An ontology for risk management of digital collections

Maintaining accessibility to and understanding of digital information over time is a complex challenge that often requires contributions and interventions from a variety of individuals and organizations. The processes of preservation planning and evaluation are fundamentally implicit and share similar complexity. Both demand comprehensive knowledge and understanding of every aspect of to-be-preserved content and the contexts within which preservation is undertaken. Consequently, means are required for the identification, documentation and association of those properties of data, representation and management mechanisms that in combination lend value, facilitate interaction and influence the preservation process. These properties may be almost limitless in terms of diversity, but are integral to the establishment of classes of risk exposure, and the planning and deployment of appropriate preservation strategies. We explore several research objectives within the course of this thesis. Our main objective is the conception of an ontology for risk management of digital collections. Incorporated within this are our aims to survey the contexts within which preservation has been undertaken successfully, the development of an appropriate methodology for risk management, the evaluation of existing preservation evaluation approaches and metrics, the structuring of best practice knowledge and lastly the demonstration of a range of tools that utilise our findings. We describe a mixed methodology that uses interview and survey, extensive content analysis, practical case study and iterative software and ontology development. We build on a robust foundation, the development of the Digital Repository Audit Method Based on Risk Assessment. We summarise the extent of the challenge facing the digital preservation community (and by extension users and creators of digital materials from many disciplines and operational contexts) and present the case for a comprehensive and extensible knowledge base of best practice. These challenges are manifested in the scale of data growth, the increasing complexity and the increasing onus on communities with no formal training to offer assurances of data management and sustainability. These collectively imply a challenge that demands an intuitive and adaptable means of evaluating digital preservation efforts. The need for individuals and organisations to validate the legitimacy of their own efforts is particularly prioritised. We introduce our approach, based on risk management. Risk is an expression of the likelihood of a negative outcome, and an expression of the impact of such an occurrence. We describe how risk management may be considered synonymous with preservation activity, a persistent effort to negate the dangers posed to information availability, usability and sustainability. Risk can be characterised according to associated goals, activities, responsibilities and policies in terms of both their manifestation and mitigation. They have the capacity to be deconstructed into their atomic units and responsibility for their resolution delegated appropriately. We continue to describe how the manifestation of risks typically spans an entire organisational environment, and as the focus of our analysis risk safeguards against omissions that may occur when pursuing functional, departmental or role-based assessment. We discuss the importance of relating risk-factors, through the risks themselves or associated system elements. To do so will yield the preservation best-practice knowledge base that is conspicuously lacking within the international digital preservation community. We present as research outcomes an encapsulation of preservation practice (and explicitly defined best practice) as a series of case studies, in turn distilled into atomic, related information elements. We conduct our analyses in the formal evaluation of memory institutions in the UK, US and continental Europe. Furthermore we showcase a series of applications that use the fruits of this research as their intellectual foundation. Finally we document our results in a range of technical reports and conference and journal articles. We present evidence of preservation approaches and infrastructures from a series of case studies conducted in a range of international preservation environments. We then aggregate this into a linked data structure entitled PORRO, an ontology relating preservation repository, object and risk characteristics, intended to support preservation decision making and evaluation. The methodology leading to this ontology is outlined, and lessons are exposed by revisiting legacy studies and exposing the resource and associated applications to evaluation by the digital preservation community.

The role of context in image annotation and recommendation

With the rise of smart phones, lifelogging devices (e.g. Google Glass) and popularity of image sharing websites (e.g. Flickr), users are capturing and sharing every aspect of their life online producing a wealth of visual content. Of these uploaded images, the majority are poorly annotated or exist in complete semantic isolation making the process of building retrieval systems difficult as one must firstly understand the meaning of an image in order to retrieve it. To alleviate this problem, many image sharing websites offer manual annotation tools which allow the user to “tag” their photos, however, these techniques are laborious and as a result have been poorly adopted; Sigurbjörnsson and van Zwol (2008) showed that 64% of images uploaded to Flickr are annotated with < 4 tags. Due to this, an entire body of research has focused on the automatic annotation of images (Hanbury, 2008; Smeulders et al., 2000; Zhang et al., 2012a) where one attempts to bridge the semantic gap between an image’s appearance and meaning e.g. the objects present. Despite two decades of research the semantic gap still largely exists and as a result automatic annotation models often offer unsatisfactory performance for industrial implementation. Further, these techniques can only annotate what they see, thus ignoring the “bigger picture” surrounding an image (e.g. its location, the event, the people present etc). Much work has therefore focused on building photo tag recommendation (PTR) methods which aid the user in the annotation process by suggesting tags related to those already present. These works have mainly focused on computing relationships between tags based on historical images e.g. that NY and timessquare co-exist in many images and are therefore highly correlated. However, tags are inherently noisy, sparse and ill-defined often resulting in poor PTR accuracy e.g. does NY refer to New York or New Year? This thesis proposes the exploitation of an image’s context which, unlike textual evidences, is always present, in order to alleviate this ambiguity in the tag recommendation process. Specifically we exploit the “what, who, where, when and how” of the image capture process in order to complement textual evidences in various photo tag recommendation and retrieval scenarios. In part II, we combine text, content-based (e.g. # of faces present) and contextual (e.g. day-of-the-week taken) signals for tag recommendation purposes, achieving up to a 75% improvement to precision@5 in comparison to a text-only TF-IDF baseline. We then consider external knowledge sources (i.e. Wikipedia & Twitter) as an alternative to (slower moving) Flickr in order to build recommendation models on, showing that similar accuracy could be achieved on these faster moving, yet entirely textual, datasets. In part II, we also highlight the merits of diversifying tag recommendation lists before discussing at length various problems with existing automatic image annotation and photo tag recommendation evaluation collections. In part III, we propose three new image retrieval scenarios, namely “visual event summarisation”, “image popularity prediction” and “lifelog summarisation”. In the first scenario, we attempt to produce a rank of relevant and diverse images for various news events by (i) removing irrelevant images such memes and visual duplicates (ii) before semantically clustering images based on the tweets in which they were originally posted. Using this approach, we were able to achieve over 50% precision for images in the top 5 ranks. In the second retrieval scenario, we show that by combining contextual and content-based features from images, we are able to predict if it will become “popular” (or not) with 74% accuracy, using an SVM classifier. Finally, in chapter 9 we employ blur detection and perceptual-hash clustering in order to remove noisy images from lifelogs, before combining visual and geo-temporal signals in order to capture a user’s “key moments” within their day. We believe that the results of this thesis show an important step towards building effective image retrieval models when there lacks sufficient textual content (i.e. a cold start).

Identifying health inequalities in individuals with major mental illness (MMI) using routine data

Abstract and Summary of Thesis: Background: Individuals with Major Mental Illness (such as schizophrenia and bipolar disorder) experience increased rates of physical health comorbidity compared to the general population. They also experience inequalities in access to certain aspects of healthcare. This ultimately leads to premature mortality. Studies detailing patterns of physical health comorbidity are limited by their definitions of comorbidity, single disease approach to comorbidity and by the study of heterogeneous groups. To date the investigation of possible sources of healthcare inequalities experienced by individuals with Major Mental Illness (MMI) is relatively limited. Moreover studies detailing the extent of premature mortality experienced by individuals with MMI vary both in terms of the measure of premature mortality reported and age of the cohort investigated, limiting their generalisability to the wider population. Therefore local and national data can be used to describe patterns of physical health comorbidity, investigate possible reasons for health inequalities and describe mortality rates. These findings will extend existing work in this area. Aims and Objectives: To review the relevant literature regarding: patterns of physical health comorbidity, evidence for inequalities in physical healthcare and evidence for premature mortality for individuals with MMI. To examine the rates of physical health comorbidity in a large primary care database and to assess for evidence for inequalities in access to healthcare using both routine primary care prescribing data and incentivised national Quality and Outcome Framework (QOF) data. Finally to examine the rates of premature mortality in a local context with a particular focus on cause of death across the lifespan and effect of International Classification of Disease Version 10 (ICD 10) diagnosis and socioeconomic status on rates and cause of death. Methods: A narrative review of the literature surrounding patterns of physical health comorbidity, the evidence for inequalities in physical healthcare and premature mortality in MMI was undertaken. Rates of physical health comorbidity and multimorbidity in schizophrenia and bipolar disorder were examined using a large primary care dataset (Scottish Programme for Improving Clinical Effectiveness in Primary Care (SPICE)). Possible inequalities in access to healthcare were investigated by comparing patterns of prescribing in individuals with MMI and comorbid physical health conditions with prescribing rates in individuals with physical health conditions without MMI using SPICE data. Potential inequalities in access to health promotion advice (in the form of smoking cessation) and prescribing of Nicotine Replacement Therapy (NRT) were also investigated using SPICE data. Possible inequalities in access to incentivised primary healthcare were investigated using National Quality and Outcome Framework (QOF) data. Finally a pre-existing case register (Glasgow Psychosis Clinical Information System (PsyCIS)) was linked to Scottish Mortality data (available from the Scottish Government Website) to investigate rates and primary cause of death in individuals with MMI. Rate and primary cause of death were compared to the local population and impact of age, socioeconomic status and ICD 10 diagnosis (schizophrenia vs. bipolar disorder) were investigated. Results: Analysis of the SPICE data found that sixteen out of the thirty two common physical comorbidities assessed, occurred significantly more frequently in individuals with schizophrenia. In individuals with bipolar disorder fourteen occurred more frequently. The most prevalent chronic physical health conditions in individuals with schizophrenia and bipolar disorder were: viral hepatitis (Odds Ratios (OR) 3.99 95% Confidence Interval (CI) 2.82-5.64 and OR 5.90 95% CI 3.16-11.03 respectively), constipation (OR 3.24 95% CI 3.01-3.49 and OR 2.84 95% CI 2.47-3.26 respectively) and Parkinson’s disease (OR 3.07 95% CI 2.43-3.89 and OR 2.52 95% CI 1.60-3.97 respectively). Both groups had significantly increased rates of multimorbidity compared to controls: in the schizophrenia group OR for two comorbidities was 1.37 95% CI 1.29-1.45 and in the bipolar disorder group OR was 1.34 95% CI 1.20-1.49. In the studies investigating inequalities in access to healthcare there was evidence of: under-recording of cardiovascular-related conditions for example in individuals with schizophrenia: OR for Atrial Fibrillation (AF) was 0.62 95% CI 0.52 - 0.73, for hypertension 0.71 95% CI 0.67 - 0.76, for Coronary Heart Disease (CHD) 0.76 95% CI 0.69 - 0.83 and for peripheral vascular disease (PVD) 0.83 95% CI 0.72 - 0.97. Similarly in individuals with bipolar disorder OR for AF was 0.56 95% CI 0.41-0.78, for hypertension 0.69 95% CI 0.62 - 0.77 and for CHD 0.77 95% CI 0.66 - 0.91. There was also evidence of less intensive prescribing for individuals with schizophrenia and bipolar disorder who had comorbid hypertension and CHD compared to individuals with hypertension and CHD who did not have schizophrenia or bipolar disorder. Rate of prescribing of statins for individuals with schizophrenia and CHD occurred significantly less frequently than in individuals with CHD without MMI (OR 0.67 95% CI 0.56-0.80). Rates of prescribing of 2 or more anti-hypertensives were lower in individuals with CHD and schizophrenia and CHD and bipolar disorder compared to individuals with CHD without MMI (OR 0.66 95% CI 0.56-0.78 and OR 0.55 95% CI 0.46-0.67, respectively). Smoking was more common in individuals with MMI compared to individuals without MMI (OR 2.53 95% CI 2.44-2.63) and was particularly increased in men (OR 2.83 95% CI 2.68-2.98). Rates of ex-smoking and non-smoking were lower in individuals with MMI (OR 0.79 95% CI 0.75-0.83 and OR 0.50 95% CI 0.48-0.52 respectively). However recorded rates of smoking cessation advice in smokers with MMI were significantly lower than the recorded rates of smoking cessation advice in smokers with diabetes (88.7% vs. 98.0%, p<0.001), smokers with CHD (88.9% vs. 98.7%, p<0.001) and smokers with hypertension (88.3% vs. 98.5%, p<0.001) without MMI. The odds ratio of NRT prescription was also significantly lower in smokers with MMI without diabetes compared to smokers with diabetes without MMI (OR 0.75 95% CI 0.69-0.81). Similar findings were found for smokers with MMI without CHD compared to smokers with CHD without MMI (OR 0.34 95% CI 0.31-0.38) and smokers with MMI without hypertension compared to smokers with hypertension without MMI (OR 0.71 95% CI 0.66-0.76). At a national level, payment and population achievement rates for the recording of body mass index (BMI) in MMI was significantly lower than the payment and population achievement rates for BMI recording in diabetes throughout the whole of the UK combined: payment rate 92.7% (Inter Quartile Range (IQR) 89.3-95.8 vs. 95.5% IQR 93.3-97.2, p<0.001 and population achievement rate 84.0% IQR 76.3-90.0 vs. 92.5% IQR 89.7-94.9, p<0.001 and for each country individually: for example in Scotland payment rate was 94.0% IQR 91.4-97.2 vs. 96.3% IQR 94.3-97.8, p<0.001. Exception rate was significantly higher for the recording of BMI in MMI than the exception rate for BMI recording in diabetes for the UK combined: 7.4% IQR 3.3-15.9 vs. 2.3% IQR 0.9-4.7, p<0.001 and for each country individually. For example in Scotland exception rate in MMI was 11.8% IQR 5.4-19.3 compared to 3.5% IQR 1.9-6.1 in diabetes. Similar findings were found for Blood Pressure (BP) recording: across the whole of the UK payment and population achievement rates for BP recording in MMI were also significantly reduced compared to payment and population achievement rates for the recording of BP in chronic kidney disease (CKD): payment rate: 94.1% IQR 90.9-97.1 vs.97.8% IQR 96.3-98.9 and p<0.001 and population achievement rate 87.0% IQR 81.3-91.7 vs. 97.1% IQR 95.5-98.4, p<0.001. Exception rates again were significantly higher for the recording of BP in MMI compared to CKD (6.4% IQR 3.0-13.1 vs. 0.3% IQR 0.0-1.0, p<0.001). There was also evidence of differences in rates of recording of BMI and BP in MMI across the UK. BMI and BP recording in MMI were significantly lower in Scotland compared to England (BMI:-1.5% 99% CI -2.7 to -0.3%, p<0.001 and BP: -1.8% 99% CI -2.7 to -0.9%, p<0.001). While rates of BMI and BP recording in diabetes and CKD were similar in Scotland compared to England (BMI: -0.5 99% CI -1.0 to 0.05, p=0.004 and BP: 0.02 99% CI -0.2 to 0.3, p=0.797). Data from the PsyCIS cohort showed an increase in Standardised Mortality Ratios (SMR) across the lifespan for individuals with MMI compared to the local Glasgow and wider Scottish populations (Glasgow SMR 1.8 95% CI 1.6-2.0 and Scotland SMR 2.7 95% CI 2.4-3.1). Increasing socioeconomic deprivation was associated with an increased overall rate of death in MMI (350.3 deaths/10,000 population/5 years in the least deprived quintile compared to 794.6 deaths/10,000 population/5 years in the most deprived quintile). No significant difference in rate of death for individuals with schizophrenia compared with bipolar disorder was reported (6.3% vs. 4.9%, p=0.086), but primary cause of death varied: with higher rates of suicide in individuals with bipolar disorder (22.4% vs. 11.7%, p=0.04). Discussion: Local and national datasets can be used for epidemiological study to inform local practice and complement existing national and international studies. While the strengths of this thesis include the large data sets used and therefore their likely representativeness to the wider population, some limitations largely associated with using secondary data sources are acknowledged. While this thesis has confirmed evidence of increased physical health comorbidity and multimorbidity in individuals with MMI, it is likely that these findings represent a significant under reporting and likely under recognition of physical health comorbidity in this population. This is likely due to a combination of patient, health professional and healthcare system factors and requires further investigation. Moreover, evidence of inequality in access to healthcare in terms of: physical health promotion (namely smoking cessation advice), recording of physical health indices (BMI and BP), prescribing of medications for the treatment of physical illness and prescribing of NRT has been found at a national level. While significant premature mortality in individuals with MMI within a Scottish setting has been confirmed, more work is required to further detail and investigate the impact of socioeconomic deprivation on cause and rate of death in this population. It is clear that further education and training is required for all healthcare staff to improve the recognition, diagnosis and treatment of physical health problems in this population with the aim of addressing the significant premature mortality that is seen. Conclusions: Future work lies in the challenge of designing strategies to reduce health inequalities and narrow the gap in premature mortality reported in individuals with MMI. Models of care that allow a much more integrated approach to diagnosing, monitoring and treating both the physical and mental health of individuals with MMI, particularly in areas of social and economic deprivation may be helpful. Strategies to engage this “hard to reach” population also need to be developed. While greater integration of psychiatric services with primary care and with specialist medical services is clearly vital the evidence on how best to achieve this is limited. While the National Health Service (NHS) is currently undergoing major reform, attention needs to be paid to designing better ways to improve the current disconnect between primary and secondary care. This should then help to improve physical, psychological and social outcomes for individuals with MMI.

Query routing in cooperative semi-structured peer-to-peer information retrieval networks

Conventional web search engines are centralised in that a single entity crawls and indexes the documents selected for future retrieval, and the relevance models used to determine which documents are relevant to a given user query. As a result, these search engines suffer from several technical drawbacks such as handling scale, timeliness and reliability, in addition to ethical concerns such as commercial manipulation and information censorship. Alleviating the need to rely entirely on a single entity, Peer-to-Peer (P2P) Information Retrieval (IR) has been proposed as a solution, as it distributes the functional components of a web search engine – from crawling and indexing documents, to query processing – across the network of users (or, peers) who use the search engine. This strategy for constructing an IR system poses several efficiency and effectiveness challenges which have been identified in past work. Accordingly, this thesis makes several contributions towards advancing the state of the art in P2P-IR effectiveness by improving the query processing and relevance scoring aspects of a P2P web search. Federated search systems are a form of distributed information retrieval model that route the user’s information need, formulated as a query, to distributed resources and merge the retrieved result lists into a final list. P2P-IR networks are one form of federated search in routing queries and merging result among participating peers. The query is propagated through disseminated nodes to hit the peers that are most likely to contain relevant documents, then the retrieved result lists are merged at different points along the path from the relevant peers to the query initializer (or namely, customer). However, query routing in P2P-IR networks is considered as one of the major challenges and critical part in P2P-IR networks; as the relevant peers might be lost in low-quality peer selection while executing the query routing, and inevitably lead to less effective retrieval results. This motivates this thesis to study and propose query routing techniques to improve retrieval quality in such networks. Cluster-based semi-structured P2P-IR networks exploit the cluster hypothesis to organise the peers into similar semantic clusters where each such semantic cluster is managed by super-peers. In this thesis, I construct three semi-structured P2P-IR models and examine their retrieval effectiveness. I also leverage the cluster centroids at the super-peer level as content representations gathered from cooperative peers to propose a query routing approach called Inverted PeerCluster Index (IPI) that simulates the conventional inverted index of the centralised corpus to organise the statistics of peers’ terms. The results show a competitive retrieval quality in comparison to baseline approaches. Furthermore, I study the applicability of using the conventional Information Retrieval models as peer selection approaches where each peer can be considered as a big document of documents. The experimental evaluation shows comparative and significant results and explains that document retrieval methods are very effective for peer selection that brings back the analogy between documents and peers. Additionally, Learning to Rank (LtR) algorithms are exploited to build a learned classifier for peer ranking at the super-peer level. The experiments show significant results with state-of-the-art resource selection methods and competitive results to corresponding classification-based approaches. Finally, I propose reputation-based query routing approaches that exploit the idea of providing feedback on a specific item in the social community networks and manage it for future decision-making. The system monitors users’ behaviours when they click or download documents from the final ranked list as implicit feedback and mines the given information to build a reputation-based data structure. The data structure is used to score peers and then rank them for query routing. I conduct a set of experiments to cover various scenarios including noisy feedback information (i.e, providing positive feedback on non-relevant documents) to examine the robustness of reputation-based approaches. The empirical evaluation shows significant results in almost all measurement metrics with approximate improvement more than 56% compared to baseline approaches. Thus, based on the results, if one were to choose one technique, reputation-based approaches are clearly the natural choices which also can be deployed on any P2P network.