Internet Computing and Big Data ICOMP 2013

Big data including life logs, are attracting attention because of a number of recent developments: ever-increasing volume of data being generated every day due to advances in the broadband environment, increasing sophistication of mobile terminals, and growth of social networking services. If these can be fully exploited, useful added value can be created. Collecting, analyzing and managing such a huge volume of diverse data require innovative ideas and technologies.
 

In many large organizations, the huge amount of data is generated beyond their imagination and companies need to invest a great deal of money in securing and retaining the data. Most of the time, the data is repeatedly written into several different locations and causing the system administrators and management a hard time to restore the data during the disaster. The backup timing window is increasing daily with the new amount of data being added to the system.
 

As most key-value stores partition and replicate data to support high availability with no strong consistency guarantee among replicas, users may suffer from data inconsistency. Although previous research has been done to support strong consistency among replicated data, most existing approaches suffer from potential hotspots and load imbalance. Neither do they consider dynamic data access patterns that may largely vary over time. In this paper, we propose a new approach, called ACRE (Adaptive Chain REplication), to support strong consistency among data replicas, hotspot avoidance and load balancing, and adaptivity to dynamic data access patterns.
 

Many scientific simulations generate bulky data sets that must be mined for observable features. It is often not computationally feasible to do this in real time and the data must be saved for subsequent “off-line” analysis either by separate software or sometimes by direct human visualisation. We present some scoping analysis and preliminary software approaches for mining medium to large scale data sets in the form of time slices or model configurations. We report on current storage and visualisation technology response and interaction times for mining scientific simulations on regular lattices using hyper-bricks of model  configurations.

The analysis of urban mobility information lets us improve different services within a city. However, due to the interaction between different systems logically interconnected generating information on a daily basis, the very large volume of data collected brings challenges for the processing, visualization and simulation of data. Moreover the people through the use of mobile devices, act as a networked sensors collecting data that are convenient to analyze in order to get valuable information about real urban dynamics.
 

There is a vast amount of information on the Internet. A user can directly access any Website listed on the pages of a search engines. However, submitting a query is integral to the use of search engines. Therefore, the accessible information largely depends on the user's own vocabulary and creativity abilities. Consequently, the user may not be able to access adequate information. To solve this problem, a search suggestion function appeared. However, this function is mainly for input support, and it is not suitable for developing a search. In this article, we propose a search assist method by developing a search engine using the "Mandal-Art", a creativity technique.

Recognition of semantic similarity between words plays an important role in text information management, information retrieval and natural language processing. There are two major approaches to recognizing the semantic similarity, among which one way is extracting similarity relationships based on a structured semantic dictionary, while the other way is learning the semantic similarity from a large corpus. Building a semantic dictionary is a time consuming task which also requires much expertise, while the learning method alone cannot extract precise similarity between words. This paper proposes to expand the semantic dictionary by learning the word similarity from heterogenous knowledge bases statistically. This method can not only expand the semantic dictionary from the open knowledge bases, but also achieve accurate semantic similarity. In the evaluation of semantic relatedness competition held by CCF, the proposed system ranks the 3rd place according to the macro average F1 and the 2nd place according to the micro average F1.

Modern advances in computing and infocommunication technologies make it possible to obtain, process and forecast the state of megalopolis traffic for the near future, while ensuring the activity of all groups of potential users of urban road networks. In this study we are created three-tier client-server system, where clients are mobile infocommunication tools - smartphones. Problems for difficult socially-technical systems where processes are characterized by dynamics on time and distribution of a large territory are considered.
 

An evolution model with requirement, personalization and knowledge solution domain was proposed for providing the creative knowledge services. Based on this model, the different between KAAS and KM from the dimension of the core technique, knowledge target, and knowledge extraction were compared and a BNF definition about KAAS was proposed. In cloud computing environment, a KAAS Cloud Architecture was introduced to make the knowledge available on a self-service, social network-based collective intelligence and on-demand service with three knowledge service layers as follows: knowledge resource layer, service delegate layer, and solution delivery layer. All these three key layers for providing knowledge services are how to improve the access of unstructured and scattered information for the non-specialist users, how to provide adequate information to knowledge workers and how to provide the information in situations requiring highly domain-specific, related and time critical information. Then, a research case about KAAS, Eknoware platform, was developed based on the architecture above-mentioned.

A special Genetic Algorithm (GA) with simplex optimization method is proposed for the problem of selecting an optimal cloud service composition plan from a lot of composite plans on the basis of some global Quality-of-Service (QoS) constraints. In this GA, some Simplex Method (SM) operations and a fitness function are provided. The design of the SM operations is made in the light of the characteristics of cloud service composition. After analyzing the types of different QoS attributes, objective function and fitness function are built. SM operations enhance local search capability of GA. Because the design of the hybrid algorithm accords with the characteristics of cloud service selection very well, excellent composite cloud service plan can be gotten from a lot of composite plans on the basis of global QoS constraints. Some tests and analyses show that the proposed algorithm can be a good method for QoS-based cloud service selection.

The number of services in home network environments has been growing increasingly, and therefore users must manage multiple user names and passwords daily. The previous works like SAML (Security Assertion Markup Language) standard and the commercial software called .NET Passport provide web services’ single sign-on function; however, the SAML system not only increases the heavy loading of servers but also costs the Internet flow. Furthermore, it may cause the potential attacks like the replay and man-in-the-middle attacks. Also, in .NET Passport the privacy of users may suffer the risk of eavesdropping. Therefore, this paper develops an agent-based single sign-on system in web services based home network environments. The proposed single sign-on scheme can reduce the number of communications among users and servers, enhance the security of home network services, provide the privacy of users, and promote the efficiency of system in the web services based home network environments.

The relevance of Web Services (WS) has received important interest among businesses. Along with the growing interest in WS comes also the users’ concerns about reliability, security, and privacy vis-à-vis these services provided via the Web. The objective of the present paper is to discuss managerial mechanisms that may help to mitigate these issues in building a trustworthy environment over the Internet. First, an examination of the ASP decision model with its related factors is presented. Second, the Outsourcing decision model with its associated advantages and drawbacks is analyzed. Finally, the presence of Web assurance seals that aim to reduce users’ concerns about security and privacy is examined. The consideration of these managerial mechanisms may helps managers to make sound decisions related to Web Services (WS), develop trust among potential and existing users of WS, and eventually lead to a decision model framework for WS.

Threats against computer systems continue to multiply, but existing security solutions that attempt to keep the attacker out of the system are becoming unable to keep pace with these challenges. In this paper we discuss the application of military deception to defend computer systems. Deception techniques enable the defender to influence the attacker's selection of targets and thus direct him to perform actions that reveal his presence and intentions. We discuss techniques that mislead attackers and cause them to take specific actions that aid in the defense of a computer system. We then focus on web servers, that are frequently attacked often as a first step of a deeper intrusion into a computer network, and present an architecture integrating deception into a popular web server.

In this study, we implemented a novel recommendation application for webpages. Our proposed application based on network traffic can recommend more valuable webpages to users. In this application, we used a special router that captured the packet stream directly. The scoring index used to make effective recommendations was based on the similarity, appearance frequency, and viewing time, which could only be captured by the network router. Thus, it was necessary to identify users to analyze their viewing time of webpages, which was achieved using their IP address and HTTP headers. We evaluated our proposed application by A/B testing and checking whether the browsing history of a user was tracked correctly. This demonstrated that our method identified users and that they preferred the recommended webpages while using our proposed application based on the viewing time.

Lately, a revolutionary method to synthesize maximally permissive with fewest monitors has emerged. It relies on reachability analysis to find minimal sets of legal and forbidden markings. A number of linear constraints are constructed [by solving an integer linear programming problem (ILPP)] to forbid all forbidden markings in the set, while guaranteeing all legal markings reachable. Due to the state explosion problem, it cannot handle very large systems. We propose earlier a method without reachability analysis and minimal siphon extraction; hence it is scalable to large systems. This paper illustrates such by applying the method to very large k-th order systems.

In recent years, large amounts of redundant data are often delivered in a short time. For example, a video streaming sender sends data to multiple receivers concurrently. We proposed network nodes in order to reduce such redundant traffic in TCP/IP network with packet cache. An upstream node compresses data into a small identifier and a downstream node reconstructs. In this paper, we present format of a packet and cache, techniques and strategies to search and replace elements in the cache. We implemented the method to computer network for experiment and carried out measurements. We obtained enough hit rates and high reduction rates that agree with ideal case.

Text and e-mail are widely open to (mis)interpretation by the recipient. It is widely accepted that humor,
sarcasm, double entendres do not work well or are interpreted correctly without the smile, tone, or expression given by the sender. This paper looks at the “mood” or level of happiness of the recipient in the correct interpretation of a text or email message.
 

College students are a highly connected population, and also one with a potential for problem alcohol use. Young adults of college age frequently have cell phones and other mobile electronic devices and spend significant time maintaining their social networks of friends via Facebook, Instagram, Twitter and other instantaneous means of communication such as text messaging. Alcohol use is not uncommon among this same group, and the newfound freedoms experienced in a college environment can lead some young adults to develop problems with alcohol use that go beyond occasional social drinking. This paper reports on the design of an interactive text messaging platform that enables therapeutic alcohol use intervention treatment to go beyond the counselor’s office and directly into the mobile devices students typically carry and use. The design of this interactive system is detailed, technology decisions are discussed, and a summary of the results of a pilot study using the system is presented. Future plans for extension of the platform onto smart phone technology and with a higher degree of interactivity are explored.

Sentimental analysis on web-mined data has an increasing impact on most of the studies. Sentimental influence of any content on the web is one of the most curios questions by the content creators and publishers. Also the impact of the webmined data is completely another issue than the sentiments. For example categorizing a blog post into positive or negative sentiment is a parameter and measuring the like and dislike numbers of the blog post is completely another issue. In this study, the impact and sentimental of the comments collected from five different web sites in Turkish with more than 300,000 comments in total. The web sites are from newspapers, movie reviews, e-marketing web site and a literature web site. All the comments are mixed into a single file. The comments have a like or dislike number, which are used as ground proof of the impact of the comment. The ground proof of the sentiment of the blog is the smileys in the post text.

Different from traditional emotion analysis which focuses on the identification of emotions from the text, this research aims to predict the reader’s emotion for given text. Regarding reader emotion as the response to the text, emotion prediction may be transferred to a classification problem which classifies the text into the categories causing different emotions. In this study, we propose an emotion prediction approach based on Partitioned Latent Dirichlet Allocation (PLDA) model. Through providing the supervised information to the training process of LDA model, PLDA model associates the words from one certain type of emotion to one certain partition of topics. The outputs of PLDA model are used as the features of a multi-label classifier for predicating the reader’s emotion. Evaluations on a large community emotion corpus show that PLDA model achieves much better performance compared to bag of words model
and LDA model.
 

In today’s e-society, the role of the social media is increasingly gaining momentum. It is known to play a vital role in collaboration, community building, participation, and sharing of information. As digital applications, several social media exist and vary in their purposes. In particular, the educational section is one giant beneficiary of this development involving impact creation on students and instructors. However, despite the widespread use of social media by students and instructors, very little empirical evidence exist regarding its impact on student’s collaboration and engagement. In this study, a case study was performed in University of Venda where students offering the module Foundation Information Technology (FIT) were used as participant.

Healthcare generates 30 percent of the world's data with a value of $300 billion in the next decade. Despite this, many healthcare providers have not developed a strategy for handling this data, and realizing the full set of opportunities to reduce costs and improve quality of care delivery. This Research Roadmap is set in major sections of structured and unstructured data and incorporates analytical decision making to improve the healthcare industry through Big Data.

SPM+DMA architecture is commonly employed in many core systems. Loop tiling is an effective way to partition data space for SPM+DMA based data blocking transfers.We observe that DMA based data transfer may induce heavy NoC congestion when the data block is very large. Furthermore, the NoC delay under congestion presents significant differential for the cores in different NoC locations. This paper considers the unbalanced distance-to-data property in the NoC and proposes a NoC-aware adaptive loop tiling (NALT) scheme to improve DMA transfer performance. In the NALT scheme, cores are grouped into different core families taking into account their distancetodata in the NoC. Then tiling sizes are determined for different core families accordingly. On one hand, the NALT scheme can adaptively hide DMA transfer cost into computation cost and reduce the overall execution time. On the other hand, it can avoid bulk DMA data blocks bursting at the same time and thus mitigate NoC congestion. We evaluate the NALT scheme on the NIRGAM platform. The results show that it achieves an average of 21% execution time improvement compared to the uniform loop tiling method in DMA transfer.
 

Computer games on mobile platforms are increasingly popular and modern devices offer games developers unconventional user interface technologies based on device orientation. We describe a maze game implemented both for Android mobile phone and tablet devices that supports navigation though a generated landscape using tilt motions accessed via the devices accelerometer sensors. We report on implementation issues including device sensitivity, resolution and appropriate ways to utilise tilt motion in a practical game or simulation. We discuss future directions for this technology and possible uses in simulations as well as games.
 

With a myriad of new opportunities arising from the Big Data paradigm, we propose a model to enhance user experience. Our adaptive user-centric model capitalizes on fluidity of online and offline realms and autonomous environments that are sensitive to the changing data fluxes. This model is based on a prototype of an ad hoc media company which for a more than a decade has been using social media to enhance user experience. We propose a prototype that expands an existing vision of Big Data by interlinking fluctuating social media streams and external web information with the media company’s data. Theoretically we focused on the veracity and value as the Big Data constructs being the most pertinent to argue for a usercentric perspective.

Since every day millions of posts are published inside the blogosphere a huge collection of web documents develops. Clustering this ever-changing collection is a very time consuming task. Therefore some certain challenges has to be accomplished because a clustering cannot be executed from scratch all the time. The presented fuzzy, incremental and hierarchical clustering algorithm tries to succeed these challenges with both, clustering terms and documents in a meaningful way and keep them up-to-date all the time. Furthermore we take a critical look at the performance, which is crucial on such a live data collection.

Database applications are used in many elds, for example, in education, mathematics, and hospitals. Clinical database is one type of database. It is often used in hospitals, medicine factories, and colleges of medicine. Some clinical databases include patients symptoms and prescriptions. Since clinical databases cover such important information, they help researchers develop new theories and methods of using drugs. At the same time, they are used to predict the risk of death; however, few of the databases are combined to be considered in single cases of patients. In this paper, we create a relational clinical database used by doctors choosing drugs for a patient. The results of comparing testing cases by using either single clinical database or the relational databases will be reported in this paper. They show that the relational databases provide more details to be considered in each case.

When it comes to transferring Big Data, there are two main areas of concern: network performance and storage performance. The primary focus of recent work has been devoted to the problems of network connectivity and bandwidth. Different transfer techniques have been proposed to quickly move massive amounts of data between computers. The goal of these techniques is to maximize bandwidth consumption by any means necessary. The network performance of these techniques has been analyzed; however their impact on storage performance is not thoroughly investigated. In this study, Big Data transfers are evaluated from the storage viewpoint. Particular attention is focused on the granularity of request sizes issued to a storage node. This paper illustrates that there is a significant impact on performance when small portions of a data set are requested in place of a single large request.
 

Resource discovery is one of the most important underpinning problems behind producing a scalable, robust and efficient global infrastructure for e-Science. A number of approaches to the resource discovery and management problem have been suggested in various computational grid environments and prototypes over the last decade. Computational resources and services in modern grid and cloud environments can be modelled as an overlay network superposed on the physical network structure of the Internet and World Wide Web.

Multiple time-series data streams exist in industrial processes, commercial activities, etc. A method of mining dynamic association rules which change over time and whose meta-patterns which have only one item maybe have different length of time slot in different rules in a sliding window from multiple time-series data streams is proposed. As streams flow, the contents of streams are preprocessed for rule discovery. The preprocessing includes piecewise linear approximation, segmenting linearized time series to let each stream have only one line segment in one time slot, and then increamental cluster these line segments, symbolic representation of the data, and merging preprocessed streams into transactions from which rules are mined. After preprocessing we use a rule finding method to obtain rules. Through periodically pruning the obsolete and infrequent patterns are deleted. To differentiate the patterns of the latest generated transactions from those of historic transactions, decay factor is introduced.
 

It is important how wireless hosts find other hosts efficiently for load and web service purposes because hosts in an ad-hoc network moves dynamically. This paper proposes a three-tier architecture, which includes content network, social network and service network. It presents a new structure for web and load
services in ad-hoc computer networks, which is a new system architecture using SOA (Service Oriented Architecture) and Web 2.0 concepts to implement functions for web and load services. It is a three-tier system structure based on Web service functions to implement services seeking and load distribution. Furthermore, this project would construct a knowledge sharing and learning platform based on the mentioned threetier architecture. Different communities can provide their services to each other using our
new knowledge platform and this forms a “virtual community.” This will leads to our desired accomplishments of “service reusability” and “service innovation” too. In addition, it can also propose the frameworks of new SOA, and evolve in other application in web2.0 style, and further more provides a platform and more resources in order to enhance the interactions between academia and industry.

In this paper, we propose a new approach to manage data distribution in a relationnal data warehouse environment. This approach deals with the dynamic data splitting on a set of interconnected machines. The data distribution that we consider is different from the “classical” one which depends on the data use. The distribution in our approach consists in splitting data when the machine reaches his storage capacity limit. This distribution assures the scalability and exploits the storage and processing resources available in the organization using the data warehouse. It is worth noting that our approach is based on a multi-agent model mixed with the scalability distribution proposed by the Scalable Distributed Data Structures. In this paper, we focus on the global dynamic for the data splitting operation based on Branch and Bound algorithm.

In the era of information explosion, enormous amounts of data have become available on hand to decision makers. Big data refers to datasets that grow so huge that they become difficult to handle using traditional tools and techniques. Due to the rapid growth of such data, solutions need to be studied and provided in order to handle and extract value and knowledge from these datasets. Such value can only be provided by using big data analytics, which is the application of advanced analytics techniques on big data. This paper aims to analyze the different methods and tools which can be applied to big data, as well as the opportunities provided and the challenges which much be faced.

Collaborative filtering (CF) is an effective technique addressing the information overload problem, where each user is represented as a set of rating scores on items. Given a target user, conventional CF algorithms measure similarity between two users by utilizing each pair of rating scores on common rated items but discarding scores rated by either of them. In this paper, we called the former as dual ratings while the latter as singular ratings. Our experiments show that only about 10% ratings are dual ones and can be used for similarity evaluation while the left 90% are singular ones and discarded. For making full use of the limited data resource, in this paper, we present SingCF, which attempts to incorporate singular ratings for accuracy improvement of CF algorithms. In particular, we first estimate the unrated scores for singular ratings and transform them into dual ones. Then we perform a CF process to discover neighborhood users and make predictions for each user. Experiments in comparison with the state-of-the-art methods demonstrate the promise of our approach.

Cloud computing provides the benefits of scalability, agility, reliability, security, performance, and maintenance to enterprises and has emerged to become a reality. While cloud computing brings in many benefits it also introduces several security issues. One of the most serious security issues is the side channel attacks which are attacks based on information gained from the physical implementation of a machine. In this paper, we aim at providing a platform to prevent various types of side channel attacks, e.g., cache-based and timing attacks.

The current tourism service is transferring from the passive mode to active one. More and more people expect to schedule the tourism plan by themselves according to the personalized requirements. This paper proposes a service organization approach of cloud community based on destination. The approach firstly classifies the tourism web services by destination and type etc. and clusters them by service function. Then by the aid of user preference model all the required services are gotten and clustered to be service pools with the same function. And the dynamic binding is also accomplished through service pool. To satisfy the requirement of service composition, the paper designs an individual service selecting algorithm based on the community. The experiments demonstrate that the approach of cloud service community can greatly improve service discovery efficiency and precision. And with the completion of the user preference library, the precision of service recommending also has a great improvement.