The increasing number of devices owned by a single user makes it increasingly difficult to access, organise and visualise personal information (PI), i.e. documents and media, across these devices. The primary method that is currently used to organise and visualise PI is the hierarchical folder structure, which is a familiar and widely used means to manage PI. However, this hierarchy does not effectively support personal information management (PIM) across multiple devices. Current solutions, such as the Personal Information Dashboard and Stuff I’ve Seen, do not support PIM across multiple devices. Alternative PIM tools, such as Dropbox and TeamViewer, attempt to provide a means of accessing PI across multiple devices, but these solutions also suffer from several limitations. The aim of this research was to investigate to what extent enhanced information visualisation (IV) techniques could be used to support accessing PI across multiple devices. An interview study was conducted to identify how PI is currently managed across multiple devices. This interview study further motivated the need for a tool to support visualising PI across multiple devices and identified requirements for such an IV tool. Several suitable IV techniques were selected and enhanced to support PIM across multiple devices. These techniques comprised an Overview using a nested circles layout, a Tag Cloud and a Partition Layout, which used a novel set-based technique. A prototype, called MyPSI, was designed and implemented incorporating these enhanced IV techniques. The requirements and design of the MyPSI prototype were validated using a conceptual walkthrough. The design of the MyPSI prototype was initially implemented for a desktop or laptop device with mouse-based interaction. A sample personal space of information (PSI) was used to evaluate the prototype in a controlled user study. The user study was used to identify any usability problems with the MyPSI prototype. The results were highly positive and the participants agreed that such a tool could be useful in future. No major problems were identified with the prototype. The MyPSI prototype was then implemented on a mobile device, specifically an Android tablet device, using a similar design, but supporting touch-based interaction. Users were allowed to upload their own PSI using Dropbox, which was visualised by the MyPSI prototype. A field study was conducted following the Multi-dimensional In-depth Long-term Case Studies approach specifically designed for IV evaluation. The field study was conducted over a two-week period, evaluating both the desktop and mobile versions of the MyPSI prototype. Both versions received positive results, but the desktop version was slightly preferred over the mobile version, mainly due to familiarity and problems experienced with the mobile implementation. Design recommendations were derived to inform future designs of IV tools to support accessing PI across multiple devices. This research has shown that IV techniques can be enhanced to effectively support accessing PI across multiple devices. Future work will involve customising the MyPSI prototype for mobile phones and supporting additional platforms.

An index in a database system interacts with many of the software modules in the system. For systems supporting a wide range of index structures, interfacing the index code with the rest of the system poses a great problem. The problems are an order of magnitude more for adding new access methods to the system. These problems could be reduced manifold if common interfaces could be specified for different access methods. It would be even better, if these interfaces could be made database-system independent. This thesis addresses the problem of defining generic index interfaces for access methods in database systems. It concentrates on two specific issues: First, specification of a complete set of abstract interfaces that would work for all access methods and for all database systems. Second, optimized query processing for all data types including userdefined data types. An access method in a database system can be considered to be made up of three specific parts: Upper interfaces, lower interfaces, and type interfaces. An access method interacts with a database system through its upper interfaces, lower interfaces and type interfaces. Upper interfaces consist of the functions an index provides to a database system. Lower interfaces are the database-system dependent software modules an index has to interact with, to accomplish any system related functions. Type interfaces consist of the set of functions an index uses, which interpret the data type. These three parts together characterize an access method in a database system. This splitting of an access method makes it possible to define generic interfaces. In this thesis, we will discuss each of these three different interfaces in detail, identify functionalities and design clear interfaces. The design of these interfaces promote development of type-independent and database-system independent access methods.

The first step of the program is to input the statistical information concerning the relations of th· database. This information is stored in the log file and the file matrix data structures. Next, the query itself is read and stored in an array called the query matrix. The program examines the various fields of this matrix and decides which relations in the database are necessary to answer the query. For these relations it determines those attributes which should be eliminated and those which should be preserved for further processing. The key attributes are identified and are projected along with the other attributes. After the initial projection is completed the sizes of the new temporary relations are evaluated and stored in the appropriate fields of the file matrix structure. The program then examines that part of the query which contains the various restrictions on the attributes. The values of the attributes are sorted and those values which do not match the restrictions are eliminated from the log file. Again, the sizes of the new relations are estimated according to the method described by Egyhazy et al. [6]. A second projection is performed to eliminate attributes which were required by the selection phase but are not part of the final answer to the query. The remaining relations are those relations which need to be joined to form a relation with the required information. In order to decide upon which relations to join, a special table, the join matrix, is created. This table contains pairs of relations which have common attributes and common values and therefore are joinable. The LP algorithm is used to determine the least expensive join out of all the possible joins. This process is repeated until all of the relations are joined to form a single relation which answers the query. As in the case of projection and selection the size of the temporary relations after each join is estimated. As a last step, we remove the key attributes which helped in joining the files but are not part of the answer to the query. / Master of Engineering

A dissertation submitted to the Faculty of Science, University of the Witwatersrand, Johannesburg, in fulfilment of the requirements for the degree of Master of Science., University of the Witwatersrand, Johannesburg, 2018 / Social network sites such as Twitter have created vibrant and diverse communities in which users express their opinions and views on a variety of topics such as politics. Extensive research has been conducted in countries such as Ireland, Germany and the United States, in which text mining techniques have been used to obtain information from politically oriented tweets. The purpose of this research was to determine if text mining techniques can be used to uncover meaningful information from a corpus of political tweets collected during the 2014 South African General Election. The Twitter Application Programming Interface was used to collect tweets that were related to the three major political parties in South Africa, namely: the African National Congress (ANC), the Democratic Alliance (DA) and the Economic Freedom Fighters (EFF). The text mining techniques used in this research are: sentiment analysis, clustering, association rule mining and word cloud analysis. In addition, a correlation analysis was performed to determine if there exists a relationship between the total number of tweets mentioning a political party and the total number of votes obtained by that party. The VADER (Valence Aware Dictionary for sEntiment Reasoning) sentiment classifier was used to determine the public’s sentiment towards the three main political parties. This revealed an overwhelming neutral sentiment of the public towards the ANC, DA and EFF. The result produced by the VADER sentiment classifier was significantly greater than any of the baselines in this research. The K-Means cluster algorithm was used to successfully cluster the corpus of political tweets into political-party clusters. Clusters containing tweets relating to the ANC and EFF were formed. However, tweets relating to the DA were scattered across multiple clusters. A fairly strong relationship was discovered between the number of positive tweets that mention the ANC and the number of votes the ANC received in election. Due to the lack of data, no conclusions could be made for the DA or the EFF. The apriori algorithm uncovered numerous association rules, some of which were found to be interest- ing. The results have also demonstrated the usefulness of word cloud analysis in providing easy-to-understand information from the tweet corpus used in this study. This research has highlighted the many ways in which text mining techniques can be used to obtain meaningful information from a corpus of political tweets. This case study can be seen as a contribution to a research effort that seeks to unlock the information contained in textual data from social network sites. / MT 2018

<div>Vast amounts of sensitive personal information are collected by companies, institutions and governments. A key technological challenge is how to effectively extract knowledge from data while preserving the privacy of the individuals involved. In this dissertation, we address this challenge from the perspective of privacy-preserving data collection and analysis. We focus on investigation of a technique called local differential privacy (LDP) and studied several aspects of it. </div><div><br></div><div><br></div><div>In particular, the thesis serves as a comprehensive study of multiple aspects of the LDP field. We investigated the following seven problems: (1) We studied LDP primitives, i.e., the basic mechanisms that are used to build LDP protocols. (2) We then studied the problem when the domain size is very big (e.g., larger than $2^{32$), where finding the values with high frequency is a challenge, because one needs to enumerate through all values. (3) Another interesting setting is when each user possesses a set of values, instead of a single private value. (4) With the basic problems visited, we then aim to make the LDP protocols practical for real-world scenarios. We investigated the case where each user's data is high-dimensional (e.g., in the census survey, each user has multiple questions to answer), and the goal is to recover the joint distribution among the attributes. (5) We also built a system for companies to issue SQL queries over the data protected under LDP, where each user is associated with some public weights and holds some private values; an LDP version of the values is sent to the server from each user. (6) To further increase the accuracy of LDP, we study how to add post-processing steps to protocols to make them consistent while achieving high accuracy for a wide range of tasks, including frequencies of individual values, frequencies of the most frequent values, and frequencies of subsets of values. (7) Finally, we investigate a different model of LDP which is called the shuffler model. While users still use LDP algorithms to report their sensitive data, now there exists a semi-trusted shuffler that shuffles the users' reports and then send them to the server. This model provides better utility but at the cost of requiring more trust that the shuffler should not collude with the server.</div>

XML probabiliste est un modèle probabiliste pour les bases de données incertaines semi-structurées, avec des applications telles que l'intégration incertaine de données, l'extraction d'informations ou le contrôle probabiliste de versions. Nous explorons dans cette thèse une solution efficace pour l'évaluation des requêtes tree-pattern avec jointures sur ces documents, ou, plus précisément, pour l'approximation de la probabilité d'une requête booléenne sur un document probabiliste. L'approche repose sur, d'une part, la production de la provenance probabiliste de la requête posée, et, d'autre part, la recherche d'une stratégie optimale pour estimer la probabilité de cette provenance. Cette deuxième partie s'inspire des approches des optimiseurs de requêtes: l'exploration de différents plans d'évaluation pour différentes parties de la formule et l'estimation du coût de chaque plan, suivant un modèle de coût établi pour les algorithmes de calcul utilisés. Nous démontrons l'efficacité de cette approche sur des jeux de données utilisés dans des travaux précédents sur l'interrogation des bases de données XML probabilistes, ainsi que sur des données synthétiques. / Probabilistic XML is a probabilistic model for uncertain tree-structured data, with applications to data integration, information extraction, or uncertain version control. We explore in this dissertation efficient algorithms for evaluating tree-pattern queries with joins over probabilistic XML or, more specifically, for approximating the probability of each item of a query result. The approach relies on, first, extracting the query lineage over the probabilistic XML document, and, second, looking for an optimal strategy to approximate the probability of the propositional lineage formula. ProApproX is the probabilistic query manager for probabilistic XML presented in this thesis. The system allows users to query uncertain tree-structured data in the form of probabilistic XML documents. It integrates a query engine that searches for an optimal strategy to evaluate the probability of the query lineage. ProApproX relies on a query-optimizer--like approach: exploring different evaluation plans for different parts of the formula and predicting the cost of each plan, using a cost model for the various evaluation algorithms. We demonstrate the efficiency of this approach on datasets used in a number of most popular previous probabilistic XML querying works, as well as on synthetic data. An early version of the system was demonstrated at the ACM SIGMOD 2011 conference. First steps towards the new query solution were discussed in an EDBT/ICDT PhD Workshop paper (2011). A fully redesigned version that implements the techniques and studies shared in the present thesis, is published as a demonstration at CIKM 2012. Our contributions are also part of an IEEE ICDE

No description available.

No description available.

With the proliferation of inexpensive cameras and the availability of high-speed wired and wireless networks, networks of distributed cameras are becoming an enabling technology for a broad range of interdisciplinary applications in domains such as public safety and security, manufacturing, transportation, and healthcare. Today’s live video processing systems on networks of distributed cameras, however, are designed for specific classes of applications. To provide a generic query processing platform for applications of distributed camera networks, we designed and implemented a new class of general purpose database management systems, the live video database management system (LVDBMS). We view networked video cameras as a special class of interconnected storage devices, and allow the user to formulate ad hoc queries over real-time live video feeds. In the first part of this dissertation, an Internet scale framework for sharing and dissemination of general sensor data is presented. This framework provides a platform for general sensor data to be published, searched, shared, and delivered across the Internet. The second part is the design and development of a Live Video Database Management System. LVDBMS allows users to easily focus on events of interest from a multitude of distributed video cameras by posing continuous queries on the live video streams. In the third part, a distributed in-memory database approach is proposed to enhance the LVDBMS with an important capability of tracking objects across cameras

No description available.