Abstract: The rapid development of web applications has created many security problems related to intrusions not just on computer, network systems, but also on web applications themselves. In Web Intrusion Systems (WIS), most techniques used nowadays are not able to deal with the dynamic and complex nature of cyber-attacks on web applications and related issues. However, web intrusion techniques based on machine learning approaches with statistical analysis of data enable autonomous detect intrusive and non-intrusive traffic with low false-positive errors. In this paper, we present the survey of various machine learning techniques used to build WIS. In addition, we develop the experimental framework for comparative analysis of some machine learning techniques applying on the well-known benchmark data set - CSIC 2010 HTTP [13].

Abstract: Demographic attributes of customers such as gender, age, etc. provide important information for e-commerce service providers in marketing and personalization of web applications. However, online customers often do not provide this kind of information due to privacy issues and other security-related reasons. In this paper, we proposed a method for predicting the gender of customers based on their catalog viewing data on e-commerce systems, such as the date and time of access, list of categories and products viewed, etc. We employ a machine learning approach and investigate a number of features derived from catalog viewing information to predict the gender of viewers. Experiments were conducted on datasets provided by the PAKDD'15 Data Mining Competition and achieved the good result. The results 81.2% on balanced accuracy and 81.4% on macro F1 score showed that basic features such as viewing time, products/categories features used together with more advanced features such as products/categories sequence and transfer features effectively facilitate gender prediction of customers.

Abstract: Hiding High Utility Sequential Patterns (HUSPs) is the task of finding the ways how to hide high utility sequential patterns appearing in sequence databases so that the adversaries cannot discover them after hiding. It has become an important research topic in recent years and has been applied in various domains such as business, marketing, stock, health and security, etc. However, few methods have been proposed for hiding high utility sequential patterns. The traditional approach is first using mining algorithms to discover all high utility sequential patterns in a specific user threshold and then apply hiding algorithms to conceal them. Generally, these algorithms are usually time-consuming when performing in the large datasets. To address this issue, this paper presents an integrated algorithm named MHHUSP (Mining with Hiding High Utility Sequential Patterns) which combines mining process with hiding process in a common process. An extensive experimental evaluation is conducted on large-scale datasets to evaluate the performance of the proposed algorithm in terms of execution time and memory consumption. Experimental results show that MHHUSP outperforms the state-of-the-art HHUSP algorithm [2].

Abstract: In recent years, dependency parsing is a fascinating research topic and has a lot of applications in natural language processing. In this paper, we present an effective approach to improve dependency parsing by utilizing supertag features. We performed experiments with the transition-based dependency parsing approach because it can take advantage of rich features. Empirical evaluation on Vietnamese Dependency Treebank showed that, we achieved an improvement of 18.92% in labeled attachment score with gold supertags and an improvement of 3.57% with automatic supertags.

Abstract: This paper presents background removal methods to increase the accuracy of saliency-based person re-identification. After evaluating the current global salience algorithm, we found that wrong matching appears when (1) images of different people have a similar or the same background and/or (2) salience on the backgrounds of various images are similar. To prove the maximum theoretical accuracy of the global saliency method when using background removal, we use a manual method with support of an interactive segmentation tool. Another method is to use an ellipse to localize a human body region. This method is a preliminary step for confirming a possibility of applying an automatic technique. The human body region is automatically determined by utilizing a local salience method named GBVS with an adaptive threshold for every image in VIPeR dataset. Preliminary results show that when applying the three background removal methods, the accuracy at rank 1 of CMC curve increases from 20.00% to 27.18%, 24.81% and 23.80% for interactive segmentation, elliptical window and adaptive local salience methods, respectively.

Abstract: Research in knowledge management (KM) has recently revealed that barriers to knowledge sharing (KS) can significantly influence KS and reduce creativity. KS is a critical contributor to creativity and innovation among individuals in organizations. This paper introduces a framework using the Lotus Blossom technique to classify KS barriers. This technique emphasizes the power of brainstorming on the field of interest by the application of a visual representation of ideas. An exploration of steps to classify barriers is discussed. One of the key aims of the framework is to ensure that barriers can be classified in ways that best identify in order to find ways to remove them. A review of a large number of KM articles in the literature has identified 160 barriers to KS in a variety of organizations. These were classified into eight themes: Social, Individual, Culture, Technology, Political, Organization, Content, and Routine, procedure and process. The paper, then, discussed the most significant barriers to KS: Psychological ownership, Lack of a motivation, Lack of trust, Lack of incentive and reward systems, Lack of organization culture, Lack of leadership, Lack of technical support, Insufficient technology infrastructure. Implications and future research in this area are also proposed.

Abstract: Applying object-oriented domain-driven design in practice requires bridging the gaps that exist among the perceived domain class models of the key stakeholders involved. In this paper, we propose a domain-driven design method that uses meta-attributes with an aim to fill these gaps. Our method extends and generalises a previous work to use meta-attributes to build the domain class model. The meta-attributes are designed to not only make it easier for the designer and domain expert to collaboratively capture the domain-specific requirements in the model, but to ease the translation of the model to design specification. This specification is written in an object-oriented, internal DSL. To increase productivity, we define a generator function that realises the meta-mapping between the state and behaviour spaces of a domain class to automatically generate its behavioural specification. We demonstrate our method with an implementation in a prototyping tool for the domain class model.

Abstract: Quantum neural networks (QNNs) have been explored as one of the best approach for improving the computational efficiency of neural networks. Because of the powerful and fantastic performance of quantum computation, some researchers have begun considering the implications of quantum computation on the field of artificial neural networks (ANNs).The purpose of this paper is to introduce an application of QNNs in construction of cryptography system in which two networks exchange their outputs (in qubits) and the key to be synchronized between two communicating parties. This system is based on multilayer qubit QNNs trained with back-propagation algorithm.

Abstract: The evolution of Facebook social network with event feature helps improve the quality of interaction among users in real life. Different from other objects such as movies and books, recommendation problem for events is inherently cold-start and affected by the natures of each social network platform (Facebook, Meetup, etc). In this paper, we discuss the character of Facebook social network and how the event recommendation problem in this case is different to others. We argue that rather than just trying to place as many as possible the most suitable events (often based on similarity measurement) on top recommendations, it's better to remove the unsuitable ones and reorder the remaining in a way that improves the user experience. From that, we propose a new method for event recommendation divided into two consecutive stages - classification and re-ranking. For the first phase, we use a blending model of probabilistic classifiers to predict positive and negative probabilities for each user-event pair, then evaluate on those results to eliminate all bad cases before passing the rest to the next phase. We also propose a new optimization procedure in this evaluation process. For the second phase, we treat the positive probability as a measurement of similarity and make some comparisons across several reranking techniques to choose the best based on the objective of improving quality of recommended lists. Experimental results on crawled Facebook public events show the effectiveness of the proposed method.

Abstract: Audio fingerprint is the digital fingerprint that can help to identify the audio content. The most usage of audio fingerprint is to detect the illegal contents to support the artists to protect their copyrights. Nowadays, there are millions of audio and video contents uploaded to internet, so the searching speed with the big database size is a big problem for all systems. In this paper, we propose a new hierarchy searching system that can detect the meta information for fingerprint in real time by using the advantages of K-modes and Localitive Sensitive Hashing (LSH). With the power of multiple GPGPU devices, we can obtain the meta information for a query within 2 milliseconds for 10 million songs' database.

Abstract: A grand challenge in knowledge representation is building the intelligent systems for Science Technology Engineering and Math (STEM) Education. In math education, the intelligent problem solver (IPS) must have sufficient knowledge to solve problems automatically, and their solutions are natural, step-by-step and can be understand by the learners. Besides that, Solid geometry is a hardly subject of mathematics to study for the high school studens. In this paper, an IPS in solid geometry is designed. The knowledge base of this system is represented based on Rela-model which is a knowledge model about relation. The inference engine of this system has been also built based on the algorithms to solve problems on objects and model. It shorn solution clearly and step-by-step. This system has been tested on various kinds of solid geometrical exercises in high-school mathematics of Vietnam education.

Abstract: Recently, with the development of the online social network, sentiment classification (SC) which determines opinions of people is a significant task in natural language processing. In this research, we propose a model which combines deep learning and sub-tree mining to resolve sentiment classification problem. Stanford Parser is used to extract the relation from the beginning to the end of the sentences and each sentence is represented as a tree. Afterwards, FindBestSub-tree algorithm with sub-tree mining technique eliminates outliers in the dataset. Then, the order of the words in a sentence changes according to DFS (Depth First Search) from a tree after outlier removal phase. Finally, the association between all words in a sentence and all sentences in a document is captured by LSTM and GRNN, respectively. Document sentiment classification experiment is conducted on multi-domain sentiment dataset. The elimination of outliers leads to higher performance in this model1. In our experiment, the proposed method achieves improvements in term of accuracy in a range of 0.14% – 6.93% over LSTM + GRNN model.

Abstract: Analyzing source code to solve software engineering problems such as fault prediction, cost, and effort estimation always receives attention of researchers as well as companies. The traditional approaches are based on machine learning, and software metrics obtained by computing standard measures of software projects. However, these methods have faced many challenges due to limitations of using software metrics which were not enough to capture the complexity of programs. The aim of this paper is to apply several natural language processing techniques, which deal with software engineering problems by exploring information of programs' abstract syntax trees (ASTs) instead of software metrics. To speed up computational time, we propose a pruning tree technique to eliminate redundant branches of ASTs. In addition, the k-Nearest Neighbor (kNN) algorithm was adopted to compare with other methods whereby the distance between programs is measured by using the tree edit distance (TED) and the Levenshtein distance. These algorithms are evaluated based on the performance of solving 104-label program classification problem. The experiments show that due to the use of appropriate data structures although kNN is a simple machine learning algorithm, the classifiers achieve the promising results.

Abstract: Currency exchange rates forecasting is paid a considerable attention of the researchers in the field of forecasting. The neural network is a well-known tool in machine learning. However, two issues are always interested by the scientists: getting toward to global convergence of extreme solutions and determining the optimal weight of the network. This paper proposes the multi-objective method of ensemble learning techniques based on the non-dominated sorting differential evolution (NSDE, a kind of direction-based methods) for training neural networks and application in Foreign Exchange forecasting problems. Two objectives of the selected model are defined based on the Mean Squared Errors and Diversity respectively, in which we used the concept of fitness-sharing based diversity. We experimented the model on four data sets of currency and compared with some of the others that the research community has announced. Through the performance forecasting indicators to show that our new method gives outstanding forecasting results.

Abstract: This paper deals with a new time-dependent model of public transportation system in the urban context that allows sharing a taxi between a passenger and parcels with speed widows consideration. We solve a realistic dynamic scenario in which requests are accepted or declined at the time of their calls. We classify vehicle speeds by different time windows during a day. Different speed windows induce the dynamic graph model for road networks and make the problem much more difficult to solve. Because of the complex model, the preprocessing steps on data as well as on dynamic graphs are very important. We use a greedy algorithm to initiate the solution and then use some local search techniques to improve the solution quality. The experimental data set is recorded by Tokyo-Musen Taxi company. The data set includes more than 20,000 requests per day, more than 4,500 served taxis per day and more than 130,000 crossing points on the Tokyo map. Experimental results are analyzed on various factors such as the total benefit, the accumulating traveling time during the day, the number of used taxis and the number of shared requests. We expect that our solution has chances to apply directly to real-life situations.

Abstract: Opinion mining and sentiment analysis has recently become a hot topic in the field of natural language processing and text mining. This paper addresses the problem of overall rating for comments in Vietnamese language. The traditional approach of using bag-of-words for feature representation would cause a very high dimensional feature space and doesn't reflect relationship between words. To capture more linguistic information, this paper provides a new neural network model containing three layers: (1) word embedding; (2) comment representation (i.e. comment feature vector); and (3) comment rating prediction. In which, the word embedding layer is designed to learn word embeddings which can capture semantic and syntactic relations between words, the second layer uses a semantic composition model for comment representation, and the third layer is designed as a perceptron and it stands for predicting overall rating of a comment. In experiment, we use a Vietnamese data set which contains comments on the domain of mobile phone products. Experimental results show that our proposed model outperforms traditional neural network models with comment representations based on bag of word model or word vector averaging.

Abstract: This paper proposes neural networks approaches to recognize logical parts in Vietnamese legal documents. We utilize four models based on recurrent neural networks including Long Short Term Memory (LSTM), Bidirectional LSTM and their combination with Conditional Random Fields. The experimental results on the Vietnamese Business Law data set shows the promising of this approach. Although, these approaches don't use any engineering features like traditional approaches, they can produce the state-of-the-art performance.

Abstract: The strength of the association rule-based approach compared to other approaches in building recommender systems is that it can provide the deep explanations. Besides, evaluating the quality of generated rules to obtain the better recommendations is also necessary. This can be completed by using the statistical implicative cohesion measure - a measure used for finding the rules with strong implicative relationships. The higher the cohesion value of a rule is, the better the quality of that rule is. This paper proposes a new approach based on the association rules and the cohesion measure to discover the tendencies in a data set and recommend the top items to a user. The proposed recommender system is tested on the data sets Groceries and CourseRegistration. Depending on the purpose of users, they can change the thresholds on the measure to observe the tendencies as well as to get the top recommendations.

Abstract: Language modeling plays a critical role in many natural language processing (NLP) tasks such as text prediction, machine translation and speech recognition. Traditional statistical language models (e.g. n-gram models) can only offer words that have been seen before and can not capture long word context. Neural language model provides a promising solution to surpass this shortcoming of statistical language model. This paper investigates Recurrent Neural Networks (RNNs) language model for Vietnamese, at character and syllable-levels. Experiments were conducted on a large dataset of 24M syllables, constructed from 1,500 movie subtitles. The experimental results show that our RNN-based language models yield reasonable performance on the movie subtitle dataset. Concretely, our models outperform n-gram language models in term of perplexity score.

Abstract: Web-based social system enables new community-based opportunities for participants to engage, share and interact. The rapid growth of Facebook has triggered a dramatic increase in spam volume and sophistication. Spammers post their status or comment in Page to send spam content to their friends or other users in the network. In this paper, we consider the problem of detecting spam accounts on Facebook based on comment content and user social behavior. We will propose a hybrid approach using Maximum Entropy (Maxent) model for classifying user comments as either spam or non-spam. We carefully conducted an empirical evaluation for our model on a large collection of comments in Vietnamese Facebook Pages and achieved promising results with an average accuracy of more than 90%.

Abstract: The study refers to the problem of the presence of the crisis in psychosocial development of young people – adolescents and young adults. Both the youth and young adults go through numerous, naturally present in human development, moments of increased tension resulting from the appetite for independence and self-sufficiency. This seems to be a common feature for both groups; both adolescents and young adults experience the burden due to overlapping obligations and commitments made to oneself and to the world and which comes from the specific social roles they assume as well as the development – related tasks they perform. The challenges are taken up to find self-fulfillment in numerous new roles, to achieve ambitions of everyday life as well as the future ones. Not surprisingly, in this period a man is believed to be, on the one hand, exposed to the experience of crisis (relating to the search for oneself and one’s own place in life, in social relationships and professional life) and, on the other hand, a young adult most intensely makes its network of social support for further years. In the study, the author refers to the research in which young adults confirm the presence of the crises in their lives. They identify and name these crisis situations and formulate their expectations of the sources and forms of support in the crisis.

Abstract: In precise real-time systems, scheduling algorithms without considerable complexity are more effective to be applied. Virtual release advancing [3] for shorter response times is a technique that is valuable for the Earliest Deadline First (EDF) scheduling, but not adaptive to precise systems due to its high time complexity. A new scheduling algorithm, called enhanced virtual release advancing, that mitigates the time complexity, is presented in this paper. Applied in the Total Bandwidth Server (TBS) context, the new algorithm significantly improves the time complexity while guaranteeing the responsiveness and schedulability. Simulation results show that the runtime overhead is approximately 56% lower than that of the original algorithm under the periodic processor utilization of 95%.

Abstract: The article introduces the application of the Cellular Automata and Predator-Prey model to simulate effects of natural enemies to the growth and spread of the brown planthopper in rice fields. The article also mentions the use of GIS data for spatial simulation, thus helping the simulating results more visual and more realistic. The model can support prediction and prevention based on natural enemies hoppers without affecting the ecological environment and less costly. The results of the model is tested with real data about BPH in Dong Thap in 2010.

Abstract: In conventional algorithms a data point can be assigned to only a single cluster. However, in reality there are several types of data that a data point belongs to multiple categories and causes ground-truth clusters overlap. In this circumstance, conventional clustering cannot work effectively. To handle this problem, several algorithms are proposed and referred as “overlapping clustering”. One of state-of-the-art partition-based overlapping clustering technique is “Non-exhaustive, Overlapping K-Means” or “NEO-K-Means” in short, which is an extension of K-Means clustering algorithm. Although NEO-K-Means works effectively for most real-world multi-category data, however, distance between clusters that is essential parameter for overlapping clustering is not included in the process of algorithm. This is a huge drawback of NEO-K-Means that makes clustering accuracy lower than it should be. In this paper, we aim to overcome this limitation in NEO-K-Means with our assumption that closer clusters should have higher probability to overlap than others that are faraway. To achieve the goal, the parameter that represents distances between clusters centroids is included in objective function before prioritizing and assigning data points to the overlapped regions of clusters. The experimental results show that our method significantly outperforms NEO-K-Means on all nine real multi-category data sets in terms of F 1 .

Abstract: Object localization is the first requirement for many applications such as navigation, obstacle avoidance, object grasping. In this paper, we present a new method that combines two techniques of localization: RFID (Radio Frequency Identification) based and RGB-D camera based. In our method, each RFID tag with an unique ID will be assigned to one object. Based on the RSSI (Received Signal Strength Indication) received from RFID readers, we make a coarse localization of the object. This localization result is then projected on the image captured by a Kinect sensor to limit the region of search (RoS). If the Kinect sensor provides depth in this RoS, depth distribution of the RoS will be computed and served to narrow again the RoS. Finally, object position is refined by applying a HOG-SVM detector [1] on the RoS of the RGB image. The combination of RFID and RGB-D is twofold. It avoids both false positives and negatives when using only RGB-D information. It reduces the computational time. We have evaluated our method in a real-scene with different positions of object. The combination of RFID and RGB-D helps to reduce the localization error from 1.02m to 0.16m in average compared to using solely RFID. The HOG-SVM detector applied on the RoS obtained higher precision (100%) than applied on the whole RGB image (72.86%) while keeping the same recall (98.96%). It also reduced the computational time from 1.038s per image to 0.39s.

Abstract: Generated from the urgent demand of innovative teaching methods, the teaching strategy should be developed based on ability development of the students. An approach of designing for a fully instructive and smart system which builds optimizations that helps solve the current planar geometrical problems will be presented in this article. The technical solution for building a helpfully instructive and smart towards outputting the geometrical problems tends to comprise the basic knowledge base of geometry which includes abundantly all sides of geometrical knowledge and the reservation of recorded and categorized problems that have been associated with the knowledge base. Via the practical optimizing designation, this intellectually designed method will certainly contribute all technical supports to the users in order to find answers and help build instructing supports in various spheres such as physics and chemistry. As highly recommended, this optimization has been put into tests and has got significant and remarkable outcomes.

Abstract: Supplier selection and evaluation plays an importance role for companies to gain competitive advantage and achieve the objectives of the whole supply chain. To select the appropriate suppliers, many qualitative and quantitative criteria are needed consider in the decision process. Therefore, supplier selection and evaluation can be seem as a multi-criteria decision making (MCDM) problem in vague environment. However, most existing fuzzy MCDM approaches have been developed using normal fuzzy numbers or converting generalized fuzzy numbers into normal fuzzy numbers through normalization process. This leads to a restriction in the application of the fuzzy MCDM approaches. In this study, a generalized fuzzy MCDM approach is proposed to select and evaluate suppliers. In the proposed approach, the ratings of alternatives and important weights of criteria are expressed in linguistic terms using generalized fuzzy numbers. Then, the membership functions of the final fuzzy evaluation value are developed. To make procedure easier and more practical, the weighted ratings are defuzzified into crisp values by employing the maximizing and minimizing set ranking approach to determine the ranking order of alternatives. Finally, a numerical example is presented to illustrate the applicability and efficiency of the proposed method.

Abstract: It has been widely recognized that decision-making is a crucial part of hospital management which goes through a process of medical behavior. Even though data mining and knowledge discovery techniques have been used to clinical medicine frequently, little research has been conducted on hospital decision-making especially hospital-cost analysis on treatment therapies among inpatients which is considered as an important aspect of annual hospital evaluation and accreditation. In this paper, we propose a novel semi-supervised fuzzy co-clustering method for hospital-cost analysis from electronic medical records. Fuzzy co-clustering is a well-known technique that performs simultaneous fuzzy clustering of objects and features which result in dynamic dimensionality reduction mechanism for categorizing high-dimensional data. However, in many real-world applications, prior knowledge of a dataset is actually available to the users; thus it is necessary to integrate this information in the clustering process. This approach is called semi-supervised fuzzy co-clustering which has been investigated and developed further within an application to hospital-cost analysis of Hanoi Medical University Hospital (HMUH), Vietnam. The findings of the paper suggest the most crucial factors for medical expense in HMUH which are significant to gradually reduce the cost of treatment meanwhile improve the quality of services.

Abstract: This study proposes a system that allows for the quantitative creation of music using a few parameters identified to be relevant to the association between music and emotion. One proposed emotional model of music suggests that musical emotion depends on a realization of and a deviation from an expectation. Expectations and their realizations and deviations are regarded as effects of probabilities, which allows us to quantitatively treat musical events as information. Additionally, as musical events have relationships of physical characteristics, they can be expressed as a certain probabilistic distributions using these relationships. In this study, we focus on “pitch” and “sound timing” of notes. The relationships of these physical characteristics are represented using ratio. Using “ratio relationships” and “normal distribution,” based on information theory, we propose a generative music system with quantitative controllers. In terms of pitch, the system provides a controllable “musical scale” by adjusting the variance of a normal distribution and controllable “modulation” by adjusting the median of a normal distribution. Additionally, using two normal distributions, the system provides controllable “musical modes.” In terms of sound timing, the system provides controllable “rhythmic complexity” by adjusting a variance of a normal distribution. The system can create not only music, but also complex music that listeners recognize as non-music. The features of this system equip it to contribute toward future studies that aim to reveal the relationship between music and emotion. Moreover, the system creates complex music, such as contemporary music and free jazz from simple music, such as children's songs and local folk songs. Therefore, the system can be used for creating art works, and we will attempt to use it thus.