Recommender Systems Research Articles

Evaluating the quality of medical content on YouTube using large language models

YouTube has become a dominant source of medical information and health-related decision-making. Yet, many videos on this platform contain inaccurate or biased information. Although expert reviews could help mitigate this situation, the vast number of daily uploads makes this solution impractical. In this study, we explored the potential of Large Language Models (LLMs) to assess the quality of medical content on YouTube. We collected a set of videos previously evaluated by experts and prompted twenty models to rate their quality using the DISCERN instrument. We then analyzed the inter-rater agreement between the language models’ and experts’ ratings using Brennan–Prediger’s (BP) Kappa. We found that LLMs exhibited a wide range of inter-rater agreements with the experts (ranging from −1.10 to 0.82). All models tended to give higher scores than the human experts. The agreement on individual questions tended to be lower, with some questions showing significant disagreement between models and experts. Including scoring guidelines in the prompt has improved model performance. We conclude that some LLMs are capable of evaluating the quality of medical videos. If used as stand-alone expert systems or embedded into traditional recommender systems, these models can mitigate the quality issue of health-related online videos.

Clustering recommendation systems: Performance and accuracy comparison

Nowadays, there is an increasing trend in movie creation and consumption through video-on-demand platforms like Netflix. It is vital for movie industry companies of this kind to adopt a recommendation-based marketing strategy, as improving customer experience directly contributes to increased profits. In our previous research work, we improved the efficiency of the standard k-NN (K-Nearest Neighbor)-based recommendation system in both accuracy and speed by first clustering customers based on their preferred genres. For the clustering step, we used the K-Means algorithm. In this research work, we expanded our experiments by using other than K-Means clustering algorithms to implement the clustering step of our proposed recommendation system. We used the DBSCAN Clustering algorithm, the Agglomerative Clustering, the Affinity Propagation, the Spectral Clustering, the Mean Shift, the Gaussian Mixture and the K-Means Split Merge algorithms. K-Means clustering is one of the most widely used partitional clustering algorithms in recommender systems due to its simplicity and low computational cost, despite some inherent limitations. Its popularity persists due to its simplicity and its minimal computational requirements. This research aims to improve K-Means-based recommender systems by experimenting with alternative clustering methods.

Intrinsic and Extrinsic Factor Disentanglement for Recommendation in Various Context Scenarios

In recommender systems, the patterns of user behaviors (e.g., purchase, click) may vary greatly in different contexts (e.g., time and location). This is because user behavior is jointly determined by two types of factors: intrinsic factors , which reflect consistent user preference, and extrinsic factors , which reflect external incentives that may vary in different contexts. Differentiating between intrinsic and extrinsic factors helps learn user behaviors better. However, existing studies have only considered differentiating them from a single, pre-defined context (e.g., time or location), ignoring the fact that a user's extrinsic factors may be influenced by the interplay of various contexts at the same time. In this paper, we propose the Intrinsic-Extrinsic Disentangled Recommendation (IEDR) model, a generic framework that differentiates intrinsic from extrinsic factors considering various contexts simultaneously, enabling more accurate differentiation of factors and hence the improvement of recommendation accuracy. IEDR contains a context-invariant contrastive learning component to capture intrinsic factors, and a disentanglement component to extract extrinsic factors under the interplay of various contexts. The two components work together to achieve effective factor learning. Extensive experiments on real-world datasets demonstrate IEDR’s effectiveness in learning disentangled factors and significantly improving recommendation accuracy by up to 4% in NDCG.

Whittle Index-Based Q-Learning for Contextual Restless Bandits: A Case Study in Email Marketing

Email marketing has become an essential tool for businesses to reach out to potential customers [16]. However, the use of mass marketing raises ethical and environmental issues. In fact, this practice tarnishes domain reputation with potential spamming and overwhelms recipients with irrelevant content. Not only does this decrease user engagement, but it also results in a negative user experience. Moreover, it contributes to significant carbon emissions [5], especially with emails containing heavy attachments. To address these challenges, we propose a novel approach that frames the problem as a Contextual Restless Bandit Recommender System aiming to target relevant user segments while adapting to dynamic environments. The objective is to offer a more effective and sustainable solution, minimizing unnecessary email transmissions and ensuring that recipients receive content tailored to their profile.

A context and sequence-based recommendation framework using GRU networks

Abstract Recommendation systems play a significant contribution in e-commerce for predicting the more relevant product to the customers based on their interests. The recommendation system refers to the user-item interaction and predicts the next item by considering the similar kind of user interest or item purchased. The context-aware and sequential recommendation is built to predict the interested product based on the current context and sequential behavior pattern interactions. To fulfill the customers’ requirements, this paper proposes a new hybrid personalized recommendation system framework called Target User Context Sequential Prediction Gated Recurrent Unit (TUCSP-GRU) using deep learning methods to recommend suitable products to the users based on their interests and context. The proposed system uses the newly calculated Target User Specific Product Rating (TUS-PR) score, the proposed TUS Gated Recurrent Unit (TUS-GRU) model, and the proposed Top-N item prediction method. Here, (i) the TUS-PR score is used to improve the product rating, (ii) the new TUS-GRU model is used to find the sequence purchase behavior pattern of customers by considering their long-term and short-term interests, and (iii) the proposed Top-N item dynamic prediction method is used to adjust the next interested item list based on the response using the back propagation continuous learning method. The experiment results of the TUCSP-GRU framework show better accuracy in predicting the interested and relevant products or items when compared to existing similar recommendation systems with respect to the standard evaluation metrics.

Empowering Precise Crop Recommendation System by Accompanying Tree Covariance Matrix-Parallel Random Forest Classifier

Transformation in crop management systems, particularly in creating an environment that gives rise to sustainable farming, is achieved due to innovation and the advancement of modernized agricultural technology. Anyhow, meeting the increasing food demand is one of the great challenges that stand in front of the farmers. By taking into account, factors like soil, climate, and seasonality, the crop recommendation system plays a central role in providing customized guidance to the farmers. Current crop recommendation models are often confined by a paucity of feature selection, spatial-temporal integration shortfalls, and a finite amount of decision-tree diversity. All these shortfalls retrain their scalability and accuracy. To overcome the aforementioned blocks, an innovative framework is projected that includes the Best Incremental Random Subset (BIRS) feature selection method for choosing the best features and the Parallel Random Forest (PRF) -Tree Covariance Matrix model (PRF-TCM) encourages decision-tree diversity, permitting more accurate and efficient crop recommendations. Experimental results reveal that the proposed framework outperforms existing models with accuracy (89.7), precision (88.6), and recall (87.5). The framework shows significant improvements over current models, responsible for more viable agricultural practices.

Multi-Disease Detection with Doctor Recommendation System

AI and machine learning, from self-driving cars to health care, have turned out to be an indispensable tool for several industries. In medicine, the big availability of patient records now opens new possibilities for applying techniques of machine learning for disease detection and diagnosis at earlier stages. The main objective of this work is to provide an advanced prediction system capable of detecting multiple diseases. This will overcome one of the major disadvantages of most systems, which usually target single diseases and may or may not do so very accurately. Our system will target the following major five diseases: Heart Disease, Liver Disease, Diabetes, Lung Cancer, and Parkinson's Disease-specifically, although it has the flexibility for future gains on other conditions. The following project incorporates different disease-specific parameters, wherein a user can enter their health data to get accurate predictions of the presence of a disease. This project would have a greater influence on enabling people to monitor and take precautions for better maintenance of health, thereby helping to increase the life span of an individual. In using machine learning in this respect, such a system may support individual well-being by providing the most accurate disease predictions possible, saving lives if that be the case.

Building a Machine Learning Algorithm-Based Model to Suggest Tourist Attractions in Response to Travelers' “Slow Life” Requirements

In order to promote the development of personalized recommendation system in the tourism industry, this study adopts a machine learning collaborative filtering algorithm model, and uses data provided by users and data mining technology to reduce the overload of Internet messages and provide customized content recommendations. The construction of tourist attractions recommendation model based on machine learning is 18% better than the traditional algorithm under the demand of tourists' “slow life”. This research can effectively solve the problems of sparse data and cold start. It provides users with more personalized and people-centered travel route recommendations, which significantly enhancing the travel experience.

EXPRESS: Too Many or Too Few? Information Cues in Recommender Systems and Consequences for Search and Purchase Behavior

EXPRESS: Too Many or Too Few? Information Cues in Recommender Systems and Consequences for Search and Purchase Behavior

Personalized Self‐Directed Learning Recommendation System Based on Social Knowledge in Distributed Web

Personalized Self‐Directed Learning Recommendation System Based on Social Knowledge in Distributed Web

Distributed Sharing and Personalized Recommendation System of College Preschool Education Resources Under the Intelligent Education Cloud Platform Environment

The emergence of Intelligent Education Cloud Platforms has revolutionized how educational resources are shared and utilized, enabling a more inclusive and adaptive approach to learning. In the context of college preschool education, the integration of distributed sharing and personalized recommendation systems addresses critical challenges in resource accessibility and learner engagement. Traditional methodologies often rely on static and generalized frameworks, lacking the flexibility to cater to diverse learning needs and the scalability for dynamic resource allocation. These limitations hinder their capacity to provide tailored educational pathways and real-time adaptability, which are essential in preschool education’s highly individualized context. To overcome these barriers, we propose a Knowledge-Adaptive Education Network (KAEN) augmented by the Adaptive Learning Pathway Strategy (ALPS), tailored for deployment on Intelligent Education Cloud Platforms. KAEN leverages graph-based knowledge representation, dynamic content alignment networks, and reinforcement learning to optimize resource recommendation and personalize learning experiences. ALPS complements this system by generating individualized learning pathways, integrating multi-modal content, and real-time feedback mechanisms to enhance engagement and educational outcomes. Experimental validation demonstrates significant improvements in resource utilization efficiency, learner engagement metrics, and adaptive content delivery quality. These findings underscore the potential of integrating AI-driven frameworks into Intelligent Education environments, offering scalable and effective solutions for preschool education resource sharing and personalization.

Applying Optimization Techniques for Enhancing Personalization to Recommend a Web Page

Web page recommendation system has been emerging as the most important area in Service computing. Web pages are analyzed and selected for recommendation in order to favor end users while searching for information. Collaborative filtering and content based approaches are two predominant techniques for recommending web pages. Traditional Naive Bayes based probabilistic approach has also shown drastic improvement in achieving personalization during Web page recommendations. However, to improve the accuracy and enhance user satisfaction, we have analyzed optimization techniques such as Ant Colony Optimization and Particle Swarm Optimizations for enhancement of personalization in web search. Here, user profiles comprising of Usage-based and Content-Based attributes are clustered based on similarity in search history. Optimization algorithms are applied to select final web pages from the set of users within the matching cluster. Experiments were carried out with datasets covering 7175 web pages accessed by 287 different users. Result shows that Particle Swarm Optimization outperforms other traditional methods with improved performance.

Resolving the Cold-Start Issue in Recommender Systems with Reinforcement Learning

The cold-start problem faced by recommender systems is a serious problem, mainly because of the lack of historical data for new users or items. Traditional recommendation techniques, such as collaborative filtering and content-based filtering, are prone to fail in making good recommendations under such conditions. This paper explores the use of reinforcement learning (RL) as a remedy for cold-start problems based on active learning methods and multi-armed bandit models. We propose a novel RL-based approach that learns user preferences incrementally from interaction and improves recommendations in an exploration-exploitation setting. The setting shows improved performance in personalization and user engagement compared to baseline methods. This paper also provides the first-order implications of the cold-start problem in the real world and assesses the challenges faced by industries impacted by this problem. Keywords - Recommender systems, cold-start problem, reinforcement learning, Markov decision processes, deep reinforcement learning, user modelling

Leveraging social circles and algorithmic processes in digital mental health tools for college students' stress management

Introduction. College students frequently face stress from academic pressures, social demands, and the challenges of daily life. Digital mental health (DMH) interventions have shown promise in helping manage these stressors. However, many students quickly abandon DMH tools, which limits their effectiveness. Previous research has identified social interaction as a key factor in improving user engagement with DMH tools. Still, most existing DMH tools have not been designed with these sharing practices in mind. Moreover, research on recommender systems (RSs) has mainly focused on single-user interactions, leaving a gap in the development of RSs for group settings, particularly in the health domain. Method. To address this gap, we developed a technological probe that integrates social features with algorithmic processes to provide recommendations for both individuals and groups, aimed at supporting students' stress management efforts. We conducted a user study with 15 college students and their social circles, who interacted with the probe over the course of a day. Results. Our findings revealed that students appreciated the opportunity to share daily activities with friends and felt motivated to engage in well-being-enhancing behaviours. Participants also gave valuable feedback on the stress management recommendations. We discuss and provide recommendations for future research.

Taming TikTok: how BIPOC individuals perceive and interact with algorithmically generated content

Introduction. TikTok’s recommendation algorithm plays a crucial role in shaping user experiences, raising concerns about algorithmic bias, content suppression, and misinformation, particularly for BIPOC users. This study explores how BIPOC individuals perceive and interact with TikTok’s algorithm, focusing on content visibility, algorithmic manipulation, and experiences with problematic content. Method. This pilot study utilized semi-structured interviews with 10 BIPOC TikTok users, recruited via social media. Participants discussed their experiences with the platform’s recommendation system, and responses were transcribed and thematically analyzed to identify patterns in content exposure, engagement strategies, and perceptions of algorithmic bias. Analysis. A qualitative thematic coding approach was applied to categorize responses based on key themes, including content representation, misinformation, and strategies for mitigating algorithmic harm. Results. Findings reveal that BIPOC users frequently encounter discriminatory content, misinformation, and algorithmic suppression. Many employ strategies such as selective engagement, the "Not Interested" feature, and algorithm resets to curate their feed. However, skepticism remains regarding TikTok’s data practices and fairness in content moderation. Conclusions. BIPOC users actively shape their digital experiences to counteract algorithmic bias, highlighting systemic inequities in AI-driven platforms. Future research should expand participant diversity and investigate long-term algorithmic trends to promote inclusive digital spaces.

Predictive Anlaysis in Publishing Machin Learning Models for Best Seller

Book reading will play a vital in personality development ,creativity and thinking of a person.This project aims to analyse the books which have been published and based on their ratings and similarity between the books.The GUI can recommend the book of similar author or same type .Such that the based on the ratings of users and the similarity between them the recommendation system can recommend the books for the user type.It can be helpful in promoting the book of users and it will helpful in success of publishing of a book .With the help of machine learning.

Symmetric Graph Contrastive Learning against Noisy Views for Recommendation

Graph Contrastive Learning (GCL) leverages data augmentation techniques to produce contrasting views, enhancing the accuracy of recommendation systems through learning the consistency between contrastive views. However, existing augmentation methods, such as directly perturbing interaction graph (e.g., node/edge dropout), may interfere with the original connections and generate poor contrasting views, resulting in sub-optimal performance. In this paper, we define the views that share only a small amount of information with the original graph due to poor data augmentation as noisy views (i.e., the last 20% of the views with a cosine similarity value less than 0.1 to the original view). We demonstrate through detailed experiments that noisy views will significantly degrade recommendation performance. Further, we propose a model-agnostic Symmetric Graph Contrastive Learning (SGCL) method with theoretical guarantees to address this issue. Specifically, we introduce symmetry theory into graph contrastive learning, based on which we propose a symmetric form and contrast loss resistant to noisy interference. We provide theoretical proof that our proposed SGCL method has a high tolerance to noisy views. Further demonstration is given by conducting extensive experiments on three real-world datasets. The experimental results demonstrate that our approach substantially increases recommendation accuracy, with relative improvements reaching as high as 12.25% over nine other competing models. These results highlight the efficacy of our method. The code is available at https://github.com/user683/SGCL .

A case-based reasoning and ontology-based hybrid recommender system for student orientation in higher education

A case-based reasoning and ontology-based hybrid recommender system for student orientation in higher education

Enhancing Collaborative Filtering with Game Theory for Educational Recommendations: The Edu–CF–GT Approach

In the field of education, the proliferation of e-learning platforms has considerably increased access to teaching material. However, this abundance of resources poses a serious challenge to learners in the form of information overload that hinders the learning process. To meet this challenge, effective mechanisms need to be put in place to guide learners towards resources that are tailored to their individual needs and preferences. Recommendation systems appear to be essential tools in this context, aiming to personalise the learning experience by offering targeted suggestions based on the user’s preferences. This article presents EDU–CF–GT, a new educational recommendation model, as a solution to this challenge. Based on our generic CF–GT model, EDU–CF–GT is adapted to the complexities of the educational domain, improving learning efficiency by simplifying access to resources. Through evaluation on an educational dataset, EDU–CF–GT demonstrates significant improvements in recommendation relevance and learner satisfaction compared to existing models.

Custom Clothing Design and Ordering Platform with Generative AI

The increase in customer preferences towards mainly be spoke fashion leads to the need of applications that would allow customers to easily design and order clothes online. However, as many existing platforms have shown, it is difficult to incorporate design tools that are easy to use, real time previews which are important and recommendations that consider the user preferences appropriately. To fill these voids, this paper presents the “Custom Clothing Design and Ordering Platform with Generative AI”. Being equipped with sophisticated machine learning and generative AI features, the platform provides users with an innovative environment that allows them creating designs suitable for their needs. These are such as AI-generated design proposals, live visualisations, and an intelligent recommendation system for fabrics and patterns chosen by the user as well as trend prediction. The platform taps into generative AI to improve design and colour choices with design features that support individual body fit and product size. Moreover, sentiment analysis is incorporated in order to analyse the reception of the users and the further design advice. By simplifying the ordering process, and providing some techniques for demand forecasting, it makes the process effective to both the customers and manufacturers. Through the application of an artificial intelligence mechanism, this revolutionary technology alters the customer experience to create a new model for personalized clothing production for independent fashion designers and businesses. Keywords: Custom Clothing Design, Generative AI, Machine Learning, Real-Time Preview, Fashion Individualisation, User-Experience Design