Graph embedding, aiming to learn low-dimensional representations for nodes in graphs, has attracted increasing attention due to its critical application including node classification, link prediction and clustering in social network analysis. Most existing algorithms for graph embedding only rely on the topology information and fail to use the copious information in nodes as well as edges. As a result, their performance for many tasks may not be satisfactory. In this thesis, we proposed a novel and general framework for graph embedding with rich text information (GERI) through constructing a heterogeneous network, in which we integrate node and edge content information with graph topology. Specially, we designed a novel biased random walk to explore the constructed heterogeneous network with the notion of flexible neighborhood. Our sampling strategy can compromise between BFS and DFS local search on heterogeneous graph. To further improve our algorithm, we proposed semi-supervised GERI (SGERI), which learns graph embedding in an discriminative manner through heterogeneous network with label information. The efficacy of our method is demonstrated by extensive comparison experiments with 9 baselines over multi-label and multi-class classification on various datasets including Citeseer, Cora, DBLP and Wiki. It shows that GERI improves the Micro-F1 and Macro-F1 of node classification up to 10%, and SGERI improves GERI by 5% in Wiki.

This study (Wassell & Young 2007; Wassell et al., 2010a) shows that behenic (C22:0) fatty acid rich Monoacylglycerol (MAG), or its significant inclusion, has a pronounced effect on crystallisation (Wassell et al., 2010b; 2012; Young et al., 2008) and interfacial kinetics (3.0; 4.0). New interfacial measurements demonstrate an unusual surface-interactive relationship of long chain MAG compositions, with and without Polyglycerol Polyricinoleate (PGPR). A novel MAG synthesised from Moringa oleifera Triacylglycerol (TAG) influenced textural behaviour of water-in-oil (W/O) emulsions and anhydrous TAG systems (4.0: 5.0; 6.0). Emulsifier mixtures of PGPR and MAG rich in C18:1 / 18:2 and C16:0 / C18:0 do not decrease interfacial tension compared with PGPR alone. Only those containing MAG with significant proportion of C22:0 impacted interfacial behaviour. A mixture of C22:0 based MAG and PGPR results with decreasing tension from ~20°C and is initially dominated by PGPR, then through rearrangement, the surface is rapidly dominated by C22:0 fatty acids. A Moringa oleifera based MAG showed unusual decreased interfacial behaviour not dissimilar to PGPR. All other tested MAG (excluding a C22:0 based MAG), irrespective of fatty acid composition resulted with high interfacial tension values across the measured temperature spectrum (50°C to 5°C). A relative decrease of interfacial tension, with decreased temperature, was greater, the longer the chain length (Krog & Larsson 1992). Moreover, results from bulk and interfacial rheology showed that the presence of C22:0 based MAG has a pronounced effect on both elastic modulus (G’) and viscous modulus (G’’). Through a multidisciplinary approach, results were verified in relevant product applications. By means of ultrasonic velocity profiling with pressure difference (UVP-PD) technique, it was possible to examine the effect of a C22:0 based MAG in an anhydrous TAG system whilst in a dynamic non-isothermal condition (3.0). The non-invasive UVP-PD technique conclusively validated structural events. The application of a Moringa oleifera based MAG in low TAG (35% - 41%), W/O emulsions, results in high emulsion stability without a co-surfactant (PGPR). The bi-functional behaviour of Moringa oleifera based MAG is probably attributed to miscibility (Ueno et al., 1994) of its fatty acids, ranging ~30% of saturated fatty acids (SAFA), with ~70% of C18:1 (5.0). It is concluded that the surface-interactive behaviour of Moringa oleifera based MAG, is attributed to approximately 10% of its SAFA commencing from C20:0. When examined separately and compared, results showed that physical effect of a Moringa oleifera based MAG was not dissimilar to PGPR, influencing the crystallisation kinetics of the particular anhydrous TAG system. When either was combined with a C22:0 rich MAG, enhanced gelation onset and strong propensity to form dendrite structure occurred (5.0). Macrobeam and synchrotron radiation microbeam small angle x-ray diffraction (SR-μ-SAXD) was utilized (6.0) to assess behavior of C22:0 rich MAG, with and without PGPR (Wassell et al., 2012). The C22:0 based MAG combined with PGPR promoted TAG crystallisation as observed by differential scanning calorimetry (DSC). Polarised optical microscopy (POM) observations indicated that C22:0 based MAG eliminates formation of large crystal aggregates, resulting in the likely formation of tiny Pickering TAG / MAG crystals (6.0). It is concluded that the presence and interactive behaviour of Pickering surface-active MAG, is strongly linked to increased fatty acid chain length, which induce increased textural resilience owing to viscoelasticity (4.0; 5.0). A multidisciplinary approach was able to verify structuring behaviour (4.0; 5.0), using multiple analyses (Wassell et al., 2010b; 2012; Young et al., 2008). Novel structuring solutions in reduced TAG based systems have been provided (4.0; 5.0). This study both enhances current understanding of structuring in low TAG W/O emulsions and has led to novel MAG compositions, which address emulsification, structuring and texture in TAG based food systems (Wassell et al., 2010a; 2012a; 2012b; 2012c; 2012d; 2012e; Bech et al., 2013).

Web applications have come a long way both in terms of adoption to provide information and services and in terms of the technologies to develop them. With the emergence of richer and more advanced technologies such as Ajax, web applications have become more interactive, responsive and user friendly. These applications, often called Rich Internet Applications (RIAs) changed the traditional web applications in two primary ways: Dynamic manipulation of client side state and Asynchronous communication with the server. At the same time, such techniques also introduce new challenges. Among these challenges, an important one is the difficulty of automatically crawling these new applications. Crawling is not only important for indexing the contents but also critical to web application assessment such as testing for security vulnerabilities or accessibility. Traditional crawlers are no longer sufficient for these newer technologies and crawling in RIAs is either inexistent or far from perfect. There is a need for an efficient crawler for web applications developed using these new technologies. Further, as more and more enterprises use these new technologies to provide their services, the requirement for a better crawler becomes inevitable. This thesis studies the problems associated with crawling RIAs. Crawling RIAs is fundamentally more difficult than crawling traditional multi-page web applications. The thesis also presents an efficient RIA crawling strategy and compares it with existing methods.

Nitrogen (N) is critical to the quantity and quality of agricultural yields. Excess N fertilization is costly, both economically and environmentally (nitrate leaching, eutrophication, greenhouse gas release, soil degradation). This research identifies zones that could substitute the field-long N-rich strips by using spatial analysis of the nitrogen sufficiency index (NSI) and the relation with Apparent Electrical Conductivity (ECa), Elevation, Slope and Soil. NSI calculated from ECa grouped into three classes was capable of minimizing the effects on NDVI. Correlation coefficients (R) between three-class NSI and NSI calculated from the nearest ECA values were very high for all the fields with values between 0.82< R <0.94, with the highest coefficients associated with fields in 2005 and 2007. Meanwhile, three-class NSI coefficients were consistently significant in relation to the NSI reference, with an average of R=0.79 for all the fields. The highest coefficient was detected for 2007, with R=0.89, whereas the lowest values were associated with 2006 (R=0.67). In the case of elevation grouped into four classes, the correlation results were not statistically significant, with overall average values of R<0.70. The maps elaborated from the NSI for ECa grouped into three classes show a high level of accuracy compared to the NSI reference map. The new N-rich zones not only can contribute to mitigating the environmental impact of agricultural practices (reducing 77% of N inputs) but also be an accurate source of data for the analysis of NSI and within-field N variability.

Matrix-rich strata (20-70% mud matrix) have been increasingly recognized in deep-marine systems. These beds are thought to be deposited from mud-rich flows in a distal basin-floor setting; however they remain poorly understood, partly because details of lateral lithological changes are poorly known. In this study, matrix-rich strata are common in proximal basin-floor strata of the Neoproterozoic Windermere Supergroup. The objective of this thesis is to provide detailed description and interpretation of the lithological and mineralogical make-up and lateral facies trends of matrix-rich strata in in a unit 40 m thick and 800 m wide. Here, stratigraphic and petrographic analyses identified five facies: classic turbidites; sandstones; clayey sandstones; sandy claystones and fine-grained banded couplets, which laterally are arranged systematically from matrix-poor sandstones to thin-bedded turbidites. This lateral change is interpreted to represent a depositional continuum along the margins of an efflux jet that formed immediately downflow of an avulsion node.

The generated HTTP traffic of users' interactions with a Web application can be logged for further analysis. In this thesis, we present the ``Session Reconstruction'' problem that is the reconstruction of user interactions from recorded request/response logs of a session. The reconstruction is especially useful when the only available information about the session is its HTTP trace, as could be the case during a forensic analysis of an attack on a website. New Web technologies such as AJAX and DOM manipulation have provided more responsive and smoother Web applications, sometimes called ``Rich Internet Applications''(RIAs). Despite the benefits of RIAs, the previous session reconstruction methods for traditional Web applications are not effective anymore. Recovering information from a log in RIAs is significantly more challenging as compared with classical Web applications, because the HTTP traffic contains often only application data and no obvious clues about what the user did to trigger that traffic. This thesis studies applying different techniques for efficient reconstruction of RIA sessions. We define the problem in the context of the client/server applications, and propose a solution for it. We present different algorithms to make the session reconstruction possible in practice: learning mechanisms to guide the session reconstruction process efficiently, techniques for recovering user-inputs and handling client-side randomness, and also algorithms for detections of actions that do not generate any HTTP traffic. In addition, to further reduce the session reconstruction time, we propose a distributed architecture to concurrently reconstruct a RIA session over several nodes. To measure the effectiveness of our proposed algorithms, a prototype called D-ForenRIA is implemented. The prototype is made of a proxy and a set of browsers. Browsers are responsible for trying candidate actions on each state, and the proxy, which contains the observed HTTP trace, is responsible for responding to browsers' requests and validating attempted actions on each state. We have used this tool to measure the effectiveness of the proposed techniques during session reconstruction process. The results of our evaluation on several RIAs show that the proposed solution can efficiently reconstruct use-sessions in practice.

The main objective of this work is to analyze the concepts of Business Intelligence to explore their implementation with MS Silverlight (RIA) to find out advantages and disadvantages of BI solutions using RIA and verified them on a case study of a fictitious bank. The work is divided into theoretical and practical part. The theoretical part is analyzing BI evolution and current trends. In the practical part the dashboard has been created for a fictitious bank management in MS Silverlight. The contribution of the author of this work lies mainly in the application of theoretical BI concepts in case study of a fictitious bank. There are elements of an MS Silverlight and processing techniques on the client that were used to show usability in BI. In the thesis was used Data Binding and MS Silverlight for predictive analysis in BI.

Mining activities, particularly coal mining, lead to the excavation of large volumes of pyrite rich soils. When exposed to air (oxygen) and water these pyrite complexes are oxidised to form highly acidic and corrosive wastewaters collectively termed acid mine drainage (AMD). Containing elevated levels of sulphates (SO42-) and toxic dissolved heavy metals, AMD seeping from mining sites, active or abandoned, poses a major environmental risk to aquatic bio-systems – not only in South Africa but globally. Chemical neutralization of AMD is expensive and often challenging. Biological sulphate reduction provides a promising and cheaper alternative to the treatment of sulphate rich wastewaters. Little, however, is known about the microbial communities involved in biological treatment systems and the effect of external factors thereon. Studying microorganisms in their natural environment is extremely difficult. The limitations of culture-based methods only provide a limited insight into the bacterial diversity of natural habitats and the microbial communities present. With the progressive advances in molecular biology, non culture-based tools such as DGGE, FISH and more recently t-RFLP allow easier and much more accurate studies of microbial communities within their natural as well as man-made environments. This study specifically investigated the use of t-RFLP to study microbial communities and dynamics within sulphate removing bioreactors. The set up and optimization of a t-RFLP system to specifically study microbial communities from sulphate removing bioreactors were investigated and the applicability of t-RFLP demonstrated. Copyright / Dissertation (MSc)--University of Pretoria, 2009. / Microbiology and Plant Pathology / unrestricted

Cloning and molecular characterization of new genes constitutes a useful approach in studying the symbiotic interactions between the model plant Medicago truncatula and Synorhizobium meliloti. Large numbers of expressed sequence tags (ESTs) available for Medicago truncatula, along with numerous cDNA, oligonucleotides, and Affimetrix DNA microarray chips, represent useful tools for gene discovery. In an attempt to identify a new gene that might be involved in the process of nodulation in Medicago truncatula, preliminary data reported by Fedorova et al. (2002), who identified 340 putative gene products or tentative consensus sequences (TCs) expressed only in nodules, was used. This research was focused on TC33166 (TC103185), which has 3 ESTs in the TC, and whose strongest BLASTX hit of TC103185 is the incomplete root hair elongation (IRE) protein kinase-like protein (NP_192429) from Arabidopsis thaliana. The Arabidopsis IRE gene is required for normal root hair growth, and a role in apical growth was suggested (Oyama et al., 2002). Infection thread growth can be looked at as an inward growth of the root hair. Thus, TC103185 was a good candidate for identifying a gene that may be involved in early events of nodulation. MtIRE (GenBank accession AC122727) is organized in 17 exons and 16 introns, similarly to the Arabidopsis IRE gene. MtIRE is a new member of the IRE family and it is a putative Ser/Thr protein kinase. MtIRE is a nodule- and flower-specific gene, suggesting that nodulation may have recruited it from other developmental processes. MtIRE is likely to be involved in the invasion process, or in the maturation of the symbiosome, or of the cells that contain rhizobia, rather than infection thread initiation and elongation or in nitrogen fixation. Nodule invasion precedes the onset of MtIRE expression and the expression pattern changes in time within the nodule. RNA interference results support MtIRE expression data and suggest a possible role in preventing extensive defense responses. Our study demonstrates the existence of an Arabidopsis IRE homolog in Medicago truncatula root nodules with an entirely new function and regulation.

Platelet-rich plasma (PRP) is the cell free plasma, which has an enriched concentration of platelets and clotting factors with the ability to enhance the natural healing process. PRP is often used by physicians in an office setting to accelerate the healing of a variety of sports related injuries, chronic wounds and enhance skin rejuvenation. PRP mimics the wound healing cascade by enhancing the recruitment, proliferation and differentiation of cells involved in tissue regeneration. Although PRP is used to enhance healing, the efficacy thereof is debated as no clear-cut set of parameters is available that device manufacturers and protocols should follow. The lack of uniformity in the PRP preparation methods results in differing PRP volume, platelet contents and unavoidably platelet-derived growth factors. Therefore, the aim of this study was to develop a simple and rapid method for preparing autologous PRP in an office setting using a tabletop centrifuge for point-of-care use. The simplified preparation procedure involved a single centrifugation step of 18 ml of whole blood, which sufficiently enriched the platelet content in the PRP fraction. As activated platelets express and release growth factors and cytokines that mediate the different phases of the wound healing cascade, the extracted PRP fraction was activated with an ethanol, calcium chloride (CaCl2) and platelet poor plasma (PPP) preparation in glass containers, without the collection of additional blood as required in some protocols. The activated PRP formed a fibrin clot, trapping the degranulating platelets and its released growth factors. The concentration of TGF- 1 obtained from the fibrin clot was 45.49 ± 3.80 ng/ml, in range with the available literature. During the in vitro studies, the extracted PRP by the developed method was able to significantly induce cell proliferation in a dose dependent manner. Cells enumerated with the crystal violet assay indicated that the cells treaded with 5% or 10% PRP significantly increased the percentage of viable cells to 165-176% and 156-158%, when compared to the positive controls. Cells enumerated with the MTT-assay indicated that the cells treaded with 5% or 10% PRP increased the percentage of viable cells to 79-91% and 87-105% which is comparable to that of the positive control. Data from the cellular proliferation assays indicate that sufficient plateletderived growth factors had been obtained with the preparation procedure. Furthermore, data from the in vivo studies indicated that the extracted PRP was able to augment soft tissue regeneration and bone formation. Treatment with the activated PRP resulted in symptom reduction and accelerated healing of various injuries. The simplified preparation and the use of the provided study product packaged in a kit developed during this study will enable physicians to easily obtain autologous PRP, in an office setting for point-of-care use, with the ability to induce tissue regeneration. / Dissertation (MSc)--University of Pretoria, 2013 / gm2014 / Pharmacology / unrestricted