Search

With the increase in the number of live TV channels, audiences must spend increasing amounts of time and energy deciding which shows to watch; this problem is called information overload, and recommender systems (RSs) are effective methods for addressing such problems. Due to the high update rates and low replay rates of TV programs, the item cold-start problem is prominent, and this problem seriously affects the effectiveness of the recommender and limits the application of recommendation algorithms for live TV. To solve this problem better, RSs must consider information in addition to the time slot strategy, which relies on experience. At present, no methods make good use of viewing behavior records. Therefore, in this paper, we proposed a viewing environment model called DeepTV that considers viewing behavior records and electronic program guides and includes a feature generation process and a model construction process. In the feature generation process, we defined seven key features by clustering viewing time, distinguishing positive and negative feedback, capturing continuous viewing preference and introducing the remaining time proportion of candidate programs. We normalize the continuous features and add powers of them. In the model construction process, we regard the live TV recommendation task as a classification problem and fuse the above features by using a neural network. Finally, experiments on industrial datasets show that the proposed model significantly outperforms baseline algorithms.

According to the characteristics of positive and negative edge of signed networks, a new signed networks community detection algorithm BTCN_SNCD (Signed networks Community Detection Based on the Tightness of Common Neighbors) is proposed based on the tightness of common neighbors. Firstly, in view of the shortcomings of the traditional local similarity metrics only considering the number of common neighbors, the contribution degree and the tightness of common neighbors are defined, which can discover the initial communities with tighter structure. Secondly, for the overlapping nodes exist in initial communities, we propose overlap coefficient and community density to remerge the initial communities, thus improving community detection accuracy. Finally, the correctness and efficiency of the BTCN_SNCD algorithm are verified through a comparison with classic signed networks community detection algorithms.

Unlike traditional clustering methods that focus on grouping objects with similar values on a set of dimensions, pattern-based clustering finds objects that exhibit coherent patterns in subspaces. Pattern-based clustering extends the concept of traditional clustering and benefits a wide range of applications. However, most of previous approaches based on single pattern model can only explore one of specific patterns, not both of them. This paper analyses different kinds of patterns between items, presents the conception of multi-pattern model. Based on this model, we can capture patterns of shifting, scaling, and other patterns with the same feathers simultaneously. From the study of multi-pattern model's characters and operating principles, an effective algorithm is introduced to cluster objects which are coherent with multi-patterns.

Converged Enhanced Ethernet (CEE) presents challenges to PCIe devices. The single-root I/O virtualization (SR-IOV) standard allows a PCIe device to be efficiently shared among Virtual Machines (VMs), without performance loss. We propose a design method for FCoE adapter. Based on the designed allocation scheme for Receive and Transmit queues, we put forward an allocation model from Virtual Functions (VFs) to VMs. This model can not only solve the problem of the limited number of VFs, but also distribute network bandwidth according to the VM's needs. In addition, we implement a communication mechanism between Physical Function (PF) Driver and VF Driver based on the inter-domain communication mechanism in Xen. This design scheme we proposed will not only make full use of hardware resources to meet the requirement of I/O performance in CEE, but also provide a method for PF Driver-VF Driver communication to reduce the complexity of hardware design.

In this paper, we focus on efficient keyword query processing for XML data based on the SLCA and ELCA semantics. We propose a novel form of inverted lists for keywords which include IDs of nodes that directly or indirectly contain a given keyword. We propose a family of efficient algorithms that are based on the set intersection operation for both semantics. We show that the problem of SLCA/ELCA computation becomes finding a set of nodes that appear in all involved inverted lists and satisfy certain conditions. We also propose several optimization techniques to further improve the query processing performance. We have conducted extensive experiments with many alternative methods. The results demonstrate that our proposed methods outperform previous methods by up to two orders of magnitude in many cases.

Pattern-based clustering, which capture the similarity of the patterns exhibited by objects in a subset of dimensions, has broad applications in DNA microarray data analysis, customer segmentation, e-business data analysis, etc. However, pattern-based clustering often returns a large number of highly-overlapping clusters, which makes it hard for users to identify interesting patterns from the huge mining results. Moreover, there lacks a general measurement to evaluate the quality of Clusters which pattern-based clustering obtained. In this paper, we discuss factors which cause highly-overlapping, make error analysis and pattern weighting, and propose qScore as a key evaluation parameters on quality of Clusters. A algorithm which based on qScore is presented to solve the problem of high-overlapping and get better quality clustering results.

Keyword search is a user-friendly way to query XML data,such that users do not need to understand the complex syntax of structured query languages and the complex structural information of the underlying XML data.However,existing semantics suffer from limited expressiveness, thus users cannot obtain desired information in many cases.We propose an entity-based semantics,namely ELCEA,to capture users'search intension by supporting additionally the semantics of "NOT".We verified the effectiveness of ELCEA by conducting rich experiment according to different metrics.

The main task of the existing link prediction is to predict link existence. However, this paper proposes a new method aiming at the online dating networks, which using weighted transactional information to predict link intensity. Three different supervised learning methods are used in detail, comparing the importance of attribute features, topological features, transactional features and global features based on users' profile information, as well as the influence of four different network graphs to the model performance. The experiment on Xiaonei dataset shows that the method used in this paper can be used to predict link intensity accurately, also illustrates that the global features have the greatest impact on the performance of model .

Existing link prediction methods have mostly adopted overlay network to represent social network and used topological features or attributive features between two nodes to predict the formation of links. However, the limitation of these methods is that they not only use a single feature for link prediction, but also not take into account the time factor and the importance of features. This paper considered the problem of temporal link prediction on the basis of using various features at the same time. Specifically, it presented a matrix-based method for combining temporal features, weighted attributive features and weighted topological features. Using two different datasets our experiments have confirmed that our approach achieved better performance.

Multi-relational data mining can be categorized into graph-based and logic-based approaches. In this paper, we propose some optimizations for mining graph databases with Subdue, which is one of the earliest and most effective graph-based relational learning algorithms. The optimizations improve the subgraph isomorphism computation and reduce the numbers of subgraph isomorphism testing, which are the major source of complexity in Subdue. Experimental results indicate that the improved algorithm is much more efficient than the original one.

Unlike traditional clustering methods that focus on grouping objects with similar values on a set of dimensions, clustering by pattern similarity finds objects that exhibit a coherent pattern of rise and fall in subspaces. Pattern-based clustering extends the concept of traditional clustering and benefits a wide range of applications, including large scale scientific data analysis, target marketing, Web usage analysis, etc. However, state-of-the-art pattern-based clustering methods (e.g., the sigma-pCluster algorithm), mining candidate clusters mostly by comparing each pair of attributes and objects, which have reduced the efficiency and makes them inappropriate for many real-life applications. This paper present a fast algorithm for mining candidate clusters. We called it Zero-Sub-Matrix. It has a better efficiency than previous algorithms.

We present a general approach to speeding up a family of multi-relational data mining algorithms that using ID propagation to obtain the information needed for building decision tree classifier from relational database. Preliminary results of our experiments suggest that the proposed method can yield 1-2 orders of magnitude reductions in the running time of such algorithms without any deterioration in the accuracy of results. The proposed modifications enhance the applicability of decision tree algorithms to significantly relational databases that would otherwise be not feasible in practice.

A multi-relational association rule mining algorithm with guidance of the user is proposed in this paper. A tuple ID propagation approach is used to solve the association rule mining with multiple database relations directly, and introduces the concept of the user's guidance in this paper. The approach has improved user's satisfaction with the mining result. Compared with the traditional algorithms, it improves the accuracy and supports multi-relational databases directly, so its running time is less than that of the multi-relational association rule mining methods based on ILP.