Consumer wearable devices like smartwatches enable real-time tracking of vital signs with various sensors. Accordingly, experts may leverage smart home techniques for treating sleep disorders by targeting specific sleep stages. To facilitate this scenario, this paper focuses on sleep stage classification based on body movement and heart rate signals detected by wearables in real time. Due to their limited battery capacity, it is crucial to balance the trade-off between power efficiency and classification accuracy. To address the problem, inspired by multi-tasking, we propose COPS, an innovative framework that includes a power-efficient shallow classifier for simple cases and a deep classifier for complex instances. COPS introduces an intelligent switch, CESwitch, to determine a confidence score that directs input to either the shallow or deep classifier. By selectively activating the shallow classifier, the overall expected power consumption could be lower. Two strategies of CESwitch, namely Confidence Delegation and Agreement Verification, are proposed and examined. Notably, both COPS and CESwitch can be seamlessly integrated with existing deep sleep stage classifiers. Comprehensive experimental results on two real datasets manifest that COPS outperforms state-of-the-art lightweight and deep sleep stage classifiers by reducing 77.8% computational cost in terms of FLOPs with only 1.9% accuracy drop. Moreover, adapting to an existing deep model saves up to 32.7% in FLOPs compared to its original architecture.

Live stream platforms have gained popularity in light of emerging social media platforms. Unlike traditional on-demand video platforms, viewers and streamers on the live stream platforms are able to interact in real-time, and this makes viewer interests and live stream topics mutually affect each other on the fly, which is the unique co-evolution phenomenon on live stream platforms. In this paper, we make the first attempt to introduce a novel next-topic recommendation problem for the streamers, LSNR, which incorporates the co-evolution phenomenon. A novel framework CENTR introducing the Co-evolutionary Sequence Embedding Structure that captures the temporal relations of viewer interests and live stream topic sequences with two stacks of self-attention layers is proposed. Instead of learning the sequences individually, a novel dual self-attention mechanism is designed to model interactions between the sequences. The dual self-attention includes two modules, LCA and LVA, to leverage viewer loyalty to improve efficiency and flexibility. Finally, to facilitate cold-start recommendations for new streamers, a collaborative diffusion mechanism is implemented to improve a meta learner. Through the experiments in real datasets, CENTR outperforms state-of-the-art recommender systems in both regular and cold-start scenarios.

In contrast to traditional online videos, live multi-streaming supports real-time social interactions between multiple streamers and viewers, such as donations. However, donation and multi-streaming channel recommendations are challenging due to complicated streamer and viewer relations, asymmetric communications, and the tradeoff between personal interests and group interactions. In this paper, we introduce Multi-Stream Party (MSP) and formulate a new multi-streaming recommendation problem, called Donation and MSP Recommendation (DAMRec). We propose Multi-stream Party Recommender System (MARS) to extract latent features via socio-temporal coupled donation-response tensor factorization for donation and MSP recommendations. Experimental results on Twitch and Douyu manifest that MARS significantly outperforms existing recommenders by at least 38.8% in terms of hit ratio and mean average precision.

Shopping in VR malls has been regarded as a paradigm shift for E-commerce, but most of the conventional VR shopping platforms are designed for a single user. In this paper, we envisage a scenario of VR group shopping, which brings major advantages over conventional group shopping in brick-and-mortar stores and Web shopping: 1) configure flexible display of items and partitioning of subgroups to address individual interests in the group, and 2) support social interactions in the subgroups to boost sales. Accordingly, we formulate the Social-aware VR Group-Item Configuration (SVGIC) problem to configure a set of displayed items for flexibly partitioned subgroups of users in VR group shopping. We prove SVGIC is APX-hard and also NP-hard to approximate within [EQUATION]. We design a 4-approximation algorithm based on the idea of Co-display Subgroup Formation (CSF) to configure proper items for display to different subgroups of friends. Experimental results on real VR datasets and a user study with hTC VIVE manifest that our algorithms outperform baseline approaches by at least 30.1% of solution quality.

With the rapid advent of Virtual Reality (VR) technology and virtual tour applications, there is a research need on spatial queries tailored for simultaneous movements in both the physical and virtual worlds. Traditional spatial queries, designed mainly for one world, do not consider the entangled dual worlds in VR. In this paper, we first investigate the fundamental shortest-path query in VR as the building block for spatial queries, aiming to avoid hitting boundaries and obstacles in the physical environment by leveraging Redirected Walking (RW) in Computer Graphics. Specifically, we first formulate Dual-world Redirected-walking Obstacle-free Path (DROP) to find the minimum-distance path in the virtual world, which is constrained by the RW cost in the physical world to ensure immersive experience in VR. We prove DROP is NP-hard and design a fully polynomial-time approximation scheme, Dual Entangled World Navigation (DEWN), by finding Minimum Immersion Loss Range (MIL Range). Afterward, we show that the existing spatial query algorithms and index structures can leverage DEWN as a building block to support kNN and range queries in the dual worlds of VR. Experimental results and a user study with implementation in HTC VIVE manifest that DEWN outperforms the baselines with smoother RW operations in various VR scenarios.

Recent technological advent in virtual reality (VR) has attracted a lot of attention to the VR shopping, which thus far is designed for a single user. In this paper, we envision the scenario of VR group shopping, where VR supports: 1) flexible display of items to address diverse personal preferences, and 2) convenient view switching between personal and group views to foster social interactions. We formulate the Multiview-Enabled Configuration Recommendation (MECR) problem to rank a set of displayed items for a VR shopping user. We design the Multiview-Enabled Configuration Ranking System (MEIRS) that first extracts discriminative features based on Marketing theories and then introduces a new coupled tensor factorization model to learn the representation of users, Multi-View Display (MVD) configurations, and multiple feedback with content features. Experimental results manifest that the proposed approach outperforms personalized recommendations and group recommendations by at least 30.8% in large-scale datasets and 63.3% in the user study in terms of hit ratio and mean average precision.

Online advertising has been all the rage these years. Budget control and traffic prediction turn out to be important issues for the demand-side platforms (DSPs). However, DSPs cannot easily grab the information of audiences and media platforms. Although DSPs might have the information immediately, it is still hard to response the request of advertisements in real-time due to the high volume of features. Therefore, we propose a method predicting traffic of requests from perspective of DSPs. The features we used are simple to be extracted from historical data. The prediction model we chose is regression model with closed-form solution. Both the features and regression model make our prediction adaptive in real-time systems. Our method can detect traffic anomalies and prevent it from overwhelming prediction. Moreover, our method can also keep pace of the trend. Experiment results show that our method's error rate of prediction is about 0.9% in total, and 10% per time unit.