Monitoring the preservation of security and dependability (S&D) properties of complex software systems is widely accepted as a necessity. Basic monitoring can detect violations but does not always provide sufficient information for deciding what the appropriate response to a violation is. Such decisions often require additional diagnostic information that explains why a violation has occurred and can, therefore, indicate what would be an appropriate response action to it. In this thesis, we describe a diagnostic procedure for generating explanations of violations of S&D properties developed as extension of a runtime monitoring framewoek, called EVEREST. The procedure is based on a combination of abductive and evidential reasoning about violations of S&D properties which are expressed in Event Calculus.

The effective integration of knowledge representation, reasoning and learning into a robust computational model is one of the key challenges in Computer Science and Artificial Intelligence. In particular, temporal models have been fundamental in describing the behaviour of Computational and Neural-Symbolic Systems. Furthermore, knowledge acquisition of correct descriptions of the desired system’s behaviour is a complex task in several domains. Several efforts have been directed towards the development of tools that are capable of learning, describing and evolving software models. This thesis contributes to two major areas of Computer Science, namely Artificial Intelligence (AI) and Software Engineering. Under an AI perspective, we present a novel neural-symbolic computational model capable of representing and learning temporal knowledge in recurrent networks. The model works in integrated fashion. It enables the effective representation of temporal knowledge, the adaptation of temporal models to a set of desirable system properties and effective learning from examples, which in turn can lead to symbolic temporal knowledge extraction from the corresponding trained neural networks. The model is sound, from a theoretical standpoint, but is also tested in a number of case studies. An extension to the framework is shown to tackle aspects of verification and adaptation under the SE perspective. As regards verification, we make use of established techniques for model checking, which allow the verification of properties described as temporal models and return counter-examples whenever the properties are not satisfied. Our neural-symbolic framework is then extended to deal with different sources of information. This includes the translation of model descriptions into the neural structure, the evolution of such descriptions by the application of learning of counter examples, and also the learning of new models from simple observation of their behaviour. In summary, we believe the thesis describes a principled methodology for temporal knowledge representation, learning and extraction, shedding new light on predictive temporal models, not only from a theoretical standpoint, but also with respect to a potentially large number of applications in AI, Neural Computation and Software Engineering, where temporal knowledge plays a fundamental role.

Cellular signalling is an important area of study in biology. Signalling pathways are well-known abstractions that explain the mechanisms whereby cells respond to signals. Collections of pathways form signalling networks, and interactions between pathways in a network, known as cross-talk, enables further complex signalling behaviours. Increasingly, computational modelling and analysis is required to handle the complexity of such systems. While there are several computational modelling approaches for signalling pathways, none make cross-talk explicit. We present a modular modelling framework for pathways and their cross-talk. Networks are formed by composing pathways: different cross-talks result from different synchronisations of reactions between, and overlaps of, the pathways. We formalise five types of cross-talk and give approaches to reason about possible cross-talks in a network. The complementary problem is how to handle unstructured signalling networks, i.e. networks with no explicit notion of pathways or cross-talk. We present an approach to better understand unstructured signalling networks by modelling them as a set of signal flows through the network. We introduce the Reaction Minimal Paths (RMP) algorithm that computes the set of signal flows in a model. To the best of our knowledge, current algorithms cannot guarantee both correctness and completeness of the set of signal flows in a model. The RMP algorithm is the first. Finally, the RMP algorithm suffers from the well-known state space explosion problem. We use suitable partial order reduction algorithms to improve the efficiency of this algorithm.

The thesis investigates how persons in an enterprise organisation can be ranked in response to a query, so that those persons with relevant expertise to the query topic are ranked first. The expertise areas of the persons are represented by documentary evidence of expertise, known as candidate profiles. The statement of this research work is that the expert search task in an enterprise setting can be successfully and effectively modelled using a voting paradigm. In the so-called Voting Model, when a document is retrieved for a query, this document represents a vote for every expert associated with the document to have relevant expertise to the query topic. This voting paradigm is manifested by the proposition of various voting techniques that aggregate the votes from documents to candidate experts. Moreover, the research work demonstrates that these voting techniques can be modelled in terms of a Bayesian belief network, providing probabilistic semantics for the proposed voting paradigm. The proposed voting techniques are thoroughly evaluated on three standard expert search test collections, deriving conclusions concerning each component of the Voting Model, namely the method used to identify the documents that represent each candidate's expertise areas, the weighting models that are used to rank the documents, and the voting techniques which are used to convert the ranking of documents into the ranking of experts. Effective settings are identified and insights about the behaviour of each voting technique are derived. Moreover, the practical aspects of deploying an expert search engine such as its efficiency and how it should be trained are also discussed. This thesis includes an investigation of the relationship between the quality of the underlying ranking of documents and the resulting effectiveness of the voting techniques. The thesis shows that various effective document retrieval approaches have a positive impact on the performance of the voting techniques. Interestingly, it also shows that a `perfect' ranking of documents does not necessarily translate into an equally perfect ranking of candidates. Insights are provided into the reasons for this, which relate to the complexity of evaluating tasks based on ranking aggregates of documents. Furthermore, it is shown how query expansion can be adapted and integrated into the expert search process, such that the query expansion successfully acts on a pseudo-relevant set containing only a list of names of persons. Five ways of performing query expansion in the expert search task are proposed, which vary in the extent to which they tackle expert search-specific problems, in particular, the occurrence of topic drift within the expertise evidence for each candidate. Not all documentary evidence of expertise for a given person are equally useful, nor may there be sufficient expertise evidence for a relevant person within an enterprise. This thesis investigates various approaches to identify the high quality evidence for each person, and shows how the World Wide Web can be mined as a resource to find additional expertise evidence. This thesis also demonstrates how the proposed model can be applied to other people search tasks such as ranking blog(ger)s in the blogosphere setting, and suggesting reviewers for the submitted papers to an academic conference. The central contributions of this thesis are the introduction of the Voting Model, and the definition of a number of voting techniques within the model. The thesis draws insights from an extremely large and exhaustive set of experiments, involving many experimental parameters, and using different test collections for several people search tasks. This illustrates the effectiveness and the generality of the Voting Model at tackling various people search tasks and, indeed, the retrieval of aggregates of documents in general.

This thesis investigates the problem of obtaining overview information from complex tabular numerical data sets non-visually. Blind and visually impaired people need to access and analyse numerical data, both in education and in professional occupations. Obtaining an overview is a necessary first step in data analysis, for which current non-visual data accessibility methods offer little support. This thesis describes a new interactive parametric sonification technique called High-Density Sonification (HDS), which facilitates the process of extracting overview information from the data easily and efficiently by rendering multiple data points as single auditory events. Beyond obtaining an overview of the data, experimental studies showed that the capabilities of human auditory perception and cognition to extract meaning from HDS representations could be used to reliably estimate relative arithmetic mean values within large tabular data sets. Following a user-centred design methodology, HDS was implemented as the primary form of overview information display in a multimodal interface called TableVis. This interface supports the active process of interactive data exploration non-visually, making use of proprioception to maintain contextual information during exploration (non-visual focus+context), vibrotactile data annotations (EMA-Tactons) that can be used as external memory aids to prevent high mental workload levels, and speech synthesis to access detailed information on demand. A series of empirical studies was conducted to quantify the performance attained in the exploration of tabular data sets for overview information using TableVis. This was done by comparing HDS with the main current non-visual accessibility technique (speech synthesis), and by quantifying the effect of different sizes of data sets on user performance, which showed that HDS resulted in better performance than speech, and that this performance was not heavily dependent on the size of the data set. In addition, levels of subjective workload during exploration tasks using TableVis were investigated, resulting in the proposal of EMA-Tactons, vibrotactile annotations that the user can add to the data in order to prevent working memory saturation in the most demanding data exploration scenarios. An experimental evaluation found that EMA-Tactons significantly reduced mental workload in data exploration tasks. Thus, the work described in this thesis provides a basis for the interactive non-visual exploration of a broad range of sizes of numerical data tables by offering techniques to extract overview information quickly, performing perceptual estimations of data descriptors (relative arithmetic mean) and managing demands on mental workload through vibrotactile data annotations, while seamlessly linking with explorations at different levels of detail and preserving spatial data representation metaphors to support collaboration with sighted users.

The main objective of Information Retrieval (IR) systems is to satisfy searchers’ needs. A great deal of research has been conducted in the past to attempt to achieve a better insight into searchers’ needs and the factors that can potentially influence the success of an Information Retrieval and Seeking (IR&S) process. One of the factors which has been considered is searchers’ emotion. It has been shown in previous research that emotion plays an important role in the success of an IR&S process, which has the purpose of satisfying an information need. However, these previous studies do not give a sufficiently prominent position to emotion in IR, since they limit the role of emotion to a secondary factor, by assuming that a lack of knowledge (the need for information) is the primary factor (the motivation of the search). In this thesis, we propose to treat emotion as the principal factor in the system of needs of a searcher, and therefore one that ought to be considered by the retrieval algorithms. We present a more realistic view of searchers’ needs by considering not only theories from information retrieval and science, but also from psychology, philosophy, and sociology. We extensively report on the role of emotion in every aspect of human behaviour, both at an individual and social level. This serves not only to modify the current IR views of emotion, but more importantly to uncover social situations where emotion is the primary factor (i.e., source of motivation) in an IR&S process. We also show that the emotion aspect of documents plays an important part in satisfying the searcher’s need, in particular when emotion is indeed a primary factor. Given the above, we define three concepts, called emotion need, emotion object and emotion relevance, and present a conceptual map that utilises these concepts in IR tasks and scenarios. In order to investigate the practical concepts such as emotion object and emotion relevance in a real-life application, we first study the possibility of extracting emotion from text, since this is the first pragmatic challenge to be solved before any IR task can be tackled. For this purpose, we developed a text-based emotion extraction system and demonstrate that it outperforms other available emotion extraction approaches. Using the developed emotion extraction system, the usefulness of the practical concepts mentioned above is studied in two scenarios: movie recommendation and news diversification. In the movie recommendation scenario, two collaborative filtering (CF) models were proposed. CF systems aim to recommend items to a user, based on the information gathered from other users who have similar interests. CF techniques do not handle data sparsity well, especially in the case of the cold start problem, where there is no past rating for an item. In order to predict the rating of an item for a given user, the first and second models rely on an extension of state-of-the-art memory-based and model-based CF systems. The features used by the models are two emotion spaces extracted from the movie plot summary and the reviews made by users, and three semantic spaces, namely, actor, director, and genre. Experiments with two MovieLens datasets show that the inclusion of emotion information significantly improves the accuracy of prediction when compared with the state-of-the-art CF techniques, and also tackles data sparsity issues. In the news retrieval scenario, a novel way of diversifying results, i.e., diversifying based on the emotion aspect of documents, is proposed. For this purpose, two approaches are introduced to consider emotion features for diversification, and they are empirically tested on the TREC 678 Interactive Track collection. The results show that emotion features are capable of enhancing retrieval effectiveness. Overall, this thesis shows that emotion plays a key role in IR and that its importance needs to be considered. At a more detailed level, it illustrates the crucial part that emotion can play in • searchers, both as a primary (emotion need) and secondary factor (influential role) in an IR&S process; • enhancing the representation of a document using emotion features (emotion object); and finally, • improving the effectiveness of IR systems at satisfying searchers’ needs (emotion relevance).

Since their emergence within the past decade, which has seen wireless networks being adapted to enable mobility, wireless networks have become increasingly popular in the world of computer research. A Mobile Ad hoc Network (MANET) is a collection of mobile nodes dynamically forming a temporary network without the use of any existing network infrastructure. MANETs have received significant attention in recent years due to their easiness to setup and to their potential applications in many domains. Such networks can be useful in situations where there is not enough time or resource to configure a wired network. Ad hoc networks are also used in military operations where the units are randomly mobile and a central unit cannot be used for synchronization. The shared media used by wireless networks, grant exclusive rights for a node to transmit a packet. Access to this media is controlled by the Media Access Control (MAC) protocol. The Backoff mechanism is a basic part of a MAC protocol. Since only one transmitting node uses the channel at any given time, the MAC protocol must suspend other nodes while the media is busy. In order to decide the length of node suspension, a backoff mechanism is installed in the MAC protocol. The choice of backoff mechanism should consider generating backoff timers which allow adequate time for current transmissions to finish and, at the same time, avoid unneeded idle time that leads to redundant delay in the network. Moreover, the backoff mechanism used should decide the suitable action to be taken in case of repeated failures of a node to attain the media. Further, the mechanism decides the action needed after a successful transmission since this action affects the next time backoff is needed. The Binary exponential Backoff (BEB) is the backoff mechanisms that MANETs have adopted from Ethernet. Similar to Ethernet, MANETs use a shared media. Therefore, the standard MAC protocol used for MANETs uses the standard BEB backoff algorithms. The first part of this work, presented as Chapter 3 of this thesis, studies the effects of changing the backoff behaviour upon a transmission failure or after a successful transmission. The investigation has revealed that using different behaviours directly affects both network throughput and average packet delay. This result indicates that BEB is not the optimal backoff mechanism for MANETs. Up until this research started, no research activity has focused on studying the major parameters of MANETs. These parameters are the speed at which nodes travel inside the network area, the number of nodes in the network and the data size generated per second. These are referred to as mobility speed, network size and traffic load respectively. The investigation has reported that changes made to these parameters values have a major effect on network performance. Existing research on backoff algorithms for MANETs mainly focuses on using external information, as opposed to information available from within the node, to decide the length of backoff timers. Such information includes network traffic load, transmission failures of other nodes and the total number of nodes in the network. In a mobile network, acquiring such information is not feasible at all times. To address this point, the second part of this thesis proposes new backoff algorithms to use with MANETs. These algorithms use internal information only to make their decisions. This part has revealed that it is possible to achieve higher network throughput and less average packet delay under different values of the parameters mentioned above without the use of any external information. This work proposes two new backoff algorithms. The Optimistic Linear-Exponential Backoff, (OLEB), and the Pessimistic Linear-Exponential Backoff (PLEB). In OLEB, the exponential backoff is combined with linear increment behaviour in order to reduce redundant long backoff times, during which the media is available and the node is still on backoff status, by implementing less dramatic increments in the early backoff stages. PLEB is also a combination of exponential and linear increment behaviours. However, the order in which linear and exponential behaviours are used is the reverse of that in OLEB. The two algorithms have been compared with existing work. Results of this research report that PLEB achieves higher network throughput for large numbers of nodes (e.g. 50 nodes and over). Moreover, PLEB achieves higher network throughput with low mobility speed. As for average packet delay, PLEB significantly improves average packet delay for large network sizes especially when combined with high traffic rate and mobility speed. On the other hand, the measurements of network throughput have revealed that for small networks of 10 nodes, OLEB has higher throughput than existing work at high traffic rates. For a medium network size of 50 nodes, OLEB also achieves higher throughput. Finally, at a large network size of 100 nodes, OLEB reaches higher throughput at low mobility speed. Moreover, OLEB produces lower average packet delay than the existing algorithms at low mobility speed for a network size of 50 nodes. Finally, this work has studied the effect of choosing the behaviour changing point between linear and exponential increments in OLEB and PLEB. Results have shown that increasing the number of times in which the linear increment is used increases network throughput. Moreover, using larger linear increments increase network throughput.

The purpose of this work was to investigate the relationship between a number of influential factors, including cognitive style and approach to learning, and students’ processing behaviour during their use of a particular Web-based resource for Electronics and Electrical Engineering undergraduates. This was achieved through the development of a learner profile for each student using Riding’s (1991) Revised Study Process Questionnaire (R-SPQ-2F). The quantitative component of the research was then set against a detailed analysis of students’ processing behaviour using verbal protocol data gathered through individual think-aloud sessions and post-intervention interviews. The results of the quantitative component of the research provided no compelling evidence to suggest that cognitive style was a factor that influenced student performance while using the resource or their perceptions of the package. There was however, some evidence to suggest that the package was more positively received by students who profiled as deep learners than their surface counterparts. The analysis of students’ processing behaviour from their verbal protocols highlighted a number of the resource’s shortcomings, which typically promoted a surface, goal-oriented approach to its content. It also identified problems with the design and structure of the resource, which at times had a deleterious effect on learning. The results also raised questions regarding the efficacy and use of psychometric inventories in this kind of research.

Sport video is an important video genre. Content-based sports video analysis attracts great interest from both industry and academic fields. A sports video is characterised by repetitive temporal structures, relatively plain contents, and strong spatio-temporal variations, such as quick camera switches and swift local motions. It is necessary to develop specific techniques for content-based sports video analysis to utilise these characteristics. For an efficient and effective sports video analysis system, there are three fundamental questions: (1) what are key stories for sports videos; (2) what incurs viewer’s interest; and (3) how to identify game highlights. This thesis is developed around these questions. We approached these questions from two different perspectives and in turn three research contributions are presented, namely, replay detection, attack temporal structure decomposition, and attention-based highlight identification. Replay segments convey the most important contents in sports videos. It is an efficient approach to collect game highlights by detecting replay segments. However, replay is an artefact of editing, which improves with advances in video editing tools. The composition of replay is complex, which includes logo transitions, slow motions, viewpoint switches and normal speed video clips. Since logo transition clips are pervasive in game collections of FIFA World Cup 2002, FIFA World Cup 2006 and UEFA Championship 2006, we take logo transition detection as an effective replacement of replay detection. A two-pass system was developed, including a five-layer adaboost classifier and a logo template matching throughout an entire video. The five-layer adaboost utilises shot duration, average game pitch ratio, average motion, sequential colour histogram and shot frequency between two neighbouring logo transitions, to filter out logo transition candidates. Subsequently, a logo template is constructed and employed to find all transition logo sequences. The precision and recall of this system in replay detection is 100% in a five-game evaluation collection. An attack structure is a team competition for a score. Hence, this structure is a conceptually fundamental unit of a football video as well as other sports videos. We review the literature of content-based temporal structures, such as play-break structure, and develop a three-step system for automatic attack structure decomposition. Four content-based shot classes, namely, play, focus, replay and break were identified by low level visual features. A four-state hidden Markov model was trained to simulate transition processes among these shot classes. Since attack structures are the longest repetitive temporal unit in a sports video, a suffix tree is proposed to find the longest repetitive substring in the label sequence of shot class transitions. These occurrences of this substring are regarded as a kernel of an attack hidden Markov process. Therefore, the decomposition of attack structure becomes a boundary likelihood comparison between two Markov chains. Highlights are what attract notice. Attention is a psychological measurement of “notice ”. A brief survey of attention psychological background, attention estimation from vision and auditory, and multiple modality attention fusion is presented. We propose two attention models for sports video analysis, namely, the role-based attention model and the multiresolution autoregressive framework. The role-based attention model is based on the perception structure during watching video. This model removes reflection bias among modality salient signals and combines these signals by reflectors. The multiresolution autoregressive framework (MAR) treats salient signals as a group of smooth random processes, which follow a similar trend but are filled with noise. This framework tries to estimate a noise-less signal from these coarse noisy observations by a multiple resolution analysis. Related algorithms are developed, such as event segmentation on a MAR tree and real time event detection. The experiment shows that these attention-based approach can find goal events at a high precision. Moreover, results of MAR-based highlight detection on the final game of FIFA 2002 and 2006 are highly similar to professionally labelled highlights by BBC and FIFA.

The prediction of protein-protein interactions (PPI) has recently emerged as an important problem in the fields of bioinformatics and systems biology, due to the fact that most essential cellular processes are mediated by these kinds of interactions. In this thesis we focussed in the prediction of co-complex interactions, where the objective is to identify and characterize protein pairs which are members of the same protein complex. Although high-throughput methods for the direct identification of PPI have been developed in the last years. It has been demonstrated that the data obtained by these methods is often incomplete and suffers from high false-positive and false-negative rates. In order to deal with this technology-driven problem, several machine learning techniques have been employed in the past to improve the accuracy and trustability of predicted protein interacting pairs, demonstrating that the combined use of direct and indirect biological insights can improve the quality of predictive PPI models. This task has been commonly viewed as a binary classification problem. However, the nature of the data creates two major problems. Firstly, the imbalanced class problem due to the number of positive examples (pairs of proteins which really interact) being much smaller than the number of negative ones. Secondly, the selection of negative examples is based on some unreliable assumptions which could introduce some bias in the classification results. The first part of this dissertation addresses these drawbacks by exploring the use of one-class classification (OCC) methods to deal with the task of prediction of PPI. OCC methods utilize examples of just one class to generate a predictive model which is consequently independent of the kind of negative examples selected; additionally these approaches are known to cope with imbalanced class problems. We designed and carried out a performance evaluation study of several OCC methods for this task. We also undertook a comparative performance evaluation with several conventional learning techniques. Furthermore, we pay attention to a new potential drawback which appears to affect the performance of PPI prediction. This is associated with the composition of the positive gold standard set, which contain a high proportion of examples associated with interactions of ribosomal proteins. We demonstrate that this situation indeed biases the classification task, resulting in an over-optimistic performance result. The prediction of non-ribosomal PPI is a much more difficult task. We investigate some strategies in order to improve the performance of this subtask, integrating new kinds of data as well as combining diverse classification models generated from different sets of data. In this thesis, we undertook a preliminary validation study of the new PPI predicted by using OCC methods. To achieve this, we focus in three main aspects: look for biological evidence in the literature that support the new predictions; the analysis of predicted PPI networks properties; and the identification of highly interconnected groups of proteins which can be associated with new protein complexes. Finally, this thesis explores a slightly different area, related to the prediction of PPI types. This is associated with the classification of PPI structures (complexes) contained in the Protein Data Bank (PDB) data base according to its function and binding affinity. Considering the relatively reduced number of crystalized protein complexes available, it is not possible at the moment to link these results with the ones obtained previously for the prediction of PPI complexes. However, this could be possible in the near future when more PPI structures will be available.