Search

Micro-service is a prevalent framework for Internet services. Its loosing coupling design has gained popularity recently. Loosing coupling services offer convenience, scalability and flexibility, however bring unexpected maintenance problems. This paper focuses on the problem where service administrations attempt to monitor the runtime services performance and status with mined service logs. To achieve it, we propose to extract unstructured logs as log patterns and visualise the services temporal relations using temporal graphs. Our analysis and data mining method can help to give new insights in system management area and non-trivial analysis process. The early experiments and analysis can clearly reveal inner structures of system and help to gain deep understanding.

For traditional integrated navigation system based on SINS/GNSS, the integrated navigation system will work in standalone inertial navigation when GNSS signals are lost. In this situation, the divergence of SINS error is fast without GNNS information correction. Besides, the divergence of SINS error is related to the motion mode of the vehicle. This paper proposes a BP neural network (BPNN) aided integrated navigation method based on vehicle motion learning. The inputs of the neural network are 10-dimension, including the outputs of the IMU, the three attitude angles estimated by SINS, and the GNSS interruption lasting time. The outputs are three predictions of SINS position error. When GNSS is effective, the neural network is used to learn the SINS error divergence characteristics under several basic motion modes of the vehicle. When the GNSS signal is disturbed, the neural network is used to correct the position error of the SINS to improve the navigation accuracy. Finally, the effectiveness of the algorithm is verified by using an 8-shape motion of the vehicle.

Fraud detection is usually regarded as finding a needle in haystack, which is a challenging task because fraudulences are buried in massive normal behaviors. Indeed, a fraudulent incident usually takes place in consecutive time steps to gain illegal benefits, which provides unique clues to probing frauds by considering a complete behavioral sequence, rather than detecting frauds from a snapshot of behaviors. Also, fraudulent behaviors may entail different parties, such that the interaction pattern between sources and targets can help distinguish frauds from normal behaviors. Therefore, in this paper, we model the attributed behavioral sequences generated from consecutive behaviors, in order to capture the sequential patterns, while those deviate from the pattern can be regarded as fraudulence. Considering the characteristics of behavioral sequence, we propose a novel model, HAInt-LSTM, by augmenting traditional LSTM with a modified forget gate where interval time between consecutive time steps are considered. Meanwhile, we employ a self-historical attention mechanism to allow for long-time dependencies, which can help identify repeated or cyclical appearances. In addition, we encode the source information as an interaction module to enhance the learning of behavioral sequences. To validate the effectiveness of the learned sequential behavior representations, we experiment on real-world telecommunication dataset under both supervised and unsupervised scenarios. Experimental results show that the learned representations can better identify fraudulent behaviors, and also show a clear cut with normal sequences in the lower dimensional embedding space through visualization. Last but not least, we visualize the weights of attention mechanism to provide rational interpretation of human behavioral periodicity.

With the tremendous development of Internet, many domain-specific new words appear in various media text streams such as forums, Sina Weibo, Wechat, etc. These new words are always a group of important words in specific domains and are significant for NLP tasks. Most existing models have time-consuming processing or cannot handle out of vocabulary (OOV) words on streaming and online scenes. In this paper, we propose an unsupervised method, D-TopWords with Gaussian LDA, to perform online detection of domain-specific new words effectively. Different from traditional new words detection models, our method is a joint statistical model based on a finite word dictionary without any handcraft features. By further introducing Gaussian LDA into our model, we solve properly the problem of OOV words from new text streams. Experimental results show that our work can successfully extract domain-specific new words and it has a better performance in online detection task than some state-of-the-art methods.

In recent years, anomaly detection has become a focal point of data mining, and numerous efforts have been made to conduct extensive researches on the theories and techniques for detecting abnormal data points. Although the amount of anomaly data is relatively small, they can potentially bring huge losses to social economy, public resources and individual properties. Thus, we propose an unsupervised anomaly detection framework named AEKNN, which aims to incorporate the advantages of automatically learnt representation by deep neural network to boost anomaly detection performance. The framework combines the training of an autoencoder and a k-th nearest neighbor based outlier detection method. We further validate the performance of our proposed model with an extensive experimental study on three UCI datasets. The parameter sensitivity results demonstrate that the proposed algorithm can scale well with respect to both dataset size, data feature dimensionality and anomaly class proportion.

Online advertising has attracted lots of attention in both academic and industrial domains. Among many realworld advertising systems, click through rate (CTR) prediction plays a central role. With a large volume of user history log and given its various features, it is quite a challenge to fully extract the meaningful information inside that amount of data. What's more, for many machine learning models, in order to achieve the best performance of the CTR prediction, a lot of hyper-parameters need to be tuned, which often costs plenty of time. In this paper, we propose an ensemble model named ETCF, which cascades GBDT with gcForest to tackle the practical problems of CTR prediction and do not need much hyper-parameter tuning work to realize its best performance. Experimental results validate the effective prediction power of ETCF against classical baseline models, and the superiority of GBDT transformed features.

With the rapid development of online social media, the short text has become the prevalent format for information of Internet. Due to the severe data sparsity issue, accurately discovering knowledge behind these short texts remains a critical challenge. Since regular topic models, such as the Latent Dirichlet Allocation (LDA), can not perform well on short texts, many efforts have been put on building different types of probabilistic topic models for short texts. Inducing topics from dense word-word space instead of sparse document-word space becomes an emerging solution for avoiding data sparsity issue, and the representative one is the Word Network Topic Model (WNTM). However, the word-word space building procedure of WNTM often imports much irrelevant information. In light of this, we propose the Robust WNTM (RWNTM), which can filter out unrelated information during the sampling. The experimental results demonstrate that our method can learn more coherent topics and is more accurate in text classification, as compared with WNTM and other state-of-the-arts.

Document modeling with hidden units as known as topics are very popular. Non-negative matrix factorization(NMF) is one of the most important techniques in document representation, which decomposes a document-term matrix into a document-topic matrix and a topic-term matrix. Since orthogonal constraint would limit terms occur only in one topic, we abandon this strong constraint. Furthermore, in order to represent documents in a certain number of topics with more semantic information, we add diversifying regularization and sparse constraint into NMF, which shows a great improvement in text classification and clustering. In the end, we draw the figure of topics similarities and display the top 20 weighted words in each topic to reveal that diversifying regularization can efficiently reduce the overlapping terms.

Aspect-based opinion mining is to find elaborate opinions towards an underlying theme, perspective or viewpoint as to a subject such as a product or an event. Nowadays, with rapid growing of opinionated text on theWeb, mining aspect-level opinions has become a promising means for online public opinion analysis. In particular, the booming of various types of online media provide diverse yet complementary information, bringing unprecedented opportunities for public opinion analysis across different populations. Along this line, in this paper, we propose CAMEL, a novel topic model for complementary aspect-based opinion mining across asymmetric collections. CAMEL gains complementarity by modeling both common and specific aspects across different collections, and keeping all the corresponding opinions for contrastive study. To further boost CAMEL, we propose AME, an automatic labeling scheme for maximum entropy model, to help discriminate aspect and opinion words without heavy human labeling. Extensive experiments on synthetic multicollection data sets demonstrate the superiority of CAMEL to baseline methods, in leveraging cross-collection complementarity to find higher-quality aspects and more coherent opinions as well as aspect-opinion relationships. This is particularly true when the collections get seriously imbalanced. Experimental results also show that the AME model indeed outperforms manual labeling in suggesting true opinion words. Finally, case study on two public events further demonstrates the practical value of CAMEL for real-world public opinion analysis.

Modular test and hierarchical test of core-based System-on-Chip (SoC) are two widely used SoC test methodologies. In this paper, the hybrid test methodology that incorporates these two together is studied by using an industrial real case. Thorough experimental results are demonstrated to compare various scenarios of the hybrid hierarchical and modular tests for SoC designs. Based on the experimental results, using channel sharing based modular test technology at a group of cores combined with hierarchical test to map the patterns of core groups to the top level would result in the most efficient total test time.

Topic model has attracted much attention from investigators, as it provides users with insights into the huge volumes of documents. However, most previous related studies that based on Non-negative Matrix Factorization (NMF) neglect to figure out which topics are widespread in the documents and which are not. These widespread topics, which we refer to coarse-grained topics, have great significance for people who concentrate on common topics in a given text set. For example, after reading the massive job ads, the jobseekers are eager to learn employers' basic requirements which can be regarded as the coarse-grained topics, as well as the additional requirements that can be deemed to be the fine-grained topics. In this paper, we propose a novel method which applies two different sparseness constraints to NMF to tell coarse-grained topics and fine-grained topics apart. The experimental results of demonstrate that the new model can not only discover coarse-grained topics but also extract fine-grained topics. We evaluate the performance of the new model via text clustering and classification, and the results show the new model can learn more accurate topic representations of documents.

Recently topic model has been more and more popular in lots of fields such as information retrieval and semantic relatedness computing, but its practical application is limited to the scalability of data. It cannot be efficiently executed on large-scale datasets in a parallel way. In this paper, we introduce an improved Regularized Latent Semantic Indexing(RLSI) with L1/2 regularization and non-negative constraints. This method formalizes topic model as a problem of minimizing a quadratic loss function regularized by L1/2 and L2 norm with non-negative constraints. This formulation allows the learning process to be decomposed into a series of mutually independent sub-optimization problems which can be processed in parallel, therefore, it has the ability to handle large-scale data. The non-negative constraints and L1/2 regularization allow our model to be more practical and more conducive to information retrieval and semantic relatedness computing. Extensive experimental results show that our improved model can deal with large-scale text data, and compared with some of the-state-of-the-art topic models, it is also very effective.

This paper presents approaches to reduce scan test data volume for SoC designs in EDT environment. They target different factors impacting scan test data volume - scan channel count, pattern count and shift cycles. In the experiments on an industrial SoC design, up to 23% scan test data volume can be reduced.

Along with the development of distributed computation and the rapid growth of data, scientific research increasingly requires the support of high-efficiency relational data processing framework. According to the characteristics of scientific data, for example bulk inserts and unfrequented change, this paper proposes a streaming processing model called Matrix-Query with the matching data storage architecture for relational query. Through transforming the original relational schema to entities and key-value indexing, the data storage solution provides more localization operation and data positioning. Compare to traditional Map-Reduce model, the Matrix-Query isolates the influence between subtasks to ensure execution in a streaming and parallel manner and reduces negative impacts of writing intermediate file. We also optimize the data structure and subtask management to improve the performance of Matrix-Query. The experimental results demonstrate performance advantage of Matrix-query compared to two famous data processing systems, Hive and HadoopDB, which build on the top of Map-Reduce model.

In this study, a comprehensive system was developed to meet the demand of the Security Guarding during 2008 Beijing Olympic Games. In the system, meteorological models, namely, MM5 and RAMS6.0, and a poisonous clouds diffusion model over complex terrain (CDM) were configured in a one-way off-line nested way. In the system, MM5 runs were performed in an real-time operational way with a horizontal resolution of 3000m, which took the output from a Global scale model (T213 from the Chinese Meteorological Administration, CMA) as the initial and boundary conditions, and the output from MM5 was used to drive the RAMS6.0 runs to provide a 36-hour prediction with a horizontal resolution of 1000m. The wind and turbulent field output from RAMS6.0 was sequentially used to drive CDM runs, which can provide the prediction of the concentration field and the dose field of the chemical clouds. During 2008 Beijing Olympic Games, the system was used to provide the scenario prediction results and the security target was set as National Stadium (known as bird’s nest), and the quantitative analysis of the hazard risk was performed based on the scenario prediction results.

The Construction of the Meta-Synthetic of Army Material Facility System of Systems(MSOAMFSOS) is a complicated systems engineering, which relates to the classification & composition, architecture & organization, and operational application, etc. This paper discusses the ideal and theory based on meta-synthesis, and the meta-synthetic assessment process from qualitative hypothesis to quantitative validation, establish general path and technique route of its system analysis.

This paper describes the logic built-in self-test (BIST) architecture for test and diagnosis of ASIC devices at the system level. The proposed architecture supports the at speed staggered launch-on-capture clocking scheme and includes novel features to further increase the device’s defect coverage, place-and-route ability, ease of debug and diagnosis, and reduce test power consumption. These features include equivalent clock merging for routing considerations, programmable shift modes for overheat considerations, configurable capture modes for yield loss and IR-drop considerations, as well as BIST signature diagnosis, masked-chain diagnosis, and one-chain diagnosis at the system level. Experimental results have successfully demonstrated the feasibility of using the proposed features for system-level test and diagnosis.