While Tri-plane representation has greatly advanced the development of 3D generative models, problems rooted in its inherent structure, such as multi-face artifacts caused by sharing the same features in symmetric regions, limit its ability to generate complete 360° views. In this paper, we propose CylinderPlane, a novel representation based on the cylindrical coordinate system, to achieve high-quality, artifact-free panoramic image synthesis. Unlike the inevitable feature entanglement in the Cartesian coordinate-based representation, the cylindrical coordinate system explicitly disentangles features at different angles. Consequently, our representation effectively eliminates feature ambiguity and ensures multi-view consistency across full 360°. We further develop a nested cylinder representation that combines cylinder planes of varying radii to achieve multi-scale feature fusion. This design not only addresses the limitations of Tri-plane in modeling complex geometries and varying resolutions, but also mitigates the polar discontinuity inherent in a single cylinder plane. Moreover, our versatile representation can be seamlessly integrated into various generative frameworks and rendering pipelines. Extensive experiments on both synthetic datasets and unstructured in-the-wild images demonstrate that our representation outperforms the existing methods.

Despite recent advances in dynamic scene reconstruction, challenges from imbalanced camera distribution and inaccurate pose estimation in real-world datasets still persist, undermining the spatiotemporal consistency of reconstruction. In this paper, we propose HybridPlane, a novel representation that leverages the complementary advantages of cylindrical and Cartesian coordinate systems to achieve high-quality dynamic scene synthesis. Unlike Cartesian projection, which shares identical features in symmetric regions, cylindrical projection explicitly disentangles features from different viewpoints, thereby improving robustness against imbalanced camera distributions. Moreover, the synergy between these two coordinate systems in both projection and representational capacity enhances the model's ability to capture complex motions and fine-grained details. We further adopt the dynamic positional encoding strategy to enhance the smoothness of temporal interpolation under inaccurate camera poses by progressively regulating high-frequency signals without incurring additional computational overhead. Extensive experiments demonstrate that our versatile representation can be seamlessly integrated into various rendering pipelines, outperforming the previous methods in reconstruction quality while reducing computational and memory costs by approximately one-third.

Distributed sparse deep learning has been widely used in many Internet-scale applications. Network communication is one of the major hurdles for training performance. In-network gradient aggregation on programmable switches is a promising solution for speeding up the performance. Nevertheless, existing in-network aggregation solutions are designed for the dense deep training, and fall short when used for the sparse training. To address this gap, we present Zebra based on our key observation on the extremely biased update frequency of parameters in distributed sparse deep training. Specifically, Zebra offloads only the aggregation for “hot” parameters that are updated frequently onto programmable switches. To enable this offloading and achieve high aggregation throughput, we propose solutions to address the challenges related to hot parameter identification, parameter orchestration and gradient aggregation as well as system reliability. We implemented Zebra on Intel Tofino switches and integrated it with PS-lite. Finally, we evaluate Zebra's performance through extensive experiments and show that it can speed up the gradient aggregation by$1.5 \sim 4 \times$and the end-to-end performance by$1.4 \sim 2.6 \times$.

With the rapid development and popularization of block chain technology, some social platforms based on block chain technology have emerged. Due to the characteristics of block chain technology, the social behaviors and ways of users on the network have been affected, thus changing the process of information transmission. Based on the characteristics of information transmission in blockchain social networks, this paper proposes a new information transmission model by comprehensively considering the influence of factors such as blockchain incentive mechanism, node intimacy in traditional social networks and individual differences on the state transition probability. A new voting node has been added to the model, representing users on the blockchain network who are affected by the incentive mechanism. In the simulation experiment, by adjusting the weight of each factor, the influence on the number of spreaders in the information transmission model is analyzed. Moreover, by comparing the model proposed in this paper with the traditional social network model, the number of propagation nodes is reduced by 17.32%. The experimental results show that the blockchain environment can effectively curb the spread and spread of false news, and achieve a good ecological network public opinion communication environment.

Nowadays, online shopping has gradually become the mainstream, and the importance of user comments has gradually increased. Due to the operation of “playing” and “controlling comments”, the fake “brainless blowing” comments that overtout the products endorsed by stars have poured out in large numbers. These fake reviews are easy to lead users to make wrong purchase decisions, damage the interests of users, and also easily cause nonfan group users’ aversion, affecting the promotion and sales of merchants. Therefore, it becomes very important to identify fake comments from a large number of comments, but it is very difficult for users to identify fake comments quickly and accurately in a large number of comments. In response to this problem, this paper collects the semantic and emotional characteristics of the text content of user reviews, combines the behavior characteristics of users when making purchase decisions, and builds a false recognition model for user reviews based on LSTM (Long Short-term Memory) to assist users in identifying false reviews. So that users can make accurate decisions and purchase products that meet their needs. Based on the F1 value, accuracy rate, and recall rate as evaluation indicators, after comparison experiments with eight commonly used models, the algorithm in this paper has the best effect.

User preference modeling from heterogeneous implicit feedback (i.e. various user behaviors) is a key issue in recommender systems, especially when homogeneous implicit feedback is sparse or even not available. But different types of implicit feedback have different confidence in user preference indication, which bring difficulties in building accurate user preference models in recommender system.In this article, we propose to model user preference profiles based on heterogeneous implicit feedback from a new perspective, in which preference confidence from different user implicit feedback should be measured in a unified way. Specifically, a preference confidence transfer learning algorithm based on multiple types of user behavior is designed. We assume that preference confidence can be transferred between target behavior and auxiliary behavior based on the features of user behavior. Then we integrated the assumption into a Latent Factor Model with confidence. Furthermore, we conduct extensive empirical studies with various baseline methods on two real public datasets and find that our algorithm can perform significantly more accurate than the other methods.

As the scale of online shopping users continues expanding, comments and feedback to users of their opinions after purchase and use are of great significance to assist users in purchasing decisions. In order to encourage users to actively comment, some websites provide comment tags for users to choose. These labels only summarize the common features of the products, but do not consider the differences among different products, it is even more unable to meet the personalized expression needs of user comments.. Based on the above reasons, the paper proposes an automatic generation algorithm of comment tags oriented to the personalized needs of users. This algorithm takes a single user as the research object, combines the Seq2Seq (Sequence to sequence learning) model and the part-of-speech-syntax features analysis algorithm to comment on users analyze, capture the description style and product characteristics of a single user in the comment language, combine the user's personalized language with the specific product characteristics, and generate comment tags from the two perspectives of user needs and product characteristics. The paper verifies the effectiveness of the algorithm through real data, compared with the existing algorithm, the algorithm in this paper is more suitable for user needs.

Locality Sensitive Hashing (LSH) is widely adopted to index similar data in high-dimensional space for approximate nearest neighbor search. Demanding applications (e.g. web search) mean that LSH must exhibit low response times and high throughput. To achieve this, they tend to load balance between multiple machines. However, as the scale of concurrent queries and the volume of data grow, large numbers of index messages are required. Hence, the network is a key bottleneck. To address this gap, we propose NetSHa, which exploits the computational capacity of programmable switches. Specifically, we introduce a heuristic sort-reduce approach to drop potentially poor candidate answers while preserving search quality. Then, NetSHa aggregates good candidate answers from different index messages when transmitting them. Through this, it reduces the network communication cost. Furthermore, we introduce a best-effort replacement mechanism to improve its concurrency. We implement NetSHa on a Barefoot Tofino programmable switch and evaluate it using 7 real-world datasets. The experimental results show that NetSHa reduces the packet volume by$4\sim 10$times and improves the search efficiency by least 3× in comparison with typical LSH-based distributed search frameworks.

Scheduling and resource allocation in clouds is asked to harness the power of the underlying resource pool, and thus service providers can meet Quality of Service (QoS) requirements of tenants specified in Service Level agreements (SLAs). Improving resource allocation ensures that all tenants will receive fairer access to system resources, which improves overall utilisation and throughput and reduces traffic in an over-crowded system. Moreover, real-time applications and services require critical deadlines in order to guarantee QoS. A growing number of data-intensive applications also drive the optimisation of scheduling through the utilisation of data locality in which the scheduler attempts to locate a task and ensure that the task's relevant data is on the same server that the task is being deployed to. Choosing suitable scheduling mechanisms for running of various applications that support multi-tenancy has consistently been a major challenge. This work proposes a new adaptive Deadline constrained and Data locality aware Dynamic Scheduling Framework - 3DSF - that orchestrates different schedulers based on varied requirements. This framework considers users' deadline-based QoS requirements, a cloud system's performance and a method of resource allocation to improve resource utilisation, system throughput, reduce the completion time of jobs, and better meet their QoS requirements. Our 3DSF contains: (1) a real-time, preemptive, deadline constrained job scheduler, (2) an optimised data locality aware scheduler, (3) an improved Dominant Resource Fairness (DRF) greedy resource allocation approach and (4) an adaptive suite to integrate all above-mentioned schedulers together.

Bad code smells (also named as code smells) are symptoms of poor design choices in implementation. Existing increases the likelihood of subsequent changes (i.e., change-proness). However, to the best of our knowledge, no prior studies have leveraged smell-based metrics to predict particular change type (i.e., structural changes). Moreover, when evaluating the effectiveness of smell-based metrics in structural change-proneness prediction, none of existing studies take into account of the effort inspecting those change-prone source code. In this paper, we consider five smell-based metrics for effort-aware structural change-proneness prediction and compare these metrics with a baseline of well-known CK metrics in predicting particular categories of change types. Specifically, we first employ univariate logistic regression to analyze the correlation between each smell-based metric and structural change-proneness. Then, we build multivariate prediction models to examine the effectiveness of smell-based metrics in effort-aware structural change-proneness prediction when used alone and used together with the baseline metrics, respectively. Our experiments are conducted on six Java open-source projects with up to 60 versions and results indicate that: (1) all smell-based metrics are significantly related to structural change-proneness, except metric ANOS in hive and SCM in camel after removing confounding effect of file size; (2) in most cases, smell-based metrics outperform the baseline metrics in predicting structural change-proneness; and (3) when used together with the baseline metrics, the smell-based metrics are more effective to predict change-prone files with being aware of inspection effort.

In this paper we propose a new user purchase preference model based on their implicit feedback behavior. We analyze user behavior data to seek their purchase preference signals. We find that if a user has more purchase preference on a certain item he would tend to browse it for more times. It gives us an important inspiration that, not only purchasing behavior but also other types of implicit feedback like browsing behavior, can indicate user purchase preference. We further find that user purchase preference signals also exist in the browsing behavior of item categories. Therefore, when we want to predict user purchase preference for certain items, we can integrate these behavior types into our user preference model by converting such preference signals into numerical values. We evaluate our model on a real-world dataset from a shopping site in China. Results further validate that user purchase preference model in our paper can capture more and accurate user purchase preference information from implicit feedback and greatly improves the performance of user purchase prediction.

Code smells are used to describe the bad structures in the source code, which could hinder software maintainability, understandability and changeability. Nowadays, scholars mainly focus on the impact of smell on textual change-proneness. However, in comparison to textual changes, structural changes could better reveal the change nature. In practice, not all code change types are equally important in terms of change risk severity levels, and software developers are more interested in particular changes relevant to their current tasks. Therefore, we investigate the relationship between smells and fine-grained structural change-proneness to solve these issues. Our experiment was conducted on 11 typical open source projects. We first employed Fishers exact test and Mann–Whitney test to explore whether smelly files (affected by at least one smell type) had higher structural change-proneness than other files, and whether files with more smell instances are more likely to undergo structural changes, respectively. Multivariate logistic regression model was built to study the relation between each kind of smell and change-proneness with respect to five change categories. Our results showed that: (1) in most cases, smelly files were more prone to structural changes and files with more smell instances tend to undergo higher structural changes; (2) quite a few smell types were related to structural change-proneness, particularly, Refused Parent Bequest (RPB), Message Chains (MCH), Divergent Change (DIVC), Feature Envy (FE) and Shotgun Surgery (SS) increased structural changes for some change categories. However, when controlling the file size Lines of Code (LOC), significant change-proneness of some smells disappeared or the magnitude of significance decreased more or less.

Summary Most of the existed researches of educational service quality are concentrated on the definition of service quality in education, measurement methods and the establishment of scales, or on student perception of service quality and its implication, while researches on the causes of educational service quality are quite scarce. In this study, the discussion on the causes of educational service quality was complemented from the perspective of social network, especially through analysis at the team level. A total of 478 copies of social network questionnaires were collected from 15 educational teams. Meanwhile, targeting at the educational institutions' customers, the service quality survey was conducted with 1,487 copies of valid samples in total. The results indicate that individual‐level social capital in the form of trust relations positively influences educational service quality. Among the group level factors, group centrality and number of cliques have a significant impact on group members' service quality, which also moderate the effect of trust on service quality.

In this paper, we propose a time-sensitive probability skyline (TPS) approach to recommend services with uncertainty. We project services to n-dimensional data space and recommend services in TPS. Experimental evaluation on real data shows the great performance of TPS in service recommendation by comparing the experiment result with results of other approaches.