In this article, a novel data augmentation method is developed for Meta-Services based on large language models (LLMs) and requirement-functional-logical-physical (RFLP) models, referred to as generative LLM-driven data augmentation (GLIDE), and an application of GLIDE to e-commerce recommendation systems is presented. First, the framework of GLIDE is designed under Meta-Services, including a generalized encoder and decoder architecture, as well as designers, which captures the idea of “designers in the loop” in Meta-Services. Furthermore, the generalized encoder and decoder architecture is established based on RFLP models in systems engineering, aiming to translate requirements and knowledge into prompts and obtain ideal LLM outputs under physical constraints. Subsequently, the application of the GLIDE method to e-commerce recommendation systems is given in detail, which is committed to enriching users’ behavioral data and providing an effective way to decouple data augmentation and recommendation modules. Finally, recommendation experiments are performed on real-world and synthetic behavior data, and the empirical results demonstrate the effectiveness of our GLIDE method.

In this paper, a novel nonzero-sum game (NSG) method is developed for completely unknown nonaffine nonlinear discrete-time (DT) systems, which is referred to as model-free NSG (MNSG). First, novel dynamic control laws are developed for NSGs using parallel control, namely introducing controls into feedback. Subsequently, an augmentedN-player NSG is formulated according to the originalN-player NSG to derive the dynamic control laws. Furthermore, we show that the control stabilities of the original and augmentedN-player NSGs are equivalent. In the meantime, we prove that optimal control of the augmentedN-player NSG is equivalent to near-optimal control of the originalN-player NSG, and the Nash equilibrium of the originalN-player NSG can be achieved. Then, a model-free learning scheme is developed to obtain the solution of the augmentedN-player NSG using online policy iteration, and neither using a model network to predict unknown dynamics nor off-policy reinforcement learning (RL) is needed in the scheme. Lastly, numerical analysis, including the NSG of a DT system with unknown control-nonaffine dynamics and coupled controls, confirms the correctness of our MNSG method. The associated code is available at: https://github.com/lujingweihh/Adaptive-dynamic-programming-algorithms/tree/main/model_free_nonzero_sum_games_discrete_time.

Recommender systems play significant roles in business, especially in e-commerce. Nevertheless, users’ behaviors are usually mixing and drifting, which is hard to tackle. Current sequential methods of item-wise interest extracting suffer from the intricacy and sparsity of data. Inspired by that category-wise interests may be intrinsic drivers of user behaviors and heterogeneous actions may reveal certain behavior patterns, we introduce intention as a tuple of category and action to address the data issues. In this paper, an intention-aware Markov chain based sequential recommendation model (IMRec) is proposed. We model the overall preferences of users as the integration of long-term preferences and short-term intents. In particular, the matrix factorization method is adopted to extract the long-term user-item preference. For the modeling of the short-term intents transition, we adopt high-order Markov chain based methods. A factorized mixture transition distribution model for high-order Markov chain approximation is leveraged in this paper to reduce the algorithm complexity. An auxiliary loss on intention representation is utilized, which brings considerable performance improvements. Experiments on real-world datasets demonstrate that our model outperforms state-of-the-art baseline models in terms of three common metrics, and shows superior stability, scalability, and training efficiency. Note to Practitioners—In e-commerce, sequential recommenders are essential to facilitate decision-making and promote business. A big challenge is to capture the sequential patterns from the mixing and drifting user-item interaction sequences. Customer behaviors are often driven by their inherent intentions, which may present more stable and reliable patterns. Motivated by that, we propose a novel intention-aware next-item recommendation algorithm with high performance. Our method models the user-item preferences as the integration of long-term preferences and short-term intents. More specifically, long-term preferences show the general tastes of users, which are modeled by latent factors of users and items. Additionally, the intention is defined as a tuple of category and action. We model the short-term intents as the intention transition from the past intentions by a high-order Markov chain. We further leverage a factorization-based mixture transition distribution model for high-order Markov chain approximation and reduce the algorithm complexity. The proposed model is validated in 4 real-world datasets and shows good recommendation performance, performance stability, model scalability, and training efficiency. Our model provides an effective recommendation method at low computational costs for e-commerce companies.

Increasing demand for hospitalization requires hospitals to optimize the admission schedules of elective patients to minimize operation cost and improve service quality. In this study, we propose a robust predict-then-optimize methodology to address the elective patient admission scheduling problem under uncertainty. The objective is to minimize total cost associated with postponement and daily bed over-utilization considering uncertain patients’ length of stay (LOS). Starting from prediction models, we first predict patients’ LOS using elaborate clusterwise regression methods. Considering the distributions of the regression residuals, we propose two-stage stochastic programming (SP) and distributionally robust optimization (DRO) approaches to model and solve the elective patient admission scheduling problem. We reformulate the proposed DRO model and construct a column-and-constraint generation algorithm to solve it efficiently. Finally, using real-world data, we conduct extensive numerical experiments comparing the performance of our proposed DRO model with benchmark methods, and discuss insights and implications for elective patient admission scheduling. The results show that our proposed DRO model can help hospitals manage high quality care, i.e., proper bed occupancy rates, at a reduced cost.Note to Practitioners—This article is motivated by our collaborations with a tertiary hospital in Beijing, China. From the perspective of hospital admission centers, we consider an elective patient admission scheduling problem that must decide the admission time for elective patients within a specified planning horizon. However, this is a challenging optimization problem due to patients’ uncertain LOS. Additionally, it is difficult to accurately describe the probability distribution of patients’ LOS. The managers find it difficult to give high-quality admission schedules to patients when they are registered which may reduce patients’ satisfaction. Therefore, we propose a predict-then-optimize framework to solve this problem. In particular, a two-phase approach that consists of LOS prediction and a two-stage optimization model is designed. We show that the methods presented in this article can be used as a practical tool to help hospital managers obtain more reasonable elective patient admission scheduling solutions that can improve service quality.

In the era of Industry 5.0 and the emerging metaverses, this study proposes Meta-Services, a novel paradigm that evolves a traditional service system into a multistage ecosystem. In the proposed paradigm, we develop a parallel intelligence-based management framework that enables stakeholders to control and manage the complex dynamics of service delivery in the metaverse. The management framework involves the following three stages: 1) service systems in metaverses; 2) elementary parallel service systems; and 3) parallel intelligence empowered ecosystems. The third stage evolves from the first two stages by incorporating the approach of “designer in the loop”, contributing to more human-centered, adaptive, and resilient service systems. Moreover, a case study in the retail sector is provided to validate the effectiveness of the proposed paradigm and show its practical potential.

In this article, we study the design and innovation of smarter service systems considering the deep integration of real service systems and virtual service elements. Specifically, we consider the perspective of life-cycle service system design and innovation that must consider the “social space” factors, such as stakeholders’ behaviors, which increases the complexity of service cyber–physical–social systems (CPSSs). We propose a new methodology framework, the semiparallel service system (SPSS) framework, following the CPSS architecture and artificial societies, computational experiments, and parallel execution (ACP) methodologies emerging in complex system modeling. We design two decomposition-then-aggregation loops to reduce the complexity of service systems innovation. In the first loop, we expand the Vee model of systems engineering to the N model to show how to develop an innovative service system step by step. In the second loop, the SPSS, in which the hybrid service system is parallel with the real service system, maintains the continuous innovations of the system. Following our framework, we can design prototypes, perform experiments, and run the hybrid service system. We use two cases, an Internet hospital platform design and a new service design in a customs bonded warehouse, to demonstrate the advantages and potential of our proposed methodology. Finally, we discuss the future research challenges and directions.

Humans are an important component of human-machine systems. A better understanding of the role and the status of humans can facilitate and improve the overall human-machine system performance, as well as ensure the well-being of humans. In human factors and ergonomics (HF/E), conventional methods of human performance evaluation usually require the efforts of trained personnel for data collection, data analysis, and the explanation of results. There is an emerging need for a novel and cost-efficient method of assessing human work and status in various systems. The development of wearable technologies has improved the potential for developing a smarter and automatic solution to performing relevant evaluations. In this paper, the authors conduct a scoping review of the studies published from 2001 to 2017 that focused on wearable sensor technology and propose a framework to summarize the research topics. The main steps in the framework include data collection, data processing, and system feedback. Specifically, considering the gaps between current HF/E studies of wearables and the available resources, the authors conducted a detailed review of the application of wearables in HF/E human work evaluation. The opportunities and challenges of introducing wearable sensors into HF/E evaluation are discussed.

Long Short-Term Memory (LSTM) network and Gated Recurrent Units (GRU) network are two widely-used gated Recurrent Neural Network (RNN) architectures. Both of them usually have a huge model size and require a long time to be trained. In this paper, we first propose a singular value decomposition (SVD) based approach for fast training of LSTM. Then, the factorized model and SVD based training approach are proposed for the GRU network, which adaptively choose the rank parameter for the matrix factorization model and reduce the training time and parameters of the gated RNNs. Experiments are carried out on the image classification and sentiment classification tasks using datasets MNIST and IMDB, respectively. The results show that the proposed LSTM-SVD approach achieves up to 3.9X speedup compared with training the original LSTM model, without loss of accuracy. The approaches for training the GRU network also have about 2X speedup. And, with the factorized models the quantity of RNN cell parameters can be significantly reduced by more than 10X.

To keep pace with the developments in medical informatics, health medical data is being collected continually. But, owing to the diversity of its categories and sources, medical data has become so complicated in many hospitals that it now needs a clinical decision support (CDS) system for its management. To effectively utilize the accumulating health data, we propose a CDS framework that can integrate heterogeneous health data from different sources such as laboratory test results, basic information of patients, and health records into a consolidated representation of features of all patients. Using the electronic health medical data so created, multilabel classification was employed to recommend a list of diseases and thus assist physicians in diagnosing or treating their patients' health issues more efficiently. Once the physician diagnoses the disease of a patient, the next step is to consider the likely complications of that disease, which can lead to more diseases. Previous studies reveal that correlations do exist among some diseases. Considering these correlations, a k-nearest neighbors algorithm is improved for multilabel learning by using correlations among labels (CML-kNN). The CML- kNN algorithm first exploits the dependence between every two labels to update the origin label matrix and then performs multilabel learning to estimate the probabilities of labels by using the integrated features. Finally, it recommends the top N diseases to the physicians. Experimental results on real health medical data establish the effectiveness and practicability of the proposed CDS framework.

To better solve the passenger assignment problem, which is a subproblem of the transit network optimization problem, we build an artificial urban transit system (AUTS) and adopt a day-to-day learning mechanism to describe passengers' route and departure-time-choice behaviors. With the support of AUTS to handle the lower level assignment problem, we are able to solve the upper level transit network design problem. Compared with other bilevel models, our approach better accommodates passengers' dynamic learning behavior and their heterogeneity. Based on AUTS, we solve the frequency optimization problem and compare the results with an analytical method. We also perform some numerical experiments on AUTS and discover some interesting issues on the capacity of public transportation system and passengers' heterogeneity.

An urban transit system usually consists of several modes, including busses, streetcars, a subway, and light rail. Unfortunately, coordination among different modes remains a challenging problem. Difficulties arise when modifying the transit network structure on a strategic level or when synchronizing timetables on a tactical level. Traditional transit network design and timetabling intend to solve a network-optimization problem based on static origin-destination (OD) information, with passenger assignment as a subproblem. In this paper, we propose an artificial urban transit system (AUTS) based on agent-based modeling and simulation. With AUTS, which is a special type of artificial transportation system (ATS), we are able to dynamically model the passenger's behavior and route choice and use the system to predict transit demand on a simplified transit network. The AUTS has the following important potential applications: forecasting transit flow; setting key parameters for urban transit networks - such as service frequencies and the capacity of subway trains - evaluating alternative modifications to subway rail and bus routes; and predicting the impact of special/emergency events to the transit network. We create a demonstration system of the Beijing transit network and present its applications in experiments.