No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Two Decades of Recommender Systems at Amazon.com
Abstract
Amazon is well-known for personalization and recommendations, which help customers discover items they might otherwise not have found. In this update to their original paper, the authors discuss some of the changes as Amazon has grown.
... Video streaming platforms, such as Netflix 3 or Hulu, 4 allow users to watch large volumes of shows, movies, and documentaries. Music streaming platforms, such as Spotify 5 or Pandora, 6 provide similar capabilities. ...
... RS are ubiquitous in modern online platforms. E-commerce [2][3][4][5][6], advertising [7], video streaming [8], and music streaming [9,10] are prime examples of RS thriving behind the scenes [11]. Other applications exist across platforms, including job [12], television [13], travel [14], tourism [15], event [16], and social media recommendation (for example, friend [17] or information recommendation, such as retweets on Twitter [18,19]). ...
... This historical backdrop leads to a discuss of contemporary RS concerns in Sect. 6 as tags [41]. Poisson factorizations have become common in this domain [42][43][44]. ...
This survey is intended to inform non-expert readers about the field of recommender systems, particularly collaborative filtering, through the lens of the impactful Netflix Prize competition. Readers will quickly be brought up to speed on pivotal recommender systems advances through the Netflix Prize, informing their prospective state-of-the-art research with meaningful historic artifacts. We begin with the pivotal FunkSVD approach early in the competition. We then discuss Probabilistic Matrix Factorization and the importance and extensibility of the model. We examine the strategies of the Netflix Prize winner, providing comparisons to the Probabilistic Matrix Factorization framework as well as commentary as to why one approach became extensively used in research while another did not. Collectively, these models help to understand the progression of collaborative filtering through the Netflix Prize era. In each topic, we include ample discussion of results and background information. Finally, we highlight major veins of research following the competition.
... The key finding is that information cocoons facilitate consensus formation within groups while limiting the diversity of information individuals are exposed to. This exacerbates intergroup divergence and holds significant implications for academia, social governance, and policy formulation (Smith and Linden, 2017;Jannach and Jugovac, 2019). ...
... The advent of artificial intelligence (AI) has revolutionized ecommerce, with AI-powered recommendation systems playing a pivotal role in shaping consumer experiences and decisionmaking processes (Smith and Linden, 2017). First, AI recommendation systems excel at learning and adapting to individual preferences through continuous refinement based on user interactions and feedback (Adomavicius and Tuzhilin, 2005). ...
In today’s digital world, the inevitable phenomenon of information cocoons profoundly impacts individual and group decision-making. Although existing research has somewhat explored information cocoons in social media and politics, it lacks depth and direct evidence. This study focuses on the e-commerce environment, which differs significantly from social media and political contexts. We examine how information cocoons form among consumers in e-commerce and explore their influence on group consensus and division through a multimethod research design and simulation analysis. Key findings reveal that AI-driven recommendations strengthen the link between individual preferences and information cocoon formation. Information cocoon has dual effects on the consumer group; that is, it gathers the consensus of the members of the group and widens the differences between different groups. Simulation results illustrate the dynamics of these processes, providing insights into both consensus-building within groups and divergence across groups. Our study contributes by providing visual and direct evidence on the impact of information cocoons on group attitudes in e-commerce contexts, thereby expanding the external validity of the original theory. This has significant implications for academia and policymakers seeking to foster a more inclusive and diverse social environment. Moreover, it is crucial for developing strategies to optimize the information environment in e-commerce scenarios to fully make use of information cocoons.
... Algorithms, particularly those based on deep reinforcement learning, curate content that aligns with individual users' preferences. By analyzing user interactions such as likes, shares, and comments, and most importantly, engagement -algorithms learn to predict and present content that maximizes further future engagement (Smith & Linden, 2017;Pariser, 2011). This personalized content delivery creates a feedback loop where user behavior is continuously shaped by the content they consume, which in turn is tailored by the algorithms based on prior behavior. ...
... Modern social networks exploit these mechanisms through highly sophisticated algorithms, that are not designed in the best interest of the user, but rather to maximize and retain engagement. The reason is, that for social networking sites, more time spent on the platform translates into higher ad revenue (Pariser, 2011;Smith & Linden, 2017;Tufekci, 2015;Gillespie, 2014). This could also be supported by studies, that investigated the behavioral targeting of the algorithms, and which could confirm that personalized content, matching previous user interactions, resulted in higher click-through rates and general interaction, leading to higher advertising revenue (Bozdag, 2013;Helberger, 2019). ...
"Cognitive Nemesis" explores the intricate relationships among reinforcement loops, confirmation bias, echo chambers, cognitive dissonance, metacognition, and the Dunning-Kruger effect within the context of modern digital media. It synthesizes these established cognitive and behavioral concepts and integrates an array of scientific literature to provide original insights into how digital environments and their dynamics impact human cognition and behavior. These insights result from extensive research at the Zentrum für Medienpsychologie und Verhaltensforschung (ZeMV), and serve as an exploratory analysis offering new perspectives on the impact of technology on cognition and resulting behavior, on an individual and societal scale. Solidified through numerous real-world applications and case studies, the implications of this book can enhance the understanding of psychological explanations for phenomena, that are observed in a digitally immersed society.
... AI-based services have caused revolutionary changes in various sectors, and companies are leveraging them to drive strategies for providing customized services to consumers (Akdim and Casaló 2023;Loureiro et al. 2023;Nilashi et al. 2016). Companies are implementing AI-based recommendation services to provide customized services, enhancing user experience and increasing convenience and satisfaction (Smith and Linden 2017). The widespread adoption of smartphones and enhanced internet access has dramatically facilitated data consumption, leading to the integration of AI-based recommendation systems in various sectors. ...
... This study aimed to analyze the factors that could influence users' continuous intention to use AI-based recommendation services provided by digital platforms and the causal relationships between them. Recently, AI-based recommendation services have rapidly developed and come into the limelight as an important technology and service that understands users' preferences and patterns and provides customized recommendations (Akdim and Casaló 2023;Nilashi et al. 2016;Smith and Linden 2017). However, research investigating the continuous intention to use AI-based recommendation services is lacking. ...
This study investigates how standardization and customization influence AI service quality and their subsequent effects on user satisfaction and continuous intention to use, with a focus on AI service preference as a moderating factor. Analysis of 1032 survey responses using PLS-SEM revealed that while standardization positively affects AI service quality dimensions, customization shows a stronger positive impact. AI service satisfaction significantly influences continuous intention to use. Additionally, AI service preference demonstrates dual moderating effects: positive between AI system quality and satisfaction and negative between AI recommendation quality and satisfaction. These findings provide valuable insights for service providers seeking to enhance their market competitiveness through AI-based services.
... For 20 years, Amazon has been using artificial intelligence to recommend products after making an online purchase or just visiting the website. Amazon invented this concept using an algorithm called "item-based collaborative filtering" [25]. To suggest "frequently bought together" products, in addition to the purchase data of a particular consumer, Amazon uses the purchase history of other people who have purchased the same or similar product, as well as data on product satisfaction [26]. ...
Artificial intelligence has completely transformed social media marketing and turned social media into a place for business. The aim of this paper is to analyze the application, benefits, and risks of using artificial intelligence in social media marketing. Artificial intelligence is used in social media marketing for a range of activities including the analysis of data on consumer behavior, automation of content creation, advertising management, consumer relationship management, and consumer communication, especially by taking advantage of chatbots and virtual assistants. Marketers have discovered significant opportunities to process vast amounts of unstructured data on social media to gain invaluable consumer insights. The use of AI in social media marketing provides numerous benefits to companies, such as substantial cost and time savings, improved effectiveness and efficiency of social media advertising, improved consumer experience and personalization, as well as better risk management and crisis response. However, using AI in social media marketing also brings numerous risks and challenges, including concerns about data privacy, the potential for AI biases, misinformation, plagiarism, a lack of trained staff, and its impact on unemployment.
... Recommendation systems are used in many areas other than learning environments. Popular examples are Spotify [52], Amazon [53] and Netflix [54]. Different techniques exist to generate recommendations based on various data sources and quality. ...
In higher education, learners become increasingly heterogeneous as they differ in, e.g., knowledge levels, competences, learning styles and media preferences. It is difficult for instructors to cater for these differences individually in face-to-face courses due to the effort for individualized supervision which does not scale for larger groups. One solution is to supplement face-to-face contact with tailored learning materials for use outside of regular physical
classes. However, learners often do not know which material is most appropriate for them, which is why individualised
recommendations are necessary. This paper outlines the
reference architecture of an adaptive digital learning
environment and how it interfaces to other independent information systems, e.g., to retrieve student data which is available anyway. The digital learning platform offers personalised recommendations for learning elements, such as learning videos or quizzes, based on the behaviour and individual characteristics of the learner. It combines AI technologies with a didactic foundation based on self-directed learning. The reference architecture fulfils previously gathered requirements and integrates the platform well in an existing software landscape due to its flexible design.
... Elements such as visual appeal, intuitive navigation, one-click purchasing options, and social proof mechanisms (e.g., customer reviews and ratings) contribute to a seamless shopping experience that encourages impulsivity [29]. Moreover, personalized recommendations and streamlined purchasing processes further facilitate immediate purchases, making it easier for consumers to act on their impulses and driving higher conversion rates for online retailers [30,31]. Research has shown that the perceived quality of an offer plays a crucial role in building trust [32] and is considered one of the primary factors influencing trust [33]. ...
The swift growth of e-commerce has markedly changed how consumers shop, especially among Generation Z, which is called Digital Natives. This study examines how product presentation videos on the Shopee video platform influence impulse buying behaviors in this group, focusing on how internal stimuli, including entertainment experience (ET), educational experience (ED), escapist experience (ES), and esthetic experience (EH) influence online impulse buying (OIB) through the mediation of arousal (AR) and pleasure (PL). In addition, demographic factors, including age, gender, and income, are treated as control variables. This research adopts a quantitative methodology, and data was gathered using a Likert scale questionnaire and a non-probability sampling method, while the SmartPLS statistical tool was used to analyze the interactions of these stimuli and their effect on the impulse buying behavior of Generation Z on digital platforms. Research indicates that entertainment and recreational activities boost emotional engagement by eliciting arousal and pleasure. Educational experiences increase knowledge and also stimulate these feelings. Escapist activities provide temporary relief from daily stresses, increasing arousal, but can also highlight personal insecurities, possibly reducing pleasure. Esthetic experiences, subject to personal tastes, provoke emotional reactions that may vary in pleasure. For Generation Z, arousal and pleasure significantly influence impulsive buying decisions. The insights indicate that effectively managing internal factors can trigger emotions leading to impulsive purchases, offering strategic marketing tactics for optimizing e-commerce on platforms like Shopee video. This research advances the understanding of consumer behavior theories in the digital era, emphasizing the intricate roles of arousal and pleasure in online impulse buying.
... One of the top businesses in the field of electronic commerce and one of the pioneers of the recommendation system is Amazon [20]. Amazon's sales have increased by 60% due to its heavy reliance on suggestion algorithms, which are used on both its websites and email correspondence. ...
Improving the effectiveness of audience targeting in digital advertising campaigns is a key factor for companies' success in the digital age. With the rapid development of artificial intelligence technology, these campaigns have become more precise and engaging for the target audience. This paper reviews the importance of exploiting artificial intelligence tools to improve the effectiveness of digital advertising campaigns and e-marketing through improving personal targeting, big data analysis, and content personalization. The study includes a review of previous research on this topic and the various artificial intelligence tools used in this field, such as machine learning, big data analysis, and recommendation systems used in recommending products to users online while providing recommendations for improving digital targeting strategies.
... McKinsey reports that 35% of Amazon's consumer purchases are driven by these personalized recommendations. [4]. Alibaba's AIRec System: Alibaba's AIRec system uses big data to deliver real-time, personalized product suggestions. ...
This paper explores the transformative role of Generative Artificial Intelligence (GenAI) in enhancing the shopping experience across various industries. We delve into how GenAI personalizes interactions, offers dynamic pricing, and generates tailored product recommendations, significantly improving customer satisfaction and operational efficiency. The study highlights the profound benefits of GenAI in retail through practical case studies while addressing the challenges associated with its implementation, including privacy, data security, and ethical concerns. The research underscores the potential of GenAI to revolutionize e-commerce and suggests areas for future investigation, such as expanding GenAI applications for more accurate visual product displays, exploring cross-industry applications, and studying the long-term effects on consumer behavior. This paper contributes to the dialogue on integrating GenAI in retail, providing insights into its current successes and exploring avenues for innovative advancements.
... The algorithm framework is illustrated in Fig.1, the procedure of MOEA-MIAE is depicted in Algorithm1. 1R ← Probs(R); // Obtain a coarse ranking of candidate lists through the Probs method 2 P 1,1 , P 2,1 , P 3,1 ← Initialization(R,n); // Initialize the main evolutionary population P 1,1 and the auxiliary evolutionary populations P 2,1 , P 3,1 3 while terminal condition is not fulfilled (i <= g max ) do 4 for each generation population P i do ...
Traditional recommendation systems tend to focus on accuracy and prefer recommending popular items, resulting in non-popular items rarely being exposed to users. However, recommending non-popular items to enhance users' novelty experience is also crucial. Currently, many researchers are dedicated to multi-objective recommendation studies. Nevertheless, existing multi-objective recommendation algorithms often exhibit poor performance on the hypervolume value(HV) metric and lack effective methods to enhance novelty within evolutionary strategies. In this paper, we propose an innovative multi-objective recommendation algorithm based on a multi-population auxiliary evolution framework, abbreviated as MOEA-MIAE. Within this framework, we design three distinct optimization paths aimed at enhancing the convergence performance of the multi-objective algorithm and improving the hypervolume value metric of results. In addition to adopting the classical genetic algorithm as the main evolutionary population , we specifically introduce two auxiliary evolutionary populations. The first auxiliary population employs an HV-based multi-parent crossover method, while the second focuses on increasing the likelihood of generating highly novel solutions during crossover operations. These three evolutionary populations achieve effective complementarity and integration of their strengths through a mutual migration strategy of solution sets. Experimental results demonstrate that the proposed model exhibits superior performance in balancing accuracy and novelty, outperforming other comparable algorithms.
... Yapay zekâ, kişiselleştirilmiş pazarlama stratejilerinin geliştirilmesinde önemli bir rol oynamaktadır. Netflix ve Amazon gibi şirketler, YZ tabanlı öneri sistemleri kullanarak müşterilerine kişiselleştirilmiş içerik ve ürün önerileri sunmaktadır (Smith & Linden, 2017). Bu sistemler, müşterilerin geçmiş etkileşimlerini ve tercihlerini analiz ederek, onların ilgisini çekecek içerikleri otomatik olarak önermektedir. ...
Dijital dönüşüm, teknolojik gelişmelerin iş süreçlerine adaptasyonu ile örgütsel yapıları derinden etkilemektedir. Bu dönüşüm sürecinde insan faktörü, teknoloji ve örgüt arasındaki dinamik etkileşim kritik bir rol oynamaktadır. Dijitalleşme, iş gücünün yetkinliklerini yeniden tanımlarken, liderlik ve değişim yönetimi süreçlerine yeni boyutlar kazandırmaktadır. Organizasyonlar, yenilikçi teknolojilere uyum sağlamak için esnek ve çevik yapılar geliştirmek zorundadır. Yapay zekâ, büyük veri ve otomasyon gibi unsurlar, karar alma süreçlerinde etkinliği artırırken, insan kaynakları yönetimi açısından da yeni stratejiler gerektirmektedir. Dijital dönüşüm, yalnızca teknik bir süreç değil, aynı zamanda kültürel ve yönetsel bir paradigma değişimidir. Başarılı bir dönüşüm, teknolojiyi insan odaklı bir perspektifle ele alan bütünleşik bir yönetim yaklaşımını zorunlu kılmaktadır. Bu bağlamda, organizasyonların sürdürülebilir rekabet avantajı elde edebilmesi, insan-tekno-organizasyon uyumunu sağlamasına bağlıdır.
... This can spoil their experience on the platform and lead them to abandon it, which can have negative consequences on the turnover generated by the platform. It is to avoid such situations that recommender systems are designed and integrated into these digital platforms [11,15,18,23,24,28]. ...
submission to Episciences
Nowadays, e-commerce, streaming and social networks platforms play an important role in our daily lives. However, the ever-increasing addition of items on these platforms (items on Amazon, videos on Netflix and YouTube, posts on Facebook and Instagram) makes it difficult for users to select items that interest them. The integration of recommender systems into these platforms aims to offer each user a small list of items that match their preferences. To improve the performance of these recommender systems, some work in the literature incorporate explicit or implicit trust between platform users through trust-based recommender systems. Indeed, many of these works are based on explicit trust, when each user designates those whom they trust in the platform. But this information is rare in most real-world platforms. Thus, other work propose to estimate the implicit trust that each user can grant to another. However, work that estimates implicit trust does not take into account the temporal dynamics of users' past following actions and even less the fact that a user can influence another on one category of item and not on another. In this paper, we propose time and content aware strategies to estimate social influence of one user on another. The resulting time and content aware implicit trust are integrated to trust-based recommender systems build on K-Nearest Neighbors (KNN) and Graph-based techniques. Experiments done for rating predictions with KNN and Top-N recommendations with Graph model show that time and content aware implicit trust make it possible to improve the performance of the KNN according to the RMSE metric by 7% and 10%, and the performance of the graph model according to the NDCG@10 metric by 59% and 08% respectively on the Ciao and Epinions datasets.
De nos jours, les plateformes d’e-commerce, de streaming et de réseaux sociaux jouent un rôle important dans notre quotidien. Cependant, l’ajout toujours croissant d’éléments sur ces plateformes (articles sur Amazon, vidéos sur Netflix et YouTube, publications sur Facebook et Instagram) rend difficile pour les utilisateurs de sélectionner les éléments qui les intéressent. L'intégration de systèmes de recommandation dans ces plateformes vise à proposer à chaque utilisateur une petite liste d'éléments correspondant à ses préférences. Pour améliorer les performances de ces systèmes de recommandation, certains travaux de la littérature intègrent une confiance explicite ou implicite entre les utilisateurs de la plateforme via des systèmes de recommandation basés sur la confiance. En effet, nombre de ces travaux reposent sur une confiance explicite, lorsque chaque utilisateur désigne ceux en qui il a confiance dans la plateforme. Mais ces informations sont rares sur la plupart des plateformes du monde réel. Ainsi, d'autres travaux proposent d'estimer la confiance implicite que chaque utilisateur peut accorder à un autre. Cependant, les travaux qui estiment la confiance implicite ne prennent pas en compte la dynamique temporelle des actions passées des utilisateurs et encore moins le fait qu'un utilisateur peut en influencer un autre sur une catégorie d'item et pas sur une autre. Dans cet article, nous proposons des stratégies sensibles au temps et au contenu pour estimer l'influence sociale d'un utilisateur sur un autre. La confiance implicite qui en résulte, consciente du temps et du contenu, est intégrée aux systèmes de recommandation basés sur la confiance, construits sur les K-Nearest Neighbours (KNN) et les techniques basées sur les graphiques. Les expériences réalisées pour les prédictions de notation avec KNN et les recommandations Top-N avec le modèle Graph montrent que la confiance implicite consciente du temps et du contenu permet d'améliorer les performances du KNN selon la métrique RMSE de 7% et 10%, et les performances du modèle graphique selon la métrique NDCG@10 de 59% et 08% respectivement sur les jeux de données Ciao et Epinions.
... Particularly in the realm of recommender systems and CARS, deep learning technologies have been increasingly adopted, driving significant advancements in practical applications. For instance, YouTube has utilized deep neural networks to refine its video recommendation engine (Covington, Adams, & Sargin, 2016), Netflix has employed similar technologies to enhance its movie suggestions (Gomez-Uribe & Hunt, 2015), and Amazon has integrated these systems to improve its e-commerce recommendations (Smith & Linden, 2017). These implementations underscore the transformative impact of deep learning on modern recommender systems, not only improving the accuracy of predictions but also enhancing user engagement through more personalized content delivery. ...
In the domain of context‐aware recommender systems, understanding and leveraging feature interactions is crucial for enhancing recommendation quality. Feature interactions delve into the complex interdependencies among user characteristics, item attributes, and contextual factors like time and location. Traditional models often struggle to effectively combine these diverse features, potentially leading to suboptimal recommendations. To tackle this issue, we propose enhancing context‐aware recommender systems through deep feature interaction learning. Our model, which combines BiLSTM and Hybrid Attention mechanisms, offers a sophisticated architecture designed to exploit deep feature interactions effectively. This approach ensures that our system captures essential contextual dynamics, thereby improving the effectiveness of the recommendation process. Experimental results across multiple datasets validate the efficacy of our approach, showing significant improvements in key metrics such as and compared to traditional and contemporary models. These achievements underscore our model's ability to deliver nuanced and adaptively tailored recommendations, marking a valuable contribution to the field of recommender systems.
... Recommendation systems have been widely used in E-commerce or video portal to provide ranked lists of items/ services for users. It is reported that recommendations contribute about 80% of the traffic to Netflix [1], 60% to You-Tube [2], 20%-30% to Amazon [3,4], and 38% to Google News [5]. It is key to match consumers with most appropriate products, thereby increasing profits coupled with user satisfaction and loyalty. ...
Recommendation systems play a critical role in our daily lives. Despite great progress, existing graph-based recommendation methods still suffer from challenges including skewed data distribution, vulnerability to noises, and sparse supervision signal. We attribute the inferior performance to the limited discriminative ability of the learned representations. To remedy this, in this paper, we develop a framework termed Multi-ACG by introducing self-supervised learning, adversarial learning, and multitask learning to learn representations with higher discrimination. Specifically, self-supervised learning and adversarial learning are first employed to synthesize hard samples for training. Meanwhile, multi-task learning is adopted to balance different loss terms for optimization. Experiments are conducted on benchmark datasets and the results have demonstrated the state-of-the-art performance of the proposed method against previous ones. The code is at https://github.com/xiaoma666123/Multi-ACG.
... This architecture is designed with three key concepts: flexibility, modularity, and maintainability [1], [2]. Initially, major industry players, including Netflix [1], Amazon [3], Meta [4], and Google [5], adopted this architecture. However, as their applications scaled, they encountered an exponential increase in complexity, which led to the adoption of container orchestrators such as Kubernetes, Docker Swarm, Mesos, and Red Hat OpenShift [6]. ...
Microservice architecture is widely adopted among distributed systems. It follows the modular approach that decomposes large software applications into independent services. Kubernetes has become the standard tool for managing these microservices. It stores sensitive information like database passwords, API keys, and access tokens as Secret Objects. There are security mechanisms employed to safeguard these confidential data, such as encryption, Role Based Access Control (RBAC), and the least privilege principle. However, manually configuring these measures is time-consuming, requires specialized knowledge, and is prone to human error, thereby increasing the risks of misconfiguration. This research introduces K8s Pro Sentinel, an operator that automates the configuration of encryption and access control for Secret Objects by extending the Kubernetes API server. This automation reduces human error and enhances security within clusters. The performance and reliability of the Sentinel operator were evaluated using Red Hat Operator Scorecard and chaos engineering practices.
... • Conducting a study on the use of the Plebeian Algorithm on a selection of social media platforms and detecting the amount of misinformation over time after its implementation (ie, a real-world tested example) that would then be compared to current methods used, such as the aforementioned "Point-And-Shoot" Algorithm • Creating a type of sentiment analysis for graphical content that could examine the emotion within an image to determine if it could be misinformation (eg, Snapchat, Instagram, and TikTok) [63,64] • Determining the spread of misinformation correlated with the spread of viruses-this could be useful in predetermining locations (and users by extension) who are at higher risk of being exposed to or expounding misinformation • Exploring the applicability of the Plebeian Algorithm in surveillance contexts, including for criminal investigations, employee onboarding, and health care [65][66][67][68] • Analyzing the spread of misinformation through online vendors such as Amazon or eBay. In particular, recent audits of Amazon (as of 2021) show a dangerous disregard for reliable information, for example, presenting vaccine misinformation books along with well-cited vaccine information books in generic searches for vaccine information [69][70][71][72][73] • Applying models of higher sophistication for data analysis and visualization (which requires access to more in-depth data), including term frequency-inverse document measures [74] and Levenshtein distances [75] among others [76] • Examining the optimal method of implementation and integration for the Plebeian Algorithm with various existing networking systems and infrastructures • Continuing analysis of data collected to corroborate to prior studies on behavioral impacts of the sentiment of informative posts on social media • Analyzing the role of corporate social media platforms (ie, Slack) in the dissemination of misinformation, especially in private chat channels • Examining the misinformation containment models using juries, including the jury system implemented by Wikipedia • Analyzing the rise of audio-form content, including podcasts, Clubhouse, and Spotify Greenroom audio-chat rooms, for the potential spread of misinformation-many of these media are becoming increasingly influential sources of news and information for many [77] • Exploring the connection between location-based social media apps (eg, Foursquare) at the spread of geographic misinformation [78] COVID-19 has had substantial impacts upon modern society. ...
Background: The infodemic created by the COVID-19 pandemic has created several societal issues, including a rise in distrust
between the public and health experts, and even a refusal of some to accept vaccination; some sources suggest that 1 in 4 Americans
will refuse the vaccine. This social concern can be traced to the level of digitization today, particularly in the form of social
media.
Objective: The goal of the research is to determine an optimal social media algorithm, one which is able to reduce the number
of cases of misinformation and which also ensures that certain individual freedoms (eg, the freedom of expression) are maintained.
After performing the analysis described herein, an algorithm was abstracted. The discovery of a set of abstract aspects of an
optimal social media algorithm was the purpose of the study.
Methods: As social media was the most significant contributing factor to the spread of misinformation, the team decided to
examine infodemiology across various text-based platforms (Twitter, 4chan, Reddit, Parler, Facebook, and YouTube). This was
done by using sentiment analysis to compare general posts with key terms flagged as misinformation (all of which concern
COVID-19) to determine their verity. In gathering the data sets, both application programming interfaces (installed using Python’s
pip) and pre-existing data compiled by standard scientific third parties were used.
Results: The sentiment can be described using bimodal distributions for each platform, with a positive and negative peak, as
well as a skewness. It was found that in some cases, misinforming posts can have up to 92.5% more negative sentiment skew
compared to accurate posts.
Conclusions: From this, the novel Plebeian Algorithm is proposed, which uses sentiment analysis and post popularity as metrics
to flag a post as misinformation. This algorithm diverges from that of the status quo, as the Plebeian Algorithm uses a democratic
process to detect and remove misinformation. A method was constructed in which content deemed as misinformation to be
removed from the platform is determined by a randomly selected jury of anonymous users. This not only prevents these types of
infodemics but also guarantees a more democratic way of using social media that is beneficial for repairing social trust and
encouraging the public’s evidence-informed decision-making.
... The rapid and deep development of the internet has led to an information overload [1], and it is therefore difficult to obtain valuable information efficiently from vast amounts of data. Recommender systems [2][3][4] alleviate the problem and have achieved widespread success in numerous platforms such as Yahoo Music [5], Amazon e-commerce [6], Netflix [7], and YouTube [8]. Collaborative filtering [9][10][11], as one of the most popular recommendation techniques, plays a critical role in producing recommendations of high quality. ...
Recommender systems search the underlying preferences of users according to their historical ratings and recommend a list of items that may be of interest to them. Rating information plays an important role in revealing the true tastes of users. However, previous research indicates that natural noises may exist in the historical ratings and mislead the recommendation results. To deal with natural noises, different methods have been proposed, such as directly removing noises, correcting noise by re-predicting, or using additional information. However, these methods introduce some new problems, such as data sparsity and introducing new sources of noise. To address the problems, we present a new approach to managing natural noises in recommendation systems. Firstly, we provide the detection criteria for natural noises based on the classifications of users and items. After the noises are detected, we correct them with threshold values weighted by probabilities. Experimental results show that the proposed method can effectively correct natural noise and greatly improve the quality of recommendations.
... Now, they can use machine learning algorithms with the aim of executing predictive analytics and forecasting market trends, consumer behavior, and operational demands. In marketing, AI-powered recommendation systems such as those by Amazon and Netflix have transformed customer engagement using personal content and product preferences of each individual (Smith & Linden, 2017). Additionally, AI has transformed the development of chatbots and virtual assistants, changing the face of customer service. ...
This report delves into the integration of artificial intelligence (AI) with vision, audio, and language in the field of multimodal learning, which enables AI systems to process and analyze data coming from various sensory sources in order to gain a more overall view of the world. Multimodal AI enhances performance in tasks such as emotion recognition, image captioning, autonomous vehicle navigation, and medical diagnostics through the combination of visual, auditory, and linguistic information. Some of the notable applications of AI include personalized customer interactions via customer service, real-time decision making by autonomous vehicles, improved healthcare diagnosis and patient care, among other applications. The challenges in the responsible deployment of AI with respect to data fusion, privacy, bias, and transparency also feature within the report. Challenges notwithstanding, the report points to the enormous impact multimodal AI will make in revolutionizing industries through improved efficiency, safety, and personalization of a myriad of services. The prospect of future innovation of multimodal learning for AI promises to be path breaking and significantly advance the capabilities of AI systems in problems solving widely across domains.
Training large models for modern recommendation systems requires a substantial number of computational devices and extended periods. Since it is essential to store model checkpoints throughout the training progress for accuracy debugging or mitigating potential failures, checkpointing systems are widely used. However, given that recommendation models can scale to hundreds of gigabytes or more, existing solutions often introduce significant overhead in terms of both storage and I/O.
In this paper, we present IncrCP, a checkpointing system specifically designed for recommendation models. Given that only a small fraction of model parameters are modified in each iteration, IncrCP creatively leverages the incremental checkpointing strategy and overcomes the inherent slow recovery problem. To support recovering all states throughout the training process, while also ensuring efficient storage utilization and rapid recovery, IncrCP proposes the 2-D chunk approach. It proactively records changed parameters in the training process as well as their indexes, extracts parameters according to duplicated indexes as independent chunk files, and then orchestrates these chunks in the 2-dimensional linked list. In this way, IncrCP achieves fast recovery by loading less unnecessary parameters and performing less deduplication during recovery. Furthermore, IncrCP includes a selective extraction approach to reduce I/O by avoiding worthless extractions and a concatenate approach to reduce random disk access when recovery. Evaluations show that IncrCP achieves up to 6.6× recovery speedup compared to the naive incremental strategy and saves storage space by 60.4% with slight overhead compared to another recovery-friendly strategy.
Current movie recommendation systems often struggle to capture complex user preferences and dynamics, primarily relying on content-based or collaborative filtering techniques. This research introduces a novel deep learning-powered method to enhance movie recommendation models, addressing the limitations of existing systems. By analyzing user behavior records and utilizing movie content elements, our method guarantees the greatest degree of customisation. In this study, we employ Artificial Intelligence (AI), graph-based techniques, and text mining to accurately estimate user preferences. While PageRank ranks the films based on their importance in the individual’s history of surfing, Convolutional Neural Network (CNN) predicts the possibility that the movie would be accepted. The experiments employed a dataset of 215 users’ browsing activity in 508 movie pages for evaluation. The presented approach achieved significant enhancement in recommendation precision and recall metrics resulting in 7.15% precision expansion and 5.19% recall growth which indicates its potential implementation in personalized movie recommendation systems.
AI isn't just a concept of the sci-fi depth or a tool of the scientific forefront anymore, it has become a field of business ecosystem. AI is leading a transformation like never before from optimizing back-end processes to new age and market changing consumer engagement. Companies are realizing that AI unlocks a new way to work: one with more productivity, optimization, and innovation. The use of AI comes with challenges that deserve addressing, mainly ethics, workforce adjustment, and regulation. In this chapter the authors explore the transformation of AI in business, what lays ahead, why it matters, how to mitigate the challenges of AI, and how to architect for its transformative power— for building a sustainable platform of AI in an organization. The authors have taken three Case studies to illustrate the potential that exists with AI as well as the danger associated within various industries.
This chapter explores the integration of Artificial Intelligence (AI) and the Internet of Things (IoT) in retail, collectively referred to as AIoT, and its transformative impact on in-store customer experiences. By combining AI's ability to personalize customer interactions with IoT's capacity to connect and optimize physical environments, retailers can enhance engagement, improve operational efficiency, and create seamless, data-driven shopping experiences. The chapter examines AIoT's role in driving personalized retail experiences, its impact on customer behavior, and the technological innovations shaping future trends. It also discusses challenges and ethical considerations, including data privacy, cybersecurity, and algorithmic fairness. Ultimately, AIoT is positioned to revolutionize retail in the Experience Economy, offering significant opportunities for innovation and competitive advantage. The chapter concludes with recommendations for retailers to leverage AIoT responsibly while fostering customer trust.
User data spread across multiple modalities has popularized multi-modal recommender systems (MMRS). They recommend diverse content such as products, social media posts, TikTok reels, etc., based on a user-item interaction graph. With rising data privacy demands, recent methods propose unlearning private user data from uni-modal recommender systems (RS). However, methods for unlearning item data related to outdated user preferences, revoked licenses, and legally requested removals are still largely unexplored. Previous RS unlearning methods are unsuitable for MMRS due to the incompatibility of their matrix-based representation with the multi-modal user-item interaction graph. Moreover, their data partitioning step degrades performance on each shard due to poor data heterogeneity and requires costly performance aggregation across shards. This paper introduces MMRecUn, the first approach known to us for unlearning in MMRS and unlearning item data. Given a trained RS model, MMRecUn employs a novel Reverse Bayesian Personalized Ranking (BPR) objective to enable the model to forget marked data. The reverse BPR attenuates the impact of user-item interactions within the forget set, while the forward BPR reinforces the significance of user-item interactions within the retain set. Our experiments demonstrate that MMRecUn outperforms baseline methods across various unlearning requests when evaluated on benchmark MMRS datasets. MMRecUn achieves recall performance improvements of up to 49.85% compared to baseline methods and is up to 1.3× faster than the Gold model, which is trained on retain set from scratch. MMRecUn offers significant advantages, including superiority in removing target interactions, preserving retained interactions, and zero overhead costs compared to previous methods.
Purpose
This study aims to investigate how information quality and system quality influence the effectiveness of artificial intelligence (AI)-based recommendation service platforms. It integrates traditional information technology service quality (SQ) metrics with recommendation SQ measures, focusing on their impact on user satisfaction and behavior. This study further examines the moderating effects of standardization and customization on these relationships.
Design/methodology/approach
This study uses structural equation modeling to analyze data from 978 users of AI recommendation services. It evaluates the direct impacts of information quality (completeness, accuracy and format) and system quality (reliability, flexibility and timeliness) on recommendation quality (RQ).
Findings
The findings show significant positive effects of information quality and system quality on the quality of AI-generated recommendations, enhancing user satisfaction. This satisfaction is crucial for promoting continuous intention to use and positive word-of-mouth (WOM). This study also finds that standardization positively moderates the impact of RQ on WOM, whereas customization strengthens the relationship between satisfaction and continuous intention to use.
Originality/value
This research emphasizes the importance of quality metrics in shaping the efficacy of AI-based recommendation systems and highlights the need for a balance between standardization and customization to optimize user engagement and satisfaction. The findings offer valuable insights for AI service developers and marketers, emphasizing the significance of customized, high-quality recommendations to ensure sustained user engagement.
Many modern machine learning (ML) methods rely on embedding models to learn vector representations (embeddings) for a set of entities (embedding tables). As increasingly diverse ML applications utilize embedding models and embedding tables continue to grow in size and number, there has been a surge in the ad-hoc development of specialized frameworks targeted to train large embedding models for specific tasks. Although the scalability issues that arise in different embedding model training tasks are similar, each of these frameworks independently reinvents and customizes storage components for specific tasks, leading to substantial duplicated engineering efforts in both development and deployment. This paper presents MLKV, an efficient, extensible, and reusable data storage framework designed to address the scalability challenges in embedding model training, specifically data stall and staleness. MLKV augments disk-based key-value storage by democratizing optimizations that were previously exclusive to individual specialized frameworks and provides easy-to-use interfaces for embedding model training tasks. Extensive experiments on open-source workloads, as well as applications in eBay's payment transaction risk detection and seller payment risk detection, show that MLKV outperforms offloading strategies built on top of industrial-strength key-value stores by 1.6-12.6x. MLKV is open-source at https://github.com/llm-db/MLKV.
Artificial intelligence is the discipline that pursues the understanding, artificial replication and possible enhancement of human intelligence. Among others, AI-based recommender systems, persuasive systems as well as decision support systems are used for making decisions that have direct impact on people’s lives. In AI-supported decision making, the initiative may lie on the user’s side (such as recommender systems), or be largely automated (such as navigation systems or automotive driving). Many organizations also employ AI systems to make decisions that concern their users, customers, or citizens. But how do these systems learn about the people involved, how are these people represented, and—most importantly—how complete, realistic, reliable and fair is this representation? In this chapter, we discuss various considerations for user modeling in human-centered AI.
In this chapter, we look into the historical aspects and current trends of adaptation and personalization in human-centric AI. First, we distinguish the terms adaptation and personalization. Then we proceed on providing historical context on how adaptation evolved and personalization became a-kind-of adaptation. We further present the current trends, especially those based on human-centric aspects, which we refer to as theory-driven personalization. Finally, we critically assess the threats to adaptive and personalized human-centric AI and provide an outlook into the future.
Computerized change has gathered extensive consideration in the writing relating to data frameworks (IS) and authoritative science. Novel computerized advances are arising at a speeding up rate, subsequently it's basic to grasp what has been realized in over thirty years of exploration as well as what still needs to be perceived to use these computerized apparatuses completely. This part gives an outline of computerized change and lays out a calculated structure that fills in as the establishment for our conversation of the group of existing examination. The other exact parts in this altered version are situated involving the applied system too. Ultimately, we discuss what setting means for advanced change and point out a few varieties that nation of activity, size class, industry, and space. ARTICLE INFO
With the rapid development of e-commerce and the diversification of the computer hardware market, consumers are facing more and more difficulties in choosing computer accessories, and traditional e-commerce platforms are difficult to meet the user's demand for accurate shopping and efficient recommendation. The purpose of this paper is to build an intelligent computer accessories shopping platform (Beiji Technology), through intelligent recommendation algorithms, to improve the user shopping experience. The platform adopts mainstream front-end technology, efficient and stable back-end architecture, supports high concurrency data processing, and at the same time realizes dynamic interaction, which enhances the responsiveness and smoothness of the user interface. Its intelligent recommendation system generates personalized accessory recommendations based on the user's historical behavior, helping users find the products they need faster. The realization of this platform not only solves the pain points of traditional e-commerce platforms in the purchase of computer accessories, but also promotes the development of the e-commerce industry in the direction of intelligence and precision, and provides valuable references for the future research and practice of intelligent e-commerce.
Training sequential recommenders such as SASRec with uniform sample weights achieves good overall performance but can fall short on specific user groups. One such example is popularity bias, where mainstream users receive better recommendations than niche content viewers. To improve recommendation quality across diverse user groups, we explore three Distributionally Robust Optimization(DRO) methods: Group DRO, Streaming DRO, and Conditional Value at Risk (CVaR) DRO. While Group and Streaming DRO rely on group annotations and struggle with users belonging to multiple groups, CVaR does not require such annotations and can naturally handle overlapping groups. In experiments on two real-world datasets, we show that the DRO methods outperform standard training, with CVaR delivering the best results. Additionally, we find that Group and Streaming DRO are sensitive to the choice of group used for loss computation. Our contributions include (i) a novel application of CVaR to recommenders, (ii) showing that the DRO methods improve group metrics as well as overall performance, and (iii) demonstrating CVaR's effectiveness in the practical scenario of intersecting user groups.
We discuss the video recommendation system in use at YouTube, the world's most popular online video community. The system recommends personalized sets of videos to users based on their activity on the site. We discuss some of the unique challenges that the system faces and how we address them. In addition, we provide details on the experimentation and evaluation framework used to test and tune new algorithms. We also present some of the findings from these experiments.
Recommender systems have been evaluated in many, often incomparable, ways. In this article, we review the key decisions in evaluating collaborative filtering recommender systems: the user tasks being evaluated, the types of analysis and datasets being used, the ways in which prediction quality is measured, the evaluation of prediction attributes other than quality, and the user-based evaluation of the system as a whole. In addition to reviewing the evaluation strategies used by prior researchers, we present empirical results from the analysis of various accuracy metrics on one content domain where all the tested metrics collapsed roughly into three equivalence classes. Metrics within each equivalency class were strongly correlated, while metrics from different equivalency classes were uncorrelated.
Recommender systems apply knowledge discovery techniques to the problem of making personalized recommendations for information, products or services during a live interaction. These systems, especially the k-nearest neighbor collaborative filtering based ones, are achieving widespread success on the Web. The tremendous growth in the amount of available information and the number of visitors to Web sites in recent years poses some key challenges for recommender systems. These are: producing high quality recommendations, performing many recommendations per second for millions of users and items and achieving high coverage in the face of data sparsity. In traditional collaborative filtering systems the amount of work increases with the number of participants in the system. New recommender system technologies are needed that can quickly produce high quality recommendations, even for very large-scale problems. To address these issues we have explored item-based collaborative filtering techniques. Itembased techniques first analyze the user-item matrix to identify relationships between different items, and then use these relationships to indirectly compute recommendations for users. In this paper we analyze different item-based recommendation generation algorithms. We look into different techniques for computing item-item similarities (e.g., item-item correlation vs. cosine similarities between item vectors) and different techniques for obtaining recommendations from them (e.g., weighted sum vs. regression model). Finally, we experimentally evaluate our results and compare them to the basic k-nearest neighbor approach. Our experiments suggest that item-based algorithms provide dramatically better performance than user-based algorithms, while at the same time p...
When leveraging the crowd to perform complex tasks, it is imperative to identify the most effective worker for a particular job. Demographic profiles provided by workers, skill self-assessments by workers, and past performance as captured by employers all represent viable data points available within labor markets. Employers often question the validity of a worker's self-assessment of skills and expertise level when selecting workers in context of other information. More specifically, employers would like to answer the question, "Is worker confidence a predictor of quality?" In this paper, we discuss the state-of-the-art in recommending crowd workers based on assessment information. A major contribution of our work is an architecture, platform, and push/pull process for categorizing and recommending workers based on available self-assessment information. We present a study exploring the validity of skills input by workers in light of their actual performance and other metrics captured by employers. A further contribution of this approach is the extrapolation of a body of workers to describe the nature of the community more broadly. Through experimentation, within the language-processing domain, we demonstrate a new capability of deriving trends that might help future employers to select appropriate workers.
This article discusses the various algorithms that make up the Netflix recommender system, and describes its business purpose. We also describe the role of search and related algorithms, which for us turns into a recommendations problem as well. We explain the motivations behind and review the approach that we use to improve the recommendation algorithms, combining A/B testing focused on improving member retention and medium term engagement, as well as offline experimentation using historical member engagement data. We discuss some of the issues in designing and interpreting A/B tests. Finally, we describe some current areas of focused innovation, which include making our recommender system global and language aware.
Recommendation systems are an increasingly prominent part of the web,
accounting for up to a third of all traffic on several of the world's most
popular sites. Nevertheless, little is known about how much activity such
systems actually cause over and above activity that would have occurred via
other means (e.g., search) if recommendations were absent. Although the ideal
way to estimate the causal impact of recommendations is via randomized
experiments, such experiments are costly and may inconvenience users. In this
paper, therefore, we present a method for estimating causal effects from purely
observational data. Specifically, we show that causal identification through an
instrumental variable is possible when a product experiences an instantaneous
shock in direct traffic and the products recommended next to it do not. We then
apply our method to browsing logs containing anonymized activity for 2.1
million users on Amazon.com over a 9 month period and analyze over 4,000 unique
products that experience such shocks. We find that although recommendation
click-throughs do account for a large fraction of traffic among these products,
at least 75% of this activity would likely occur in the absence of
recommendations. We conclude with a discussion about the assumptions under
which the method is appropriate and caveats around extrapolating results to
other products, sites, or settings.
Recommendation algorithms are best known for their use on e-commerce Web sites, where they use input about a customer's interests to generate a list of recommended items. Many applications use only the items that customers purchase and explicitly rate to represent their interests, but they can also use other attributes, including items viewed, demographic data, subject interests, and favorite artists. At Amazon.com, we use recommendation algorithms to personalize the online store for each customer. The store radically changes based on customer interests, showing programming titles to a software engineer and baby toys to a new mother. There are three common approaches to solving the recommendation problem: traditional collaborative filtering, cluster models, and search-based methods. Here, we compare these methods with our algorithm, which we call item-to-item collaborative filtering. Unlike traditional collaborative filtering, our algorithm's online computation scales independently of the number of customers and number of items in the product catalog. Our algorithm produces recommendations in real-time, scales to massive data sets, and generates high quality recommendations.
The Netflix Recommender System: Algorithms, Business Value, and Innovation
- C A Gomez-Uribe
- N Hunt
Method and System for Associating Feedback with Recommendation Rules
- K Chakrabarti
- B Smith