Multi-behavior recommendation systems enhance prediction accuracy for target behavior (e.g., purchase) by integrating auxiliary behaviors such as viewing and adding to cart. However, existing models face two fundamental challenges regarding data characteristics: On one hand, they typically treat all auxiliary behaviors equally or apply simple weighting, lacking effective mechanisms to evaluate and filter inherent noise, leading to contaminated feature representations. On the other hand, when fusing sparse target features with abundant auxiliary features, the critical target signal is easily diluted, losing its dominant role in the final prediction. To address this, we propose a Target-Enhanced Gated Transformer model TEGT. Its core innovations include: a behavior-adaptive gating module that filters source-level noise through nonlinear transformations and hard thresholding while generating importance weights; a target-guided dual-modulation attention mechanism utilizes target behavior as queries to retrieve semantically relevant auxiliary signals, then applies secondary modulation by combining gated weights to ensure both noise resistance and target dominance in the fusion process; a lightweight collaborative semantic enhancement module clusters the fused representations and employs cluster-center contrastive learning to explicitly amplify collaborative signals under sparse target behavior. Extensive experiments on three real-world datasets show that TEGT consistently outperforms state-of-the-art baselines, it achieves remarkable improvements of up to 9.73% in Recall@10 and 9.16% in NDCG@10.

Existing LLM-driven recommendation systems (RS) suffer from over-reliance on a single pre-trained model, which limits adaptability across diverse scenarios due to differences in large language models' strengths in semantics, knowledge, and reasoning. To address this issue, we propose AMLrec (Adaptive Multi-LLM Recommendation), a dynamic routing-based adaptive multi-LLM collaboration framework that unifies two dominant paradigms—LLM as Recommender and LLM + Recommender—through decision knowledge complementation. For each user or item, a lightweight encoder generates embeddings that are compared with learnable LLM prototypes using cosine similarity to select the most suitable models. In the first paradigm, selected LLMs generate recommendations via structured prompts, and their outputs are aggregated to form the final recommendation list. In the second paradigm, chosen LLMs produce semantic embeddings, which are fused with learnable embeddings after PCA-based dimensionality reduction and aligned using a lightweight adapter to bridge distribution gaps. Notably, AMLrec does not require fine-tuning of the underlying LLMs, significantly reducing computational overhead. Experiments on real-world datasets demonstrate that the proposed approach consistently outperforms single-LLM baselines across all evaluation metrics, validating its effectiveness. The main contributions of this work are threefold: introducing dynamic routing for multi-LLM recommendation system collaboration, proposing a unified architecture that harmonizes both paradigms, and enabling efficient adaptation without LLM fine-tuning. The code is available at https://github.com/Jiale-12138/AMLrec.

Multi-behavior recommender systems improve prediction accuracy of target behaviors (e.g., purchases) by integrating auxiliary behaviors (e.g., page views). However, existing models face two key limitations: (1) Static propagation mechanisms and inflexible dependency modeling fail to capture dynamic changes in user preferences and cascading relationships between behaviors; (2) Sparse target behavior data usually leads to excessive influence of auxiliary signals, which degrades recommendation quality. To address these challenges, we propose the Dynamic Dependency-Aware Collaborative Contrastive Learning Multi-Behavior Recommendation Model, MBDCC. MBDCC has two dedicated modules: (1) Behavioral gating cascade and cross-attention fusion module, which dynamically models cascading dependencies between behaviors through learnable gate control transfer units controlled by behavioral attributes. This replaces static propagation with adaptive feature flow regulation, capturing evolving user preferences. Meanwhile, it uses a target-guided cross-attention mechanism to selectively fuse semantically relevant auxiliary signals using the target behavior as a query, addressing inflexible cross-behavioral dependency modeling; (2) Collaborative semantic enhancement module, it constructs a user similarity measure matrix based on co-occurrence frequency of interaction items in target behavior, and clusters nodes using a hybrid clustering strategy. By introducing contrastive learning between nodes and their clustering centers, the collaborative semantic information between similar nodes under the target behavior is effectively captured and amplified, alleviating the challenge of sparse target interaction data. Extensive experiments on three real-world datasets show that MBDCC consistently outperforms state-of-the-art baselines, it achieves remarkable improvements of up to 6.84% in Recall@10 and 5.18% in NDCG@50. Moreover, ablation experiments further demonstrate the correctness of our motivation and the necessity of the various modules of the MBDCC model.

The forecasting of Click-Through Rates (CTR) is crucial in online advertising and recommendation systems, significantly impacting revenue and user experience. Traditional CTR models rely on simple feature representations and static interactions, making it difficult to accurately capture complex user behaviors. Although deep learning has made progress in feature processing, it still faces challenges in adapting to dynamic environments and maintaining model stability. This paper introduces the Dynamic Fusion Prediction Network (DFPN), which innovatively employs a Dynamic Context Encoder (DCE) and FusionGate mechanism to enhance feature representation and interaction. The DCE adapts to varying contexts, allowing for a more personalized representation of features. FusionGate further improves feature importance identification, contributing to enhanced prediction accuracy and robustness. Additionally, dropout and layer normalization are integrated to mitigate overfitting and improve generalization. The DFPN is designed to be orthogonal to existing CTR models such as FM and DeepFM, enabling flexible integration for performance enhancement. Experiments on public benchmarks demonstrate that DFPN significantly improves accuracy, generalization, and stability, confirming its superior performance in dynamic environments.

Image super-resolution is a long-standing challenge in computer vision aimed at reconstructing high-quality images from low-quality images. In recent years, the rapid development of deep learning has driven significant advances in this field, but there are still some limitations of existing methods. They are difficult to take into account the preservation of local details when capturing global information, resulting in overly smooth reconstruction results and insufficient recovery of high-frequency details. And the recovery of high-frequency information is crucial in image reconstruction because it directly determines the perceived quality and visual fidelity of the image. To effectively address these challenges, we propose a dynamic feature fusion network DFFNet for single-image super-resolution to investigate the interaction of local and non-local information. DFFNet introduces non-local information for information interaction based on window self-attention to make up for the defect of information isolation between windows, increase the global feeling field, and realize the effective fusion of local and non-local. Meanwhile, a channel with multi-scale spatial attention module is introduced to further refine the key features and improve the image super-resolution effect. Extensive experimental evaluations show that our network achieves better performance on image super-resolution compared to existing methods.

Multibehavioral recommender systems improve the prediction accuracy of target behaviors (e.g., purchase) by integrating auxiliary user behaviors (e.g., page view), but often face the challenge of sparse data on target behaviors. Although contrastive learning can effectively alleviate this problem, existing models still have limitations: they reduce behavioral relationships to linear combinations, ignoring users’ unique behavioral combination modes for different items; meanwhile, when constructing contrastive views, they fail to adequately consider the data imbalance between auxiliary and target behaviors, resulting in learned features biased toward auxiliary behaviors. Consequently, we introduced a multibehavioral recommendation model founded on dual mode contrast learning (DMCL). DMCL innovatively defines behavior combination modes and generates interaction graphs from two perspectives: behavior combination and cascading, and designs a specialized graph encoder for each graph to learn node embedding. By introducing adversarial constraints and contrastive learning, DMCL enhances the complementary nature of node embeddings, and at the same time utilizes interaction graphs in different modes to construct a contrast view, which makes full use of the limited interaction information of the target behavior, and effectively mitigates the feature bias problem. Finally, DMCL combines multilayer perceptron to predict recommendation scores. Our comprehensive experiments on three real datasets demonstrate significant improvements in Recall and NDCG metrics, achieving increases of 2.4–10.1% and 2.6–10.3%, respectively, over current state-of-the-art methods.

Margin distribution has been proven to play a crucial role in improving generalization ability. In recent studies, many methods are designed using large margin distribution machine (LDM), which combines margin distribution with support vector machine (SVM), such that a better performance can be achieved. However, these methods are usually proposed based on single-view data and ignore the connection between different views. In this article, we propose a new multiview margin distribution model, called MVLDM, which constructs both multiview margin mean and variance. Besides, a framework is proposed to achieve multiview learning (MVL). MVLDM provides a new way to explore the utilization of complementary information in MVL from the perspective of margin distribution and satisfies both the consistency principle and the complementarity principle. In the theoretical analysis, we used Rademacher complexity theory to analyze the consistency error bound and generalization error bound of the MVLDM. In the experiments, we constructed a new performance metric, the view consistency rate (VCR), for the characteristics of multiview data. The effectiveness of MVLDM was evaluated using both VCR and other traditional performance metrics. The experimental results show that MVLDM is superior to other benchmark methods.

Caching recommended contents at the network edge can effectively alleviate the traffic pressure of the backbone network and significantly improve user experience. However, highly personalized and precise recommendations often rely on leveraging more user request records, raising serious privacy concerns. Existing recommendation-aware edge caching mechanisms typically apply a fixed level of privacy protection, without considering the personalized privacy of users. This one-size-fits-all approach often introduces significant noise, adversely impacting cache hit ratio (CHR). In this work, we propose a differential privacy-based edge caching framework supporting personalized privacy-preserving to address these challenges. We formulate a CHR maximization problem under personalized privacy constraints and reveal the NP-completeness of the problem with a rigorous mathematical proof. Subsequently, we mathematically model the relationship between personalized privacy and user preference distortion, analyzing its impact on recommendations and user requests. To solve it, we introduce an efficient heuristic algorithm named the Backhaul Traffic-Aware Caching Algorithm. This algorithm utilizes backhaul traffic as a feedback signal to make accurate caching decisions, enabling adaptive optimization of caching decisions by perceiving the impact of noise and low-quality recommendations. Extensive experiments on two typical real-world datasets validate the effectiveness of our framework, demonstrating its ability to enhance privacy protection while simultaneously improving CHR.

Point-of-Interest (POI) recommendation aims to enhance user experiences by offering tailored suggestions for places that align with their interests. Existing approaches to POI recommendation suffer from the data sparsity problem and fail to capture the temporal dependency among users’ check-in history. In this paper, we propose a novel POI recommendation approach based on graph neural networks. We encode each user’s historical check-ins using a graph structure, which preserves the temporal order of the users’ history and employs a graph neural network to learn users’ preferences. In addition, we apply self-supervised learning to facilitate data-efficient learning under the sparse setting and describe a curriculum-learning-based training strategy for POI recommendation. Experiments on two real-world, large-scale benchmark datasets demonstrate that the proposed model has strong performance, outperforming a number of state-of-the-art baseline alternatives by a wide margin. An extensive ablation study further verifies the effectiveness of each of the introduced modules.

Cloud detection is a crucial step in the preprocessing of satellite remote sensing images. Existing methods tend to have misjudgments when dealing with specific scenarios, such as challenges in distinguishing thin clouds from the background and addressing missing cloud boundaries. To solve this problem, we designed a novel Spatial–Frequency Domain Feature Enhancement Block (SFDE) embedded in a U-shaped network called SFDE-net. SFDE consists of three units: the Dual Frequency Feature Unit (DFF), the Spatial Domain Feature Unit (SDF), and the Cross-Domain Feature Fusion Unit (CDF). DFF globally learns the boundaries and overall structure of clouds in the frequency domain, SDF captures fine-grained information in the spatial domain, and CDF adaptively fuses features from both DFF and SDF. Our method’s effectiveness was evaluated on two public datasets, GF-1 WFV and LandSat8. Extensive experiments demonstrated that the proposed SFDE-net achieved accurate detection accuracy and outperformed several state-of-the-art methods.

With the development of the Internet and the application of cloud computing, the number of Web Services with similar functions is increasing. How to recommend high-quality services that satisfy their needs to users based on the quality of service (QoS) prediction results of Web Services is becoming increasingly important. Using the non-functional attributes of Web Services for QoS prediction has become a major research issue. Among the existing researches, methods based on collaborative filtering (CF) are firstly used in the field of QoS prediction. However, due to the sparsity of the call records of Web Services in the real world, traditional CF methods are greatly limited. In addition, the feature interaction of different dimensions and the importance of different feature combinations between the geographical location of users and services are rarely considered at the same time. In order to address the above, this paper proposes a new QoS prediction model that combines CF and feature interaction learning. The model innovatively uses similarity adaptive corrector to correct QoS, employs an attention factorization machine to model low-order feature interaction, and uses deep neural network to model high-order feature interaction. We conducted a comprehensive experiment with real-world datasets and achieved better QoS prediction accuracy.

With the increasing number of web services on the Internet, it has become an important challenge to recommend the optimal service that meets the needs of users. Quality of Service (QoS) plays an important role in service recommendation. Most existing methods ignore the correlation information among QoS attributes, and their prediction accuracy can be further improved. We propose Weight Adaptive Multi-Task Learning method(WAMTL). WAMTL treats the prediction process of a single QoS attribute as a sub-task, simultaneously predicting the values of multiple QoS attributes in parallel. In parallel learning, by making full use of the correlation information among QoS attributes, the prediction process of QoS attributes promotes each other and improves the accuracy of the prediction results. By introducing adaptive sub-task loss weight into the model, the problem of unbalanced learning rate of sub-tasks in multi- task learning method is alleviated. We conducted comprehensive experiments on real observed datasets. The experimental results show that our method significantly outperforms the comparison models in terms of prediction accuracy.

Convolutional neural networks (CNNs) have been successfully applied to music genre classification tasks. With the development of diverse music, genre fusion has become common. Fused music exhibits multiple similar musical features such as rhythm, timbre, and structure, which typically arise from the temporal information in the spectrum. However, traditional CNNs cannot effectively capture temporal information, leading to difficulties in distinguishing fused music. To address this issue, this study proposes a CNN model called MusicNeXt for music genre classification. Its goal is to enhance the feature extraction method to increase focus on musical features, and increase the distinctiveness between different genres, thereby reducing classification result bias. Specifically, we construct the feature extraction module which can fully utilize temporal information, thereby enhancing its focus on music features. It exhibits an improved understanding of the complexity of fused music. Additionally, we introduce a genre-sensitive adjustment layer that strengthens the learning of differences between different genres through within-class angle constraints. This leads to increased distinctiveness between genres and provides interpretability for the classification results. Experimental results demonstrate that our proposed MusicNeXt model outperforms baseline networks and other state-of-the-art methods in music genre classification tasks, without generating category bias in the classification results.

Recently, most studies in the field have focused on integrating reviews behind ratings to improve recommendation performance. However, two main problems remain (1) Most works use a unified data form and the same processing method to address the user and the item reviews, regardless of their essential differences. (2) Most works only adopt simple concatenation operation when constructing user-item interaction, thus ignoring the multilevel relationship between the user and the item, which may lead to suboptimal recommendation performance. In this paper, we propose a novel Asymmetric Multi-Level Interactive Attention Network (AMLIAN) integrating reviews for item recommendation. AMLIAN can predict precise ratings to help the user make better and faster decisions. Specifically, to address the essential difference between the user and the item reviews, AMLIAN uses the asymmetric network to construct user and item features using different data forms (document-level and review-level). To learn more personalized user-item interaction, the user ID and item ID and some processed features of user reviews and item reviews are respectively used for multilevel relationships. Experiments on five real-world datasets show that AMLIAN significantly outperforms state-of-the-art methods.

Click-through rate prediction plays a critical role in many fields, and many efforts are devoted to analyzing item behavior as a way to find the interest preferences of target user. Most previous work has not taken into account and fully explored the basic information, historical behavior and temporal information of similar users, which can result in missing information and lead to inaccurate CTR prediction. To solve the above problem, we propose Item Multi-Information Evolution Network (IMIEN). Firstly, for the basic information of users, we use RNN sequences to capture the evolutionary dynamics of users interested in the target item. Secondly, we use Evolutionary Interest Extraction Block (EIEB) to mine the evolutionary interest of target user and similar users over time. Finally, we introduce temporal information to find recent user groups with similar interests as the target user, which aids in predicting the probability of the target user clicking on the target item. We achieve the best results on five public datasets compared with previous mainstream models, which validates the effectiveness of IMIEN.

CTR prediction is an important task in personalized recommendation and online advertising. In recent years, many models have been applied in CTR prediction systems to imporve the predcition accuracy. Although the current work has achieved results, the diversity of relationships between the different features was ignored: 1) most existing models usually use the vector-product to realize the interaction of each feature pairs, which ignore the potential contexts between the different feature interactions and lack the different interactive behaviors between the same pair of features; 2) most existing models usually only consider the importance of features of the same order, which ignore the weight relationships between the different feature representations.In this paper, we propose a novel deep learning model called Multi-Relational Interaction Network (MRIN), which learns fine-grained feature interactions and captures latent relationships between features. On the one hand, we design two interactive modules that take different interactive methods to learn the feature interactions in the potential contexts. On the other hand, we flexibly apply attention network to further automatically learn the weight relationships between different feature representations. Extensive experiments conducted on two public datasets prove the effectiveness of our proposed model, which has better performance than the state-of-the-art models.

In the era of information explosion, recommender systems are widely studied and applied to explore users’ preferences. Reviews often reflect users’ real thoughts and play an important role in modeling user preferences. However, most existing review-based recommendation methods assign the same weight to words with different levels of importance, which degrades the recommendation performance. To address these problems, we propose a novel model that uses an adjusted cosine similarity to compute the preference matrix of user, then uses an incentive compression network to weight the preference matrix to focus on important local interaction features and achieve user rating prediction of items based on that interaction feature. The experimental results show that our model outperforms previous mainstream recommendation models.

Click-through rate prediction(CTR) is a critical task in recommendation system. Most previous work on modelling the interaction features between user history behavior and target item focused only on mining the single interaction feature between user’s single historical behavior and the target item, or only modeling the user behavior sequence to capture user interest, which may lead to lower prediction accuracy. To solve the above problem, we propose the Deep Interaction Behavioral Feature Network(DIBFN), which is a dual-module approach that focuses on both the interaction features of user interest and the target item and on the interaction features of user’s single historical behavior and the target item. First, the Evolving Interaction Module (EIM), which processes user behavior sequence to explore the user interest over time, captures the interaction features of the user interest with the target item. Then, the Behavioral Feature Module (BFM) is proposed to learn behavior representations from multi-space through multi-head self-attention and use a two-layer attention mechanism to mine the interaction features between user’s single historical behavior and the target item. Experimental results on real datasets show that the proposed DIBFN model outperforms previous main-stream approaches.

Abstract Recently, many deep learning-based models have been successfully applied to click-through rate prediction. However, most previous models focus only on feature-level interactions between a single user behavior and the target item or only treat the user’s historical behavior as a sequence to uncover the hidden interests behind it when mining user interests. This can lead to user interest that evolves over time dynamically being ignored or the interest shown by a single user’s behavior not being exploited. Based on the above problems, we propose evolving interest with feature co-action network (EIFCN). Specifically, we first design user dynamic interest network to treat the user’s historical behavior as a sequence of information, and tap into the user’s hidden interests over time. In this part, we use a multi-head self-attention mechanism to initially process the data and then pass it into the deep learning network. Then a feature co-action network is designed to mine the user’s single behavior and the displayed feature-level interactions of the target item. Experimental results show that the EIFCN model performs better than other models.

In recent years, Variational AutoEncoder (VAE) based methods have made many important achievements in the field of collaborative filtering recommendation system. VAE is a kind of Bayesian model which combines latent variable model with variational inference, but its optimization is often troubled by posterior collapse. By comparing the optimization process of VAE and ordinary autoencoder, we observe that the mismatch between poorly optimized encoder and decoder with too strong characterization capabilities makes it difficult to learn the mapping from the data manifold to the parameterized graph. Since the learning of a posteriori network corresponds to the encoder, we think that the problem of a posteriori collapse can be alleviated by balancing the encoder and decoder better. Therefore, we proposed Hy-VAE, which combines conventional VAE with deterministic autoencoder, and has the advantages of both VAE and deterministic autoencoder. Experiments on three real-world recommendation data sets show that our method alleviates the posterior crash problem in VAE and improves the recommendation performance.

With the development of mobile internet and social platforms, the lifestyle of check-in has become popular in people’s daily life. With the support of highly accurate positioning technology, the social platform has accumulated a large amount of user check-in data. Based on these data, the platform can provide point-of-initerest recommendation services for users.However, existing recommendation algorithms often suffer from some limitations: (1) Recommendation systems using deep learning algorithms require extremely high hardware computing power and can not be deployed on edge devices; (2) General recommendation algorithms do not fully exploit the potential of users to go to new points of interest when making recommendations. Based on the above existing problems, we propose our own collaborative filtering recommendation model: (1) The model contains simple operations and requires very little device computing power; (2) It models users’ activity areas by using a Gaussian distribution model, and then calculates the similarity between users to establish similar user groups. The similarity can be used as the probability of regional transfer of group members. Based on this method, the points-of-interest of similar users are added to the recommendation list for the purpose of exploring new interest points. The experimental results on the well-known dataset Gowalla demonstrate that the proposed approach outperforms other state-of-the-art POI recommendation methods.

With the popularization of mobile devices and the development of satellite positioning technology, it is more convenient to collect the travel data of users. A large amount of user sign-in data provides strong data support for points of interest recommendation. The rapid development of transportation and the improvement of people’s material living standards have made point-of-interest recommendations an urgent need. Because the user’s travel is a continuous behavior with a trajectory. We can dig out important information from the adjacent sign-in records. For example, a large number of users will go to place B after they have been to place A, which shows that the connection between place A and place B is very close. In this paper, the graph embedding method is used to mine users’ favorite degree of places, and the tensor decomposition method is used to mine users’ travel interests, so as to analyze the relationship between users’ travel characteristics and places. Based on the idea of collaborative filtering, we need to find similar user groups in the dataset. In this paper, a method for calculating weighted virtual users is proposed, and similar users are clustered by using Gaussian mixture clustering method to get similar user groups. Experimental results show that the method implemented in this paper has better effect than other methods.

As a representative of nonfunctional features, the Quality-of-Service (QoS) plays an important role in recommending the best service to users. To obtain the QoS value of web services, researchers have proposed many web service QoS prediction methods. However, most existing methods, e.g., the collaborative filtering method, leverage the historical user service invocation information to predict the QoS value, and only consider the QoS information of similar users or services, and ignore the attributes and characteristics of users and services. In this paper, we propose a Deep Feature Interaction Learning method, called DFIL for short, for web service QoS prediction. Specifically, (1) DFIL constructs a feature extraction method to effectively obtain the features of users and services. (2) DFIL trains a novel neural network for QoS prediction, which can accurately find the potential relationship between users and services based on the feature information, and provide good web service recommendation.

With the increasing prevalence of web services on the World Wide Web, a large number of functionally equivalent web services are provided by different providers. Quality-of-Service (QoS), representing the nonfunctional characteristics, plays an important role in dealing with how to recommend the optimal services to users among these candidates. Many existing methods for predicting QoS values of web services show that QoS values are intensively relevant to location due to the great influence of network distance and the internet connection between users and services. In this paper, we propose a novel location-aware feature interaction learning (LAFIL) method for predicting the QoS values of the user-service matrix and then making the recommendation by learning the underlying relation, which is hidden in the features concerning with location information. LAFIL can effectively solve the problems of data sparsity and cold-start by leveraging the location features of both users and services. To evaluate the performance of our proposed method, comprehensive experiments are conducted using a real-world dataset and the results show that our method achieves better QoS prediction accuracy compared to state-of-the-art approaches.

With the rapid development of location-based social networks (LBSNs), point of interest (POI) recommendation has become an important way to meet users' personalized demands. The aim of POI recommendation is to provide personalized recommendation of POIs for mobile users. However, traditional POI recommendation systems cannot satisfy users' personalized demands. The reason is that the traditional POI recommendation system cannot recommend the next POI to a user based on the user's context information. Also, the traditional POI recommendation system provides no real-time guarantee on performance. In this demo, we propose a novel real-time next POI recommendation system named R2SIGTP which provides more personalized real-time recommendation compared with existing ones. Our system has the following advantages: 1) it has real-time performance; 2) it uses a unified approach to integrate geographic and preference information; 3) it considers the feedback of each single user to provide more personalized recommendation. We have implemented our system. R2SIGTP is easy to use and can be used by the mobile terminal's browser to recommend the next POI to the user in real-time based on the automatically identified user location and current time. The experimental results on real-world LBSNs show that R2SIGTP's performance is satisfactory.

Music recommendation systems automatically suggest a track to a user according to the user's preference. Thus, effectiveness information mining approach is necessarily required to analyze and measure the multiple attributes which could affect the users' preference such that more personalized recommendation can be achieved. However, traditional music recommendation systems rarely consider the multiple attributes of users' preference, leading to a misguided recommendation. In this paper, we propose a novel music recommendation approach, named RNDM. RNDM which employs three computational attributes to describe the user's music taste: novelty, diversity and mainstream. Based on these three attributes, a virtual friendship of each user is constructed by using a modified random walk algorithm. Finally, more personalized recommendations can be achieved by exploiting the music preferences of both the target user and his virtual friends. Compared with the state-of-the art music recommendation approaches, the experimental results show that RNDM significantly improves the accuracy of recommendation and provides more personalized music recommendation for users.

Modern embedded processors provide hardware support for cache locking, a mechanism used to facilitate the WCET (Worst-Case Execution Time) calculation of a task. We investigate the problem of integrating task scheduling and cache locking for a set of preemptible tasks with individual release times and deadlines on a multi-core processor with two-level caches. We propose a novel integrated approach that schedules the task set and allocates the locked cache contents of each task to the local caches (L1 caches) and the level-two cache (L2 cache). Our approach consists of three major components, the task scheduler, the L1 cache allocator, and the L2 cache allocator. The task scheduler aims at minimizing the number of task preemptions. The L1 cache allocator converts the interference graph of all the tasks scheduled on each core into a DAG by considering the preemptions between tasks and allocates the L1 cache space to each task. The L2 cache allocator converts the interference graph of all the tasks into a DAG by using a k-longest-path-based graph orientation algorithm and allocates the L2 cache space to each task. Both cache allocators significantly improve the cache utilization for all the caches due to the efficient use of the interference graphs of tasks. We have implemented our approach and compared it with the extended version of the preemption tree-based approach and the static analysis approach without cache locking by using a set of benchmarks from the MRTC WCET benchmark suite and SNU real-time benchmarks. Compared to the extended version of the preemption tree-based approach, the maximum WCRT (Worst Case Response Time) improvement of our approach is 15%. Compared to the static analysis approach, the maximum WCRT improvement of our approach is 37%.

Using mobile sinks to collect sensed data in WSNs (Wireless Sensor Network) is an effective technique for significantly improving the network lifetime. We investigate the problem of collecting sensed data using a mobile sink in a WSN with unreachable regions such that the network lifetime is maximized and the total tour length is minimized, and propose a polynomial-time heuristic, an ILP-based (Integer Linear Programming) heuristic and an MINLP-based (Mixed-Integer Non-Linear Programming) algorithm for constructing a shortest path routing forest for the sensor nodes in unreachable regions, two energy-efficient heuristics for partitioning the sensor nodes in reachable regions into disjoint clusters, and an efficient approach to convert the tour construction problem into a TSP (Travelling Salesman Problem). We have performed extensive simulations on 100 instances with 100, 150, 200, 250 and 300 sensor nodes in an urban area and a forest area. The simulation results show that the average lifetime of all the network instances achieved by the polynomial-time heuristic is 74% of that achieved by the ILP-based heuristic and 65% of that obtained by the MINLP-based algorithm, and our tour construction heuristic significantly outperforms the state-of-the-art tour construction heuristic EMPS.

Caches are widely used in embedded systems to bridge the increasing speed gap between processors and off-chip memory. However, caches make it significantly harder to compute the worst-case execution time (WCET) of a task. To alleviate this problem, cache locking has been proposed. We investigate the WCET-aware I-cache locking problem and propose a novel dynamic I-cache locking heuristic approach for reducing the WCET of a task. For a nonnested loop, our approach aims at selecting a minimum set of memory blocks of the loop as locked cache contents by using the min-cut algorithm. For a loop nest, our approach not only aims at selecting a minimum set of memory blocks of the loop nest as locked cache contents but also finds a good loading point for each selected memory block. We propose two algorithms for finding a good loading point for each selected memory block, a polynomial-time heuristic algorithm and an integer linear programming (ILP)-based algorithm, further reducing the WCET of each loop nest. We have implemented our approach and compared it to two state-of-the-art I-cache locking approaches by using a set of benchmarks from the MRTC benchmark suite. The experimental results show that the polynomial-time heuristic algorithm for finding a good loading point for each selected memory block performs almost equally as well as the ILP-based algorithm. Compared to the partial locking approach proposed in Ding et al. [2012], our approach using the heuristic algorithm achieves the average improvements of 33%, 15%, 9%, 3%, 8%, and 11% for the 256B, 512B, 1KB, 4KB, 8KB, and 16KB caches, respectively. Compared to the dynamic locking approach proposed in Puaut [2006], it achieves the average improvements of 9%, 19%, 18%, 5%, 11%, and 16% for the 256B, 512B, 1KB, 4KB, 8KB, and 16KB caches, respectively.

Caches have been widely used in modern embedded processors to bridge the increasing speed gap between processors and off-chip memory. In real-time embedded systems, computing the Worst-Case Execution Time (WCET) of a task is essential for the task scheduler to construct a valid schedule for a task set. Unfortunately, caches make it much harder to compute the WCET of a task. Cache locking has been proposed to alleviate the timing unpredictability problem caused by caches. In this article, we investigate the following WCET-aware data-cache locking problem for a single task. Given a task, select a set of variables as locked cache contents such that the WCET of the task is minimized. We propose two dynamic full cache-locking approaches. The first formulates the problem as a global Integer Linear Programming (ILP) problem that simultaneously selects a minimum set of memory blocks of variables as locked cache contents and allocates them to the data cache. The second iteratively constructs a subgraph of the Control Flow Graph (CFG) of the task in which the lengths of all the paths are close to the longest path length, uses an ILP formulation to select a minimum set of memory blocks of variables in the subgraph as locked cache contents, and allocates the selected memory blocks to the data cache. We also propose two novel, efficient data-cache allocation algorithms for the global ILP approach and the iterative ILP approach, respectively. We have implemented both approaches and compared them with two state-of-the-art approaches, the longest path-based dynamic cache-locking approach and the static WCET analysis approach without cache locking by using a set of benchmarks from the Mälardalen WCET benchmark suite, SNU real-time benchmarks, and Powerstone benchmarks. Compared to the static WCET analysis approach, the average WCET improvements of the first approach range between 11.4% and 26.4%. Compared to the longest path--based, dynamic cache-locking approach, the average WCET improvements of the first approach range between 5.0% and 15.4%. The second approach performs slightly better than the first approach.

We investigate the latency and network lifetime-aware clustering problem for data collection in a WSN (Wireless Sensor Network) with one base station and propose a novel approach. Our approach consists of a polynomial-time heuristic for constructing clusters, a polynomial-time heuristic and an ILP (Integer Linear Programming) algorithm for constructing a routing tree for inter-cluster communication. We have performed extensive simulations on the network instances of 200, 400, 600, 800 and 1000 sensor nodes with uniform and random distributions, and compared our approach with two state-of-the-art approaches, namely MR-LEACH and DSBCV. Our simulation results show that the average network lifetime improvements of our approach using the ILP algorithm for constructing a routing tree over MR-LEACH and DSBCV are 35% and 62%, respectively, and the performance of the heuristic for constructing a routing tree is comparable to that of the ILP algorithm.

Caches have been extensively used to bridge the increasing speed gap between processors and off-chip memory. However, caches make it much harder to compute the WCET (Worst-Case Execution Time) of a program. Cache locking is an effective technique for overcoming the unpredictability problem of caches. We investigate the WCET aware D-cache locking problem for a single task, and propose two dynamic cache locking approaches. The first approach formulates the problem as a global ILP (Integer Linear Programming) problem that simultaneously selects a near-optimal set of variables as the locked cache contents and allocates them to the D-cache. The second one iteratively constructs a subgraph of the CFG of the task where the lengths of all the paths are close to the longest path length, and uses an ILP formulation to select a near-optimal set of variables in the subgraph as the locked cache contents and allocate them to the D-cache. For both approaches, we propose a novel, efficient D-cache allocation algorithm. We have implemented both approaches and compared them with the longest path-based, dynamic cache locking approach proposed in [22] and the static WCET analysis approach without cache locking proposed in [14] by using a set of benchmarks from the Mälardalen WCET benchmark suite, SNU real-time benchmarks and the benchmarks used in [27]. Compared to the static WCET analysis approach, the average WCET improvements of the first approach range between 11.3% and 31.6%, and the average WCET improvements of the second approach range between 12.3% and 32.9%. Compared to the longest path-based, dynamic cache locking approach, the average WCET improvements of the first approach range between 4.7% and 14.3%, and the average WCET improvements of the second approach range between 5.3% and 15.0%.

Caches are widely used in embedded systems to bridge the increasing speed gap between processors and off-chip memory. However, caches make it significantly harder to compute the WCET(Worst Case Execution Time) of a task. To alleviate this problem, cache locking has been proposed. We investigate the I-cache locking problem, and propose a WCET-aware, min-cut based dynamic instruction cache locking approach for reducing the WCET of a single task. We have implemented our approach and compared it with the two state-of-the-art cache locking approaches by using a set of benchmarks from the MRTC benchmark suite. The experimental results show that our approach achieves the average improvements of 41%, 15% and 7% over the partial locking approach for the 256B, 512B and 1KB caches, respectively, and 7%, 18% and 17% over the longest path based dynamic locking approach for the 256B, 512B and 1KB caches, respectively.