychological data rather than relying solely on past user interactions. A qualitative evaluation, comparing the model's output for different user personas against a standard popularity-based baseline, demonstrated a significant improvement in recommendation relevance and personalization.

The results suggest that personality-aware recommendation systems can enhance user satisfaction and engagement across multiple domains, from entertainment platforms to e-commerce, online learning, and recruitment. By integrating stable psychological traits with a content-based similarity model, such hybrid approaches represent a promising direction for the future of intelligent, human-centered recommendation systems.

Recommender systems (RSs) on platforms like Netflix and Spotify personalize user experiences but also raise concerns about their impact on aesthetic welfare. This paper evaluates two important arguments against RS-driven platforms. The satisfaction argument claims that RSs harm aesthetic welfare by steering users toward profitable content that is less satisfying because it is less aligned with their personal tastes. I argue that while RS-driven platforms may exhibit a bias toward promoting profitable content, they are unlikely to do so aggressively enough to significantly harm user satisfaction, as such a strategy carries inherent risks and offers uncertain profit advantages compared to catering to users' existing preferences. Moreover, even if RSs consistently steer engagement toward profitable content that diverges from a user's personal taste, they would likely need to reshape user preferences in the process to maintain engagement. In such cases, aesthetic enjoyment would remain largely intact, as the altered taste would align with the promoted content. The autonomy argument suggests that RSs push users into patterns of aesthetic engagement that conflict with their core aesthetic valuessuch as those who prefer niche or non-mainstream contentdue to apparent RS biases favoring popular and mainstream material. I contend that such authenticity concerns are not particularly worrisome when the core value in question lacks independent normative significance. For example, there is no strong reason to claim that aesthetic users should or need to engage with non-mainstream content. Ultimately, this paper challenges claims about the harms of RS-driven platforms, arguing that their impact on aesthetic well-being is less significant than often assumed.

Distribution companies are trying to reduce the high real loss and manage the poor voltage profile, so power loss minimization and voltage profile improvement are the main and important tasks to be faced by electrical engineers in the distribution system (DS). To achieve this goal requires knowledge for analyzing a radial distribution system (RDS) involves conducting power flow analysis, voltage regulation assessment, sensitivity analysis, and validation and verification processes to obtain reliable numerical results. Power flow analysis (PFA) determines the steadystate operating conditions by solving nonlinear equations representing power injection and consumption at each bus. The results include active and reactive power flows, indicating power demands and losses, as well as voltage magnitudes and phase angles, aiding in voltage stability assessment. (RDS) has been applied on operated example the standard IEEE 33-bus system and the results compared with the other popular algorithms. Overall, these analyses offer valuable insights for optimizing the power distribution system, identifying areas of concern, and improving system performance.

This paper investigates the colonial architecture of Shamsuddin Medical College Hostel, Sylhet, Bangladesh for its historical context, present scenario and a possible future reuse. The research analyses the transformation of the edifice from its original function to the present use by benchmarking it against other Assam-type colonial buildings in terms of architectural elements and materials used. Physical condition, deterioration and modifications of building over time are evaluated through a combination of field surveys, interviews and archival research. Neglect, institutional limitations, and environmental impacts are examined as some of the major obstacles to conservation. The research is also looking into whether the facility could be used as an adaptive reuse space, to preserve heritage values while answering contemporary needs. Incorporating a series of stakeholder perspectives such as architects, local authorities and community, the study proposes pragmatic conservation strategies. These results point to potential for the building, troubled as it is, to be reimagined into an operational space that respects its past.

Recently, attributed community search, a related but different problem to community detection and graph clustering, has been widely studied in the literature. Compared with the community detection that finds all existing static communities from a graph, the attributed community search (ACS) is more challenging since it aims to find dynamic communities with both cohesive structures and homogeneous node attributes given arbitrary queries. To solve the ACS problem, the most popular paradigm is to simplify the problem as two sub-problems, that is, structural matching and attribute filtering and deal with them separately. However, in real-world graphs, the community structure and the node attributes are actually correlated to each other. In this vein, current studies cannot capture these correlations which are vital for the ACS problem. In this paper, we propose Query-Driven Graph Convolutional Networks (QD-GCN), an end-to-end framework that unifies the community structure as well as nod...

This article explores the socio-technical imaginaries of Brazilian tech workers involved in artificial intelligence (AI) development, focusing on their views on Brazil's role in the global AI market and the societal implications of AI in Latin America. Based on semi-structured interviews, the study addresses key challenges such as algorithmic bias, diversity and inclusion (DEI) in AI, and AI governance. The findings position Brazil's complex role as a hub of AI development linked to extractive industries, challenging reductive narratives of the country as a mere data colony, while acknowledging its subordinate incorporation into the global market. It also exposes the disconnect between critical debates such as 'decolonial AI' and tech workers' imaginaries, while highlighting the potential for workers' collective agency to influence society's regulation of AI.

Customer churn is a critical concern for any business that relies on
recurring clients. Simultaneously, personalized product
recommendations have become a standard feature for improving sales
and customer satisfaction in e-commerce. This project combines
customer churn analysis with product recommendation techniques,
integrated directly into the Odoo ERP platform. The system provides
predictive insights about customer behavior while offering relevant
product suggestions using hybrid recommender algorithms. The
implementation leverages Python scripts/custom modules integrated into Odoo v16 running on Ubuntu 24. We achieved functional modules for churn prediction and recommendation based on sales data, user behavior, and historical interactions.

Machine Learning (ML) is recognized as a foundational component within the broader field of Artificial Intelligence (AI), representing a transformative technology that enables systems to autonomously learn from vast amounts of data and make adaptive decisions. This study aims to articulate the conceptual framework of machine learning, trace its historical development, and systematically analyze its contemporary application domains. The three principal learning paradigms-supervised, unsupervised, and semi-supervised learning-are comparatively examined from both theoretical and practical standpoints. Employing a descriptive methodological approach, widely adopted algorithms such as decision trees, support vector machines (SVM), and k-nearest neighbors (k-NN) are presented and critically assessed in terms of their strengths, limitations, and use-case scenarios. Particular attention is devoted to applications in automatic speech recognition, recommendation systems, and fraud detection. Furthermore, prominent ML platforms including TensorFlow, Scikit-learn, and IBM Watson are evaluated based on their performance, scalability, and user accessibility. Real-world project examples are incorporated to demonstrate the practical integration and sector-specific adaptation of machine learning technologies. The analysis underscores that machine learning is not only essential for the technology sector but is increasingly becoming indispensable in critical domains such as education, healthcare, finance, and transportation. A significant rise in the demand for qualified professionals in this interdisciplinary field is projected in the coming years.

ÖZET: Makine öğrenmesi, yapay zekâ alanının temel taşlarından biri olarak, veriden otomatik öğrenme kabiliyeti sayesinde çeşitli sektörlerde devrim yaratmaktadır. Bu çalışmada, makine öğrenmesinin temel kavramsal çerçevesi, tarihçesi ve günümüzdeki uygulama alanları ele alınmıştır. Denetimli, denetimsiz ve yarı denetimli öğrenme türleri teorik ve pratik yönleriyle karşılaştırılmıştır.

Nowadays, most of the information science inquires about a spotlight on affiliation rule to decide explicit examples and rules from huge information. Affiliation guideline is worked by basic information duration device, for example, WEKA which incorporates arrangement, bunching, affiliation runs, etc. When all is said in done, the affiliation standard could be pertinent for enormous E-Learning datasets like they were pertinent towards the formless arrangement. Within this exam, the uses for affiliating rule after gathering under straight information for a considerable length of time from LMS have been explored. Intriguing quality measurements and other applicable understudy inclinations are caught in the wake of thinking about a few affiliation rule calculations. Some imagined introductions of rules and their applicable outcomes will be exhibited in wording reasonableness in eLearning conditions.

The personalization of virtual fashion recommendations remains hindered by limited integration of chromatic and anthropometric factors, especially skin tone compatibility. This study addresses a critical research gap by proposing a voice-enabled 3D fashion recommendation system that incorporates deep learning-based skin tone classification and adaptive garment suggestion. The primary objective is to enhance aesthetic compatibility and user satisfaction through real-time, personalized recommendations. Utilizing a custom-designed Deep Convolutional Neural Network (DCNN) and reinforcement learning algorithms, the system classifies user skin tones with 89.14% accuracy and adapts recommendations based on user feedback, reducing outfit resets by 54%. A curated dataset encompassing five skin tone categories and a multi-stage image preprocessing pipeline ensures inclusive and robust performance. The results demonstrate significant improvements in recommendation relevance and user engagement, with 88% satisfaction and 93.7% dominant tone detection accuracy. These findings underscore the system's potential to set new benchmarks in personalized fashion retail while promoting inclusivity and sustainable consumer practices.

We developed a concept including a set of tools for self-management for patients suffering from axial spondyloarthritis (SpA). This concept involves patient-recorded outcome measures, both subjective assessment and clinical measurements, that are used to present recommendations. We report from experiences made while implementing a proof of this concept. Besides giving the patient a self-management tool, our work also improved the methodology for clinical measurements. Keywords—axial spondyloarthritis; self-mangement; healthcare; self-assessment; evidence-based; mobile applications. I . I N T R O D U C T I O N For a variety of chronic diseases, patients managing the condition themselves (self-management) can result in reduced costs in the health care sector and an improved clinical outcome [1]. Self-management encompasses methods where the patient participates in managing their disease through education and changes of behaviour and lifestyle [2] [3]. In evidence-based self-management...

This article represents a BERT-based anime recommendation system which delivers personalized anime recommendations. The model learns from 1.77 million filtered users alongside 148 million ratings. The model extends the publicly available implemention of BERT4Rec paper [4] by SungMin Cho and Jae-Won Chung [1] by removing positional encoding due to temporal signal absence and adding genre embeddings to enhance item representations. The model delivers performance results close to the original BERT4Rec architecture even though it lacks temporal information which proves the effectiveness of the proposed modifications. The model delivers strong performance across standard Top-K recommendation metrics when tested on large-scale data which proves its effectiveness for domain-specific recommendation tasks.

The use of a portfolio curriculum approach, when teaching a university introductory statistics and probability course to engineering students, is developed and evaluated. The portfolio curriculum approach, so called, as the students need to keep extensive records both as hard copies and digitally of reading materials, interactions with faculty, interactions with other students and work they have completed on their own, is designed to encourage active learning, mainly in the areas of cooperation and collaboration. In order to investigate the effectiveness of the portfolio curriculum, a controlled experiment applying a pre-test-post-test control group design is conducted. Two tests are conducted, one before the commencement of the course (pre-test) and one after the completion of the course (post-test). The effectiveness is evaluated by comparing within-subject post-test and pre-test scores and by comparing the scores between subjects in the experimental group, i.e., those who learned using the portfolio curriculum approach and subjects in the control group, i.e., those who learned using a traditional method of teaching. In addition to analysis of the controlled experiment, a Survey of Attitudes Toward Statistics (SATS) was completed on the first and last day of the semester by the participants so as to give a measure of student confidence, understanding, liking, and difficulty of the portfolio curriculum approach as opposed to using a traditional method of teaching and learning. The findings of these investigations are reported and discussed, as are the merits and problems encountered regarding the methodology and student attitudes regarding the portfolio curriculum approach.

This paper presents the development of a movie recommendation system based on machine learning techniques, with the goal of achieving a root mean squared error (RMSE) below 0.86490 in line with a Harvard University challenge. Using the MovieLens dataset, the system applies data preprocessing, noise reduction, and matrix factorization to capture user-to-user and movie-to-movie effects. Regularization and cross-validation are employed to prevent overfitting, particularly addressing sparse user ratings and infrequent movie ratings. The final model achieves an RMSE of 0.86482 on the untouched validation set, demonstrating the system’s predictive accuracy. This work highlights the effectiveness and limitations of collaborative filtering and matrix factorization in generating personalized movie recommendations.

Web-based recommendation strategy implemented in a cadastre information system is presented in the paper. This method forms the list of page profiles recommended to a given user. The idea of page recommendation uses the concept of a page profile for system pages containing forms with search criteria. The calculation of rank values for page profiles is based on the usage frequency and the significance weights of profile elements determined by users.

Due to the large volume of data produced by e-commerceor e-business websites, there is a need to discover business intelligence for recommending top-N items.Several methods have been employed to collaborative filtering recommender systems and still have shortcomings. Due to the success of deep learning in various areas, itcan also be used to improve the performance of recommender systems. The aim of this research work is to further promote research using deep learning techniques for learning user preferences from multiple domains and enhance collaborative filtering recommender systems.Several experiments based on two benchmark datasets (MovieLens 1M,MovieLens 10M and Jester) are carried out to verify the effectiveness of the proposed method, and the result shows that our model outperforms traditional methods that used feed-forward neural networks by a significant margin and performs very comparably with state-of-the-art methods onreal-word datasets.

Collaborative filtering is a common aspect recently used in e-commerce to increase sales and overcome information overload. One significant limitation in collaborative filtering is data sparseness. Several studies have proposed alleviating this issue by utilizing extra information such as users' reviews. However, the researchers have been concerned with using entire reviews irrespective of the users' ratings. This requires extra processing time and might perplex the recommendation decision. In this study, after analyzing the users' ratings and reviews, it was noted that when the rating values are 4 or 5, most of the reviews accompanying these ratings are positive. Otherwise, when the ratings are 1 or 2, the reviews are negative. The rating value 3, however, is a confusing one. It, consequently, might mislead users' preferences. Thus, two points are highlighted. First, analyzing and classifying the entire users' reviews is time-consuming. Second, directly considering the rating value 3 as a like or dislike decreases the model's accuracy. Hence, this method utilizes the users' reviews only in case the rating value is 3. Three scenarios were formulated and fitted to this case. The first scenario considers the rating value 3 as a like. The second scenario regards the same value as a dislike. The third one classifies the users' reviews into positive or negative regarding their content. Additionally, a probabilistic method was applied to compute the recommendation list. The proposed method was implemented on two standard datasets. Eventually, the results of the third scenario outperformed the other two in terms of accuracy.

Public libraries have become one of the biggest powerhouses for the production of knowledge and information by having various sources and references that their effective and efficient management can have a significant impact on the development of users' information literacy. The aim of the current research was to investigate the role of knowledge management in improving the information literacy of public library users and providing a suitable model. The research method was a combined method using the metacomposite technique, content analysis and Delphi to identify and categorize the components of management-knowledge and information literacy, and then the survey method. The statistical population included 284 librarians and 45582 users, 181 librarians and 497 users were selected by multi-stage cluster sampling. The data collection tool included two researchermade questionnaires for knowledge management and information literacy, derived from the theoretical and experimental background of the research,

This paper examines various illumination invariant techniques and identifies the one which works well with principle component analysis for human face recognition. Experimental results show that by applying the technique called Gradientfaces at the pre-processing stage which computes the orientation of the image gradients in each pixel of the face images and uses the computed face representation as an illumination invariant version of the input image, it can greatly improve the recognition rates. From a low recognition rate of 6.25% up to 60.75% testing on the Asian face database which has images with various illumination.

Numerous attempts have been made to devise systems that make the work of a visually impaired person easier. These researches have focused on a number of issues such as path finding, obstruction detection, face recognition, sign recognition, to name a few. The aim of this paper is to outline a system, based on Microsoft Kinect that will provide some of these features in a unified manner. The system is based on a number of open source tools such as: OpenCV, OpenKinect, Tesseract and Espeak. Features that have been incorporated building this aiding tool are object detection and recognition, face detection and recognition, object location determination, optical character recognition and audio feedback. One of the key components of this research is to ensure considerable amount of accuracy and at the same time be extremely efficient in terms of hardware resource required. Since the system is an aggregation of multiple components, their accuracies are measured independently from online and offline point of view (where applicable). The best component (face recognition) showed an accuracy of 90%. The weakest component (text recognition) yielded an accuracy of 65%. The proposed system is able to detect and recognize face/text/chair in a frame within 2.25 seconds.

Research in news recommendation, like other areas of recommender systems, focuses heavily on the utility of recommendations for the consumers of news, that is, the users of apps, sites, or feeds in which personalized news is delivered. This emphasis leaves out an important class of stakeholders, namely professional journalists themselves. We argue that these omissions diminish the value of journalists' work and perspectives and leads to lower quality news delivery. As a possible solution, we show that a multistakeholder approach to news recommendation offers an avenue for restoring the journalistic voice to news recommendation. In this paper, we argue that news recommender systems should be designed with journalists as first-class users and explore the research implications of such a commitment. We include preliminary results from an interview study exploring the deficiencies journalists find in current news recommendation platforms and what might be the benefits of greater governance over them.

Recommender systems play an important role in many scenarios where users are overwhelmed with too many choices to make. In this context, Collaborative Filtering (CF) arises by providing a simple and widely used approach for personalized recommendation. Memory-based CF algorithms mostly rely on similarities between pairs of users or items, which are posteriorly employed in classifiers like k-Nearest Neighbor (kNN) to generalize for unknown ratings. A major issue regarding this approach is to build the similarity matrix. Depending on the dimensionality of the rating matrix, the similarity computations may become computationally intractable. To overcome this issue, we propose to represent users by their distances to preselected users, namely landmarks. This procedure allows to drastically reduce the computational cost associated with the similarity matrix. We evaluated our proposal on two distinct distinguishing databases, and the results showed our method has consistently and considerably outperformed eight CF algorithms (including both memory-based and model-based) in terms of computational performance.

This research introduces an intelligent travel planning approach that creates personalized itineraries using static location data. By combining category-based filtering with a modified A* algorithm, the system tailors routes based on user preferences such as travel type, interest categories, and time availability. It generates optimized, multi-day plans that minimize travel time while maximizing relevance, all without relying on real-time connectivity. This makes the solution ideal for offline, user-centric travel experiences.

Personal debt in the United States has reached critical levels, creating widespread economic strain and limiting opportunities for financial mobility. This article presents a comprehensive AI-driven ecosystem designed to proactively identify financially distressed individuals and connect them with personalized debt relief resources through advanced machine learning and realtime data engineering. The framework integrates multiple AI models, including risk classification algorithms, propensity scoring systems, natural language processing for intent detection, and recommender systems for tailored program matching. Built on a scalable infrastructure utilizing Apache Kafka and Spark for stream processing, the system aggregates behavioral signals from diverse sources while maintaining privacy compliance. The multichannel engagement strategy encompasses on-site personalization, targeted digital remarketing, connected television campaigns, and direct communication channels to ensure inclusive reach across demographics. Through a structured five-phase journey from crisis identification to financial empowerment, the framework demonstrates significant improvements in program participation rates, debt reduction outcomes, credit score rehabilitation, and reduction in financial anxiety. The system's architecture enables nationwide deployment across varied populations and regions, offering a transformative solution to address economic inequality and promote sustainable financial recovery. This technological innovation represents a convergence of artificial intelligence, behavioral science, and social impact, providing a blueprint for large-scale financial wellness initiatives that serve the public good while advancing the field of applied AI in economic contexts.

World Health Organization publications disclose that breast cancer is one of the most common diseases amongst the women, and that it has a high death rate. Its prevalence is growing in developing nations, where the vast majority of cases are discovered late in the disease's development. Preventing this malignancy may be accomplished with safeguards and frequent examinations. Aside from that, early detection of the illness may be beneficial in terms of battling the condition. This deep learning algorithm is a strong approach for identifying features about the breast mass that are often not visible to the human eye through deep learning analysis of the mammogram. The algorithm will then be evaluated through various metrics to validate on its accuracy. Deep learning is a technology that will be used in this project. The algorithm intends to help cancer firm research centres in Kenya and globally in quicker grouping, identification and aid doctors in quicker diagnosis and research.

The Human Resources (HR) department faces significant challenges in employee retention. Traditional methods, such as performance evaluations and career development using regression, association, and clustering, have been widely used and have yielded positive results. However, these approaches are limited in predicting changes in employee behaviour and capturing complex relationships between variables. In this study, we leverage AI advancements to enhance predictive analysis by utilizing deep learning's ability to identify patterns and complex relationships while continuously adapting to employee behavior changes. Specifically, we integrate Graph Convolutional Network (GCN) deep learning-based and bipartite graph-based approaches to construct a robust link prediction model. The bipartite employee-training network serves as input to the GCN, where each convolutional layer aggregates information from neighboring nodes, leveraging observed link information at each hidden layer. During the evaluation phase, the model iteratively aggregates information until an optimal state is reached, uncovering hidden relationship patterns that facilitate employee skill development. Empirical results on a benchmark dataset demonstrate significant performance improvements, with precision, recall, and AUC metrics exceeding 80%, highlighting the model's effectiveness in enhancing employee retention.

This thesis investigates the application of artificial intelligence (AI) to enhance children's book discovery within Iceland's unique linguistic market. Addressing the challenge of maintaining reading engagement amidst digital distractions, this study developed and evaluated an AI-powered web application. The research demonstrates the effectiveness of large language models and semantic search in providing personalized book recommendations.

The system utilizes a large language model in a conversational interface to engage young readers. This approach bypasses the limitations of adult-centric platforms reliant on extensive interaction data. A Retrieval Augmented Generation (RAG) approach employs vector embeddings in Pinecone and integrates with a book catalog database. This ensures recommendations are grounded in available Icelandic children's literature. The system considers reading level, content appropriateness, and individual interests when making suggestions.

Evaluation, conducted through partnerships with Borgarbókasafnið (Reykjavik City Library) and school librarians (Seljaskóli, Flataskóli, Kársnesskóli), involved direct observation and user surveys. Testing with children across various age groups revealed high satisfaction with both the recommendation process and the resulting suggestions. Key areas identified for future improvement include finer age group distinctions, processing speed optimization, and broader book catalog integration.

This project presents a practical model for applying modern AI in educational settings for children. The findings enhance understanding of AI's potential to support literacy while respecting privacy and developmental needs. The methodology offers valuable guidance for developing similar tools in comparable linguistic environments, illustrating how AI can effectively augment, rather than replace, human expertise in education.

This paper presents a novel graph neural network (GNN)-based model for personalized learning with advanced graph neural networks, incorporating both graph convolutional networks (GCN) and graph attention Networks (GAT). Our model leverages GCN, which consists of multiple layers embedding deep learning models, to aggregate data from neighboring nodes and capture the intricate relationships between students and courses. The GAT layers refine these embeddings by dynamically assigning importance weights to connections, prioritizing relationships critical for personalized course recommendations. This dual-layered approach enables the model to account for both global structural patterns and locally significant interactions within the student-course graph. We evaluated the performance of our model using the open university learning analytics dataset (OULAD), a rich dataset encompassing student demographic information, interaction data, and course performance metrics. Experimental results achieved 78.9% F1-score, 78.3% precision, and 76.2% recall in personalized recommendations, outperforming single-layer GCN implementations by approximately 15 percentage points. These results demonstrate the model's ability to handle complex, dynamic relationships in educational data, ensuring more relevant and effective recommendations. By addressing key challenges in recommendation systems, such as the need for dynamic weighting of relationships and the handling of sparsity in educational data, our study underscores the transformative potential of GNNs in advancing personalized education. This work sets the stage for further exploration of GNN applications in e-learning, paving the way for adaptive and intelligent course recommendation systems that align with individual learning needs and preferences.

Open educational resources (OER) are valuable assets in learning and teaching. They ensure cost-effectiveness and customizability, and contribute to global collaboration in the education realm. Hence, education stakeholders face a critical challenge in locating suitable OERs that meet their needs and pedagogical objectives. To cater to this issue, researchers propose diverse digital solutions, one being recommender systems (RS). While a wide array of the suggested tools focused on learners only, this study introduces MOERID, an artificial intelligence (AI)-powered OER recommender Chatbot aimed at the teaching community. It aspires to facilitate resource discoverability, allowing instructors to save time and energy to concentrate on other pedagogical duties. MOERID engages natural language processing (NLP) and recommendation filtering techniques to locate and deliver relevant OERs to instructors. The study describes the implementation of MOERID, highlights its efficacy and provides actionable recommendations for future research by outlining the research gaps.

Цель исследования. Оценка диагностической значимости, информативности и безопасности ультразвукового исследования (УЗИ) с контрастным усилением препаратом Соновью (SonoVue) в диагностике болезни Крона (БК) и язвенного колита (ЯК). Материалы и методы. Проведено пилотное проспективное исследование, в котором приняли участие 15 пациентов с воспалительными заболеваниями кишечника (ВЗК). Все пациенты дали письменное согласие на участие в исследовании и обработку персональных данных. В исследование включали взрослых пациентов с установленным диагнозом ЯК и БК, с доказанной клинической активностью заболевания. Активность оценивалась на основании клинико-лабораторных данных по шкале Беста (CDAI >150) для пациентов с БК и по шкале Trulove-Witts (2-3-я стадия) и индексу Мейо (DAI) для пациентов с ЯК. Всем больным проведены колоноскопия с биопсией, УЗИ органов брюшной полости с исследованием васкуляризации стенки кишечника (цветной допплер, энергетический допплер, контрастное исследование)....

The integration of Internet of Things (IoT) with Enterprise Resource Planning (ERP) systems represents a transformative advancement in modern business operations, revolutionizing how organizations manage and optimize their processes. This technical exploration delves into the architecture, implementation domains, and optimization strategies of IoT-ERP integration, highlighting the synergistic relationship between edge computing and enterprise systems. The content examines core components including data collection layers, communication infrastructure, and processing frameworks, while addressing critical aspects of security and performance optimization. The implementation across inventory management, supply chain visibility, and predictive maintenance demonstrates the practical impact of these integrations on operational efficiency. Technical challenges in data management and security are presented alongside innovative solutions, showcasing how edge computing and advanced system integration techniques enhance real-time processing capabilities. The discussion illuminates the transformative potential of IoT-ERP integration in driving digital transformation across industrial sectors.

Efficient optimization approaches are required for scientific computing in cloud environments to manage largescale calculations, dynamic workloads, and probabilistic decision making. This study investigates the role of Ant Colony Optimization (ACO), Gradient Descent (GD), and Bayesian Decision Models (BDM) in improving computational efficiency. A hybrid optimization framework is proposed, which includes ACO for resource allocation, GD for iterative learning, and BDM for probabilistic inference. The performance metrics examined were execution time, accuracy, resource utilization, and scalability. The results show that the hybrid model greatly improves computational performance by improving efficiency, adaptability, and optimization capabilities, making it a viable option for cloud-based scientific computing applications.

Описано структуру спеціалізованого програмного комплексу моделювання на основі ітераційних алгоритмів методу групового урахування аргументів (МГУА) з можливістю мультидоступу через Інтернет або локальну мережу. В програмному комплексі процес моделювання реалізовано у трьох режимах діалогу – два автоматичних (стандартний та плановий) та один інтерактивний, коли можна втручатися в процес самоорганізації моделей. Програмне забезпечення працює з різними наборами даних у форматі Excel та текстовому редакторі Блокнот. Будуються моделі різної складності та структури з різним розбиттям вибірки. Найкращі моделі представляються системою для графічного і змістового аналізу та зберігаються в базі даних разом із проміжними розрахунками і результатами експериментів для подальшого застосування. Система одночасно працює з трьома базами даних: початковою, базою даних проміжних розрахунків та базою даних результатів.The structure of specialized modeling software based on iterative algorithms of the g...

This paper investigates methods to improve Internet of Things (IoT) and robotic process automation (RPA) systems by integrating Principal Component Analysis (PCA), Least Absolute Shrinkage and Selection Operator (LASSO), and Elaborative Stepwise Stacked Artificial Neural Network (ESSANN). By enhancing data pre-processing, feature selection, and predictive modelling, these cutting-edge methods solve the difficulties associated with managing big, complicated datasets. In comparison to traditional methods, the suggested methodology showed a noteworthy improvement in mean squared error (MSE) of 0.05 and overall accuracy (95%), precision (92%), and recall (90%). Additionally, the study shows increased computing efficiency, which is critical to RPA and IoT system scalability. An ablation study verified the synergistic advantages of merging LASSO, ESSANN, and PCA; their combined application performed better than each of them separately in managing complex data and automating intricate procedures. The results indicate that this strategy has a great deal of promise to improve automation accuracy and dependability across a range of sectors.

The traditional health insurance industry relies on static, one-size-fits-all policies that fail to account for individual health needs, leading to customer dissatisfaction and inefficient risk assessment. Conventional methods struggle with dynamic data, unstructured inputs, and an inability to predict evolving health risks, resulting in misaligned coverage and poor personalization. To address these limitations, this study proposes a hybrid VAE-LSTM (Variational Autoencoder-Long Short-Term Memory) model combined with predictive analytics to enhance customer personalization in health insurance plans. The VAE compresses complex customer health data into a latent representation, while the LSTM captures temporal dependencies in sequential health records, enabling dynamic risk prediction and tailored policy recommendations. The proposed model achieves an exceptional AUC of 0.98 in the ROC curve analysis, demonstrating superior classification accuracy compared to traditional methods. Feature importance analysis reveals that Product Usage (40%) and Tenure (35%) are the most influential factors, while demographic data like Age (15%) provides supplementary insights. The model improves prediction accuracy by 25-30% over conventional rule-based or regression-based approaches, significantly enhancing customer satisfaction and retention. Future work includes integrating reinforcement learning for real-time policy adjustments and addressing cold-start challenges in new customer onboarding. This research marks a significant advancement in data-driven, personalized health insurance, optimizing both customer experience and insurer profitability.

This study presents a novel method for protecting cloud-based medical apps by combining Secure Healthcare Access Control Systems (SHACS) with Automated Threat Intelligence (ATI). By utilizing machine learning algorithms, anomaly detection methods, and real-time threat intelligence, the suggested framework improves cloud healthcare security. While SHACS guarantees safe, dynamic, and context-aware access management for critical healthcare data, ATI integration offers the capacity to proactively identify and address new cyber threats. In addition to fortifying the security architecture, this two-pronged strategy makes it easier to control access in real time while adjusting to changing security threats. The solution maintains data privacy and compliance while guarding against unauthorised access by guaranteeing compliance with important regulatory requirements like HIPAA and GDPR. Empirical testing revealed that the architecture could withstand complex attacks, as evidenced by its remarkable 94.2% threat detection rate and 95.3% resilience score. Additionally, the framework produced dependable security alerts with a low false-positive rate of only 3.2%. When compared to conventional methods, the suggested alternative provides notable gains in scalability, performance, and operational efficiency. By successfully reducing cybersecurity threats and upholding high system integrity, this integrated solution meets the growing demand for strong security measures in cloud-based healthcare systems. The findings validate the framework's promise as a secure, scalable, and effective solution for the healthcare industry, protecting private patient information in intricate and dynamic cloud environments. Future studies will concentrate on increasing scalability and optimizing resource usage without sacrificing security efficacy.