Les méthodes ensemblistes constituent un sujet de recherche très populaire au cours de la dernière décennie. Leur succès découle en grande partie de leurs solutions attrayantes pour résoudre différents problèmes d'apprentissage intéressants parmi lesquels l'amélioration de l'exactitude d'une prédiction, la sélection de variables, l'apprentissage de métrique, le passage à l'échelle d'algorithmes inductifs, l'apprentissage de multiples jeux de données physiques distribués, l'apprentissage de flux de données soumis à une dérive conceptuelle, etc... Dans cette thèse nous allons dans un premier temps présenter une comparaison empirique approfondie de 19 algorithmes ensemblistes d'apprentissage supervisé proposé dans la littérature sur différents jeux de données de référence. Non seulement nous allons comparer leurs performances selon des métriques standards de performances (Exactitude, AUC, RMS) mais également nous analyserons leur diagrammes kappa-erreur, la calibration et les propriétés biais-variance. Nous allons aborder ensuite la problématique d'amélioration des ensembles de modèles par la sélection dynamique d'ensembles (dynamic ensemble selection, DES). La sélection dynamique est un sous-domaine de l'apprentissage ensembliste où pour une donnée d'entrée x, le meilleur sous-ensemble en terme de taux de réussite est sélectionné dynamiquement. L'idée derrière les approches DES est que différents modèles ont différentes zones de compétence dans l'espace des instances. La plupart des méthodes proposées estime l'importance individuelle de chaque classifieur faible au sein d'une zone de compétence habituellement déterminée par les plus proches voisins dans un espace euclidien. Nous proposons et étudions dans cette thèse deux nouvelles approches DES. La première nommée ST-DES est conçue pour les ensembles de modèles à base d'arbres de décision. Cette méthode sélectionne via une métrique supervisée interne à l'arbre, idée motivée par le problème de la malédiction de la dimensionnalité : pour les jeux de données avec un grand nombre de variables, les métriques usuelles telle la distance euclidienne sont moins pertinentes. La seconde approche, PCC-DES, formule la problématique DES en une tâche d'apprentissage multi-label avec une fonction coût spécifique. Ici chaque label correspond à un classifieur et une base multi-label d'entraînement est constituée sur l'habilité de chaque classifieur de classer chaque instance du jeu de données d'origine. Cela nous permet d'exploiter des récentes avancées dans le domaine de l'apprentissage multi-label. PCC-DES peut être utilisé pour les approches ensemblistes homogènes et également hétérogènes. Son avantage est de prendre en compte explicitement les corrélations entre les prédictions des classifieurs. Ces algorithmes sont testés sur un éventail de jeux de données de référence et les résultats démontrent leur efficacité faces aux dernières alternatives de l'état de l'art / Ensemble methods has been a very popular research topic during the last decade. Their success arises largely from the fact that they offer an appealing solution to several interesting learning problems, such as improving prediction accuracy, feature selection, metric learning, scaling inductive algorithms to large databases, learning from multiple physically distributed data sets, learning from concept-drifting data streams etc. In this thesis, we first present an extensive empirical comparison between nineteen prototypical supervised ensemble learning algorithms, that have been proposed in the literature, on various benchmark data sets. We not only compare their performance in terms of standard performance metrics (Accuracy, AUC, RMS) but we also analyze their kappa-error diagrams, calibration and bias-variance properties. We then address the problem of improving the performances of ensemble learning approaches with dynamic ensemble selection (DES). Dynamic pruning is the problem of finding given an input x, a subset of models among the ensemble that achieves the best possible prediction accuracy. The idea behind DES approaches is that different models have different areas of expertise in the instance space. Most methods proposed for this purpose estimate the individual relevance of the base classifiers within a local region of competence usually given by the nearest neighbours in the euclidean space. We propose and discuss two novel DES approaches. The first, called ST-DES, is designed for decision tree based ensemble models. This method prunes the trees using an internal supervised tree-based metric; it is motivated by the fact that in high dimensional data sets, usual metrics like euclidean distance suffer from the curse of dimensionality. The second approach, called PCC-DES, formulates the DES problem as a multi-label learning task with a specific loss function. Labels correspond to the base classifiers and multi-label training examples are formed based on the ability of each classifier to correctly classify each original training example. This allows us to take advantage of recent advances in the area of multi-label learning. PCC-DES works on homogeneous and heterogeneous ensembles as well. Its advantage is to explicitly capture the dependencies between the classifiers predictions. These algorithms are tested on a variety of benchmark data sets and the results demonstrate their effectiveness against competitive state-of-the-art alternatives

Titel: Ekologiskt som ekologiskt? – En kvantitativ studie av eko-label effekten ur de tre hållbarhetsaspekterna. Nivå: C-uppsats i ämnet företagsekonomi: Examensarbete i företagsekonomi C 15hp Författare: Malin Damberg Larsson och Anna Östlund Handledare: Dr. Jonas Kågström Datum: 2015 MAJ Syfte: Syftet med studien är att mäta om magnituden på eko-label effekten skiljer sig utifrån vilken del av de tre hållbarhetsaspekterna (miljö, social, ekonomisk) produkten fokuserar på. Genom vår studie vill vi bidra med en inblick i hur eko-label effekten påverkar synen på ekologiska produkter utifrån de olika aspekterna. Metod: I studien genomfördes ett experiment med 117 deltagare som var uppdelade i tre olika grupper. Deltagarna fick smaka på två vanliga vindruvor men där en benämndes som ekologisk. De tre grupperna skilde sig åt genom att den ”ekologiska” vindruvan påstods rikta sig mot antingen social-, ekonomisk- eller miljö/ekologisk hållbarhet. Detta för att se om eco-label effekten var starkare i någon av de tre betingelserna. Resultat & slutsats: Resultatet visar att konsumenter tenderar att glorifiera ekologiska produkter. Det går också att se att redan ekopositiva konsumenter som möts av en rikta marknadsföring mot eko- och miljö tenderar att glorifiera ekologiska produkter mer jämfört med de andra betingelserna. Förslag till fortsatt forskning: Författarna efterlyser vidare forskning som vill inrikta sig mot en äldre målgrupp och dess syn och uppfattning på ekologiska produkter. Uppsatsens bidrag: Uppsatsens bidrag består av att både bekräfta delar av tidigare studier som visat på en glorifiering av ekologiska produkter samt belysa vikten av tanken om de hälsofördelar de ekologiska produkterna anses ha bland konsumenterna. Rent praktiskt bidrar uppsatsen till kunskap kring vikten av att i sin marknadsföring belysa miljöaspekter.

Objetivos: Determinar a frequência de prescrições de medicamentos off label e não licenciados para pediatria na Atenção Primária à Saúde do município de Viamão, no Rio Grande do Sul. Método: Estudo descritivo, analisando prescrições de 323 pacientes emitidas de Agosto a Dezembro de 2012 em duas unidades de atenção primária do Sistema Público de Saúde do município de Viamão, no Rio Grande do Sul. Resultados: Durante o período estudado obteve-se a prescrição de 731 medicamentos e houve frequência de 31,7% de medicamentos prescritos off label, especialmente anti-histamínicos e antiasmáticos (32,3% e 31,5%, respectivamente). O principal tipo de prescrição off-label foi dose (38,8%), seguida de idade (31,5%) e de frequência de administração (29,3%). Com relação à prescrição off label de dose, foi mais frequente a sobredose 93,3% do que a subdose 6,7%. Não foram encontradas prescrições de medicamentos não licenciados. Discussão e Conclusão: As dificuldades relacionadas às pesquisas com crianças propiciam a prescrição de medicamentos off label, prática que, apesar de não ser ilegal, gera insegurança em relação aos possíveis efeitos adversos em uma população com características peculiares como a pediátrica. O estudo mostrou que esta prática é comum na Atenção Primária a Saúde em uma cidade do Rio Grande do Sul, a exemplo de pesquisas em cidades europeias. Este é o primeiro estudo que analisa a frequência de prescrição de medicamentos off label no âmbito ambulatorial do Sistema Único de Saúde no país e espera-se que possa contribuir para a busca de alternativas que promovam o uso racional de medicamentos na pediatria. / Objectives: To determine the frequency of prescriptions for off label and unlicensed medicines for pediatrics at Viamão Primary Health Care in Rio Grande do Sul. Method: Descriptive study analyzing prescriptions of 323 patients, issued from August to December 2012, in two primary care units of Viamão Public Health System in Rio Grande do Sul. Results: During the study period 731 prescription drugs were observed. Off label frequency was 31.7%, especially antihistamines and anti-asthmatic prescriptions (32.3% and 31.5%, respectively). The main type of off-label prescribing was dose (38.8%), followed by age (31.5%) and frequency of administration (29.3%). Regarding the prescription off label of dose, was more frequent overdose (93.3%) than underdose (6.7%). No prescriptions for unlicensed medicines were found. Discussion and Conclusion: The difficulties related to research with children favor prescription drugs off label, a practice that is not illegal, but may produce adverse effects in a population with unique characteristics such as pediatric. The study showed that this practice is common in Primary Health Care in a city in Rio Grande do Sul, as it is in Europe, as reported by other researches. This is the first study that analyzes the frequency of off-label prescribing of drugs in outpatient under the National Health System in Brazil and it is expected to be able to contribute to the search for alternatives that promote rational drug use in pediatrics.

Context: Multi-label classification concerns classification with multi-dimensional output. The Classifier Chain breaks the multi-label problem into multiple binary classification problems, chaining the classifiers to exploit dependencies between labels. Consequently, its performance is influenced by the chain's order. Approaches to finding advantageous chain orders have been proposed, though they are typically costly. Objectives: This study explored the use of filter model feature selection techniques to order Classifier Chains. It examined how feature selection techniques can be adapted to evaluate label dependence, how such information can be used to select a chain order and how this affects the classifier's performance and execution time. Methods: An experiment was performed to evaluate the proposed approach. The two proposed algorithms, Forward-Oriented Chain Selection (FOCS) and Backward-Oriented Chain Selection (BOCS), were tested with three different feature evaluators. 10-fold cross-validation was performed on ten benchmark datasets. Performance was measured in accuracy, 0/1 subset accuracy and Hamming loss. Execution time was measured during chain selection, classifier training and testing. Results: Both proposed algorithms led to improved accuracy and 0/1 subset accuracy (Friedman & Hochberg, p < 0.05). FOCS also improved the Hamming loss while BOCS did not. Measured effect sizes ranged from 0.20 to 1.85 percentage points. Execution time was increased by less than 3 % in most cases. Conclusions: The results showed that the proposed approach can improve the Classifier Chain's performance at a low cost. The improvements appear similar to comparable techniques in magnitude but at a lower cost. It shows that feature selection techniques can be applied to chain ordering, demonstrates the viability of the approach and establishes FOCS and BOCS as alternatives worthy of further consideration.

Neste trabalho são estudadas as mioglobínas de Aplysía Brasiliana (MbApB) e da tartaruga marinha \"Dermoche lis Coriacea\" (MbT) focalizando a transição ácida alcalina (TAA), a interação com metais de transição e mudanças conformacionais induzidas - termicamente com objetivo de observar diferenças estruturais destas mioglobinas. A TAA da MbApB possui um pK = 7,2 quando observada Fe3+ em espectro do Fe3+ em g1 = 5,83 e pK = 7,5 quando medida através da absorção óptica (&#955=590nm). O espectro de EPR do Fe+ a pH alcalino revela uma distorção rômbica no campo cristalino do ferro que esta de acordo com a ausência da histidina distantes ta proteína. Além disso, as linhas associadas à configuração de baixo spin são bastante alargadas. Este alargamento de linha PQ de ser explicado por uma flutuação no posicionamento da heme em relação aos eixos de simetria. A MbApB forma complexos com o Cu2+ e Mn2+ . Existe som te um sitio de ligação para esses metais na proteína. Este sítio provavelmente envolve ligantes comuns ao Mn2+ e Cu2+ pois a ligação é competitiva, indicando que o complexo com o Cu2+ é mais estável do que com Mn2+ (KAMn2+ = (11,5-0,8).103 M-1 ). O complexo MbApB: Cu2+ exibe uma transição dependente do pH indicando uma mudança na coordenação do Cu2+ . Esta transição pode ser ajustada teoricamente admitindo-se uma interconversão entre duas formas, fornecendo um pK = 11,0. Observa-se também que a complexação da MbApB com mais de um Cu2+ provoca desnaturação. A TAA da MbT foi estudada por absorção óptica (pK=8,4) e EPR (pK = 8,3). Os espectros de EPR mostram que o Fe3+ esta num campo cristalino com simetria axial e distorção rômbica. As medidas de absorção óptica revelam pontos isosbéstioos praticamente idênticos aos de outras mioglobinas e um espectro na região de Soret típico de hemoproteínas que possuem a histidina distal. Estes dados permitem antecipar que a MbT deve possuir os mesmos resíduos de aminoácidos na região da heme que a Mb(II), porem com a histidina E-7 ligeiramente inclinada em relação ao eixo de simetria da heme. As mudanças conformacionais induzidas pela temperatura foram estudadas para a MbT em várias pH\'s. Esta proteína possui uma cisteína em sua cadeia polipeptídica que permite a ligação de um marcador de spin especifico para os grupos SR. Os resultados obtidos através do espectro de EPR do marcador de spin e da absorção óptica do Fe3+ permitem concluir que esta proteína exibe o fenômeno de pré-desnaturação e a transição da forma nativa para a desnaturada é o resultado de uma seqüência de etapas indicando que existem estados intermediários. / The myoblobins of \"Applysia Brasiliana\" (MbApB) and of the sea turtle \"Dermochelis Coriacea\" (MbT) are studied in this work with special attention devoted to the acid-aIkaIyne transition (AAT) , the interaction with transition metals and temperature induced conformational changes in order to characterize structural differences in these proteins. The AAT of MbApB has a pK = 7.2 obtained from the EPR spectra of Fe3+ at g1 = 5,83 and a pK = 7,5 obtained from optical absorption (&#955=590nm). The EPR Spectrum of Fe3+ at alkaline pH shows a rhombic distortion the ion crystal field which is in agreement with the absence in this protein of the distal histidine residue. The ESR lines associated with the low spin configuration are considerably broadened. This effect can be explained by fluctuations on the heme position relative to the symmetry axis. MbApB forms complexes with both Cu2+ and Mn2+ only one binding site is obtained for both metals in the protein. This site probably has common ligands for Mn2+ and Cu2+ as the binding is competitive, suggesting also that the Cu2+ complex is more stable than the Mn2+ one (KAMn2+ = (11,5-0,8).103 M-1 ). The Cu2+: MbApB complex shows a pH dependent transition probably related to a change in the Cu2+ coordination. This transition can be adjusted assuming an interconversion between two different forms with a pK = 11,0. Addition of more than one Cu2+ ion per MbApB produces the denaturation of the protein. The AAT of MbT was studied through optical absorption (pK =: 8,4) and EPR (pK =: 8.3). The EPR spectra show that the Fe3+ ion is in a crystal field with axial symmetry with rhombic distortion. Measurements of optical absorption show isosbestic points practically identical to other hemoproteins with the distal histidina present. Our data permit to anticipate that MbT probably has the same aminoacid residues in the heme region as Mb (II), and thÉit a slight displacement of the histidine E-7 about the heme symmetry axis is observed. Temperature induced conformational changes were studied in MbT at various pH\'s. This protein has a cysteine residue in the polypeptide chain which allows the binding of a specific spin label, a maleimide derivative. Our results from the EBR spectra of the spin label attached to MbT together with the optical absorption by Fe3+ lead to the conclusion that this protein exhibits a pre-denaturation behaviour and the transition from the native to the denatured forms in a result of a sequence of steps suggesting the existence of intermediate states.

Após setenta anos de sua descoberta, o vírus da zika surgiu no Brasil, espalhou-se rapidamente pelas Américas e trouxe complicações incomuns em doenças causadas por Flavivirus, como a microcefalia. A Organização Mundial da Saúde classifica a zika como a doença viral mais preocupante da atualidade e considera urgente desenvolver novos métodos de diagnóstico para ela e doenças correlatas como a dengue. Embora existam exames para identificar infecções pelos vírus dessas duas doenças, ainda não há um método rápido, específico e de baixo custo para o diagnóstico precoce. Visando preencher essa lacuna, este trabalho teve como objetivo construir dois tipos de biossensores eletroquímicos de DNA para detecção label-free desses dois vírus. Foram fabricados eletrodos descartáveis em substrato de politereftalato de etileno metalizado com filme fino de ouro nas configurações com um e três contatos. As sequências genéticas de iniciadores e sondas de captura foram desenhadas especialmente para este trabalho com base na análise dos genomas dos vírus. O primeiro biossensor utilizou o eletrodo em uma célula eletroquímica e foi capaz de identificar sequências de DNA da zika ou da dengue. As análises por espectroscopia de impedância eletroquímica mostraram que o biossensor é seletivo à sequência alvo com limite de detecção de (9,86 ± 0,89) nmol L-1. O segundo biossensor utilizou um eletrodo de três contatos para identificação de sequências de DNA em uma gota da amostra. No contato central, usado como eletrodo de trabalho, foi imobilizada a sequência de captura e os contatos laterais funcionaram como eletrodos de referência e auxiliar. Nesse sistema as medidas de impedância indicaram limite de detecção de (25,0 ± 1,7) nmol L-1. Os biossensores desenvolvidos mostraram seletividade para identificar o material genético dos vírus da zika e da dengue nos ensaios com DNA sintético e, portanto, são promissores para a análise de amostras reais, principalmente de produtos da polimerase da cadeia reversa. / After seventy years of its discovery, zika virus has emerged in Brazil, spread rapidly throughout the Americas, bringing unusual complications in diseases caused by flaviviruses, such as microcephaly. The World Health Organization classifies zika as the most harmful viral disease today and considers urgent the development of new diagnostic methods for zika and related diseases, such as dengue. Although there are tests to identify both infections, no current diagnostic method is rapid, specific and cost-efective. This thesis describes two types of electrochemical DNA biosensors for label-free detection of these zika and dengue. Disposable electrodes were fabricated on polyethylene terephthalate substrates covered with a nanometric gold layer by thermal evaporation, manufactured in one- and three-contact configurations. Genetic sequences of primers and complementary capture probes were designed based on the analysis of the virus genomes. The first biosensor we developed used the new electrode in an electrochemical cell and was able to identify zika or dengue DNA sequences. Analyses by electrochemical impedance spectroscopy showed that these biosensors are selective for zika or dengue with a detection limit of (9.86 ± 0.89) nmol L-1. A second type of biosensor used a three-contact electrode to identify DNA sequences in a drop of sample. In the central contact, used as a working electrode, the capture sequence was immobilized and the lateral contacts acted as reference and auxiliary electrodes. In this system the impedance measurements indicated a limit of detection of (25.0 ± 1.7) nmol L-1. The developed biosensors showed selectivity for zika and dengue in the synthetic DNA assays, and therefore are promising for the analysis of real samples, especially the polymerase chain reaction amplicon.

L’utilisation des antipsychotiques (AP) pour une indication non approuvée autrement dite off-label, peut être associée avec des effets indésirables graves chez les ainés. Étant donné le peu d’études portant sur le sujet, cette étude vise à déterminer la prévalence et les facteurs associés à l’utilisation off-label des AP chez les ainés âgés ≥ 65 ans vivant dans la communauté au Québec. Des analyses secondaires des données provenant de 2 enquêtes «l’Étude sur la Santé des Aînés» (ESA) et ESA-Services, étaient effectuées sur un échantillon total de 4018 ainés. L’utilisation off-label d’AP était définie par l’absence d’une indication approuvée selon la liste d’approbation publiée par Santé Canada en juin 2016, durant la même année. Les diagnostics ont été repérés dans les registres administratifs des services médicaux et pharmaceutiques de la « Régie de l’Assurance médicale du Québec» (RAMQ) en utilisant les codes internationaux de maladies (CIM-9) validés par d’autres études, et les codes CIM-10 appropriés selon la formule de changement CIM-9/10 fournie par la RAMQ. Le cadre conceptuel d’Andersen & Newman conçu pour expliquer le comportement de l'utilisation des services de santé selon trois catégories de facteurs (prédisposant, facilitants et de besoin) a été utilisé pour étudier, avec des régressions multinomiales, l’utilisation off-label des AP comparée à la non-utilisation et à l’utilisation approuvée dite labeled. La prévalence d’utilisation d’AP dans cet échantillon d’ainés était estimée à 2.5%. Parmi les utilisateurs d’AP, 78% étaient considérés comme off-label. L’utilisation off-label des antipsychotiques comparée à la non-utilisation était associée à l’âge (RC : 0.46; 95%CI: 0.27-0.78) , le niveau de scolarité (RC: 2.68; 95%IC: 1.64- 4.40), le nombre de visites ambulatoires incluant les visites chez le médecin traitant (≥ 6) (RC: 2.39; 95%IC: 1.34- 4.25), l’utilisation d’antidépresseurs ou de benzodiazépines (RC: 5.81; 95%IC: 3.31- 10.21), et la présence d’un syndrome cérébral organique et de la maladie d’Alzheimer (RC: 5.73; 95%IC: 1.74- 18.89). L’utilisation off-label des AP comparée à l’utilisation labeled était associée avec l’insomnie (RC: 0.13; 95%IC: 0.02- 0.91) et la dépression majeure (RC: 0.02; 95%IC: <0.01-0.11). Cette étude a montré une prévalence élevée de l’utilisation off-label des AP chez les ainés vivant dans la communauté au Québec. En résumé, l’utilisation off-label des AP était associée à la présence de visites ambulatoires répétées, l’utilisation d’autres médicaments psychotropes et la présence d’un syndrome cérébral organique, reflétant potentiellement un profil des cas cliniques plus complexes. / Abstract: The risk of using antipsychotics (AP) for off-label indications may outweigh the benefits in older adults. Due to the scarcity of studies describing off-label use of AP in older adults, this study aimed to determine the prevalence of the off-label use of AP in Quebec, and to determine, using Andersen & Newman’s model of healthcare seeking behaviour, the predisposing, enabling and need factors associated with the off-label use of AP. We used data from a socio-demographically (age, sex, and postal code representing population density) weighted sample (n = 4018) of older adults living in the community and who participated in the «Enquête sur la Santé des Aînés» (ESA) and ESA-Services health studies. AP use was identified from the RAMQ pharmaceutical registry. The presence of off-label use of AP (yes/no) was identified from validated ICD-9/10 diagnoses codes found in the RAMQ and MED-ÉCHO medical services and hospitalisations databases. The off-label use of AP was defined by the absence, during the same one-year period, of an approved indication for a delivered AP, according to Health Canada’s approval database as of June 2016. Multinomial logistic regression was used to study the off-label use of AP as compared to both labeled use and non-use as a function of predisposing, enabling and need factors. In this study, the prevalence of AP use reached 2.5%, of which 78% was off-label. Compared to non-use, off-label AP use was associated with age (OR: 0.46; 95%CI: 0.27-0.78); education level (OR: 2.68; 95% CI: 1.64- 4.40), a higher number of outpatient visits including physicians visits (OR: 2.39; 95%CI: 1.34- 4.25), antidepressant or benzodiazepine use (OR: 5.81; 95%CI: 3.31- 10.21), and the presence of an organic brain syndrome & Alzheimer’s disease (OR: 5.73; 95%CI: 1.74- 18.89). Compared to labeled use, off-label use was less likely in those with major depression (OR: 0.02; 95%CI: <0.01-0.11) and insomnia (OR: 0.13; 95%CI: 0.02- 0.91). In summary, off-label use is highly prevalent among community living older adults using AP. Off-label use was associated with the presence of an organic brain syndrome & Alzheimer’s disease, other psychotropic drug use and with increased outpatient visits, suggesting the off-label use of AP in more complex and severe clinical cases. Future longitudinal studies should focus on health related changes among incident off-label users of AP.

L’objectif de l’étude est d’évaluer la durabilité de la filière laitière de petits ruminants au Liban et d’identifier ses faiblesses afin de proposer des solutions garantissant sa pérennité. Les résultats d’une enquête englobant 129 éleveurs (Srour, 2006) ainsi que de trois enquêtes portant sur 15 transformateurs, 83 distributeurs et 250 consommateurs ont été agrégés en 12 paramètres représentant les trois volets de la durabilité (environnementale, sociale et économique). Les résultats ont été utilisés pour concevoir un cahier des charges englobant des solutions aux problèmes de la durabilité ; celui-ci a été validé auprès de 16 producteurs, 15 transformateurs, 17 distributeurs et 20 consommateurs. Les producteurs sont responsables de l’impact environnemental le plus important de la filière et souffrent d’une fragilité économique élevée. Les transformateurs sont caractérisés par un manque d’investissement et de politique de traitement des déchets, mais présentent les meilleures conditions sociales à leurs employés et sont les plus performants au niveau de la valeur ajoutée malgré un taux faible de croissance. Chez les distributeurs, les épiceries sont en régression face à des grandes surfaces en pleine expansion et offrant de meilleures conditions sociales à leurs employés. Les consommateurs ignorent le concept de la durabilité et basent leur confiance sur les vendeurs pour l’achat des produits traditionnels. La mise en place de cahier des charges est soutenue par son originalité, l’absence de produits similaires et la situation géographique, mais elle doit prendre en considération la réceptivité du marché (local et extérieur) et l’absence d’infrastructure spécialisée / The aim of this study is to evaluate the sustainability of dairy small ruminant supply chain in Lebanon and to identify its weaknesses in order to suggest solutions guaranteeing its durability. The results relevant to a survey covering 129 breeders (Srour, 2006) and to three other surveys covering 15 processors, 83 distributors and 250 consumers were combined in 12 parameters representing the three facets of sustainability (environmental, social and economical). The results were used to design a labelling scheme containing solutions to sustainability problems; the latter was then validated with 16 breeders, 15 processors, 17 distributors, and 20 consumers. The breeders are responsible for the most important environmental impact of the supply chain and suffer from a high economical fragility. The processors are characterized with a shortage of investment and waste treatment policy but present the best social conditions to their employees and hold the best added value performance despite a weak growth rate. At the distributors’ level, groceries are in regression, facing large retailers in full expansion and offering better social conditions to their employees. The consumers ignore the sustainability concept and base their trust on the salesmen in buying traditional products. The application of the labelling scheme is supported by its originality, the absence of similar products and the geographical situation, but it has to take into consideration the receptiveness of the market (local and foreign) and the absence of specialized infrastructure

La non-discrimination au travail constitue un sujet de société et une préoccupation des entreprises en France depuis l'importation de l'approche anglo-saxonne du management de la diversité au début des années 2000. Face à la nécessité d'appréhender les spécificités sociétales et managériales de ce phénomène en contexte français, cette thèse explore le processus de traduction de la question de la non-discrimination au travail en pratiques managériales au sein des entreprises. Pour comprendre et illustrer ce processus, la thèse propose une analyse de données mixtes et l'étude d'un cas original : le Label Diversité. Le recueil des données s'illustre notamment par une immersion participante au sein d'un réseau d'entreprises (FACE) dans le cadre d'une collaboration CIFRE ainsi que par une enquête auprès des concepteurs du Label Diversité et de plusieurs entreprises labellisées. Une grille d'analyse intégrative combinant les théories néo-institutionnalistes et la littérature sur l'instrumentation de gestion est également mobilisée. Ces choix méthodologiques et théoriques permettent de répondre à l'objectif poursuivi par ce travail doctoral à quatre niveaux. En appréhendant la non-discrimination au travail comme un champ organisationnel, la thèse permet de mieux comprendre les spécificités du contexte dans lequel s'insère cette question en France. A l'aune des théories du changement institutionnel, elle offre également une meilleure compréhension du processus par lequel l'approche de la diversité a favorisé la prise en compte des discriminations au travail par les entreprises. En troisième lieu, la thèse permet de comprendre comment ces entreprises s'approprient concrètement le management de la diversité. Une première évaluation des effets des pratiques auxquelles le management de la diversité donne lieu au sein des entreprises est enfin développée. Elle propose notamment une analyse du Label Diversité sous le prisme des perspectives appropriatives des instruments de gestion. En conclusion, ce travail doctoral questionne l'institutionnalisation de la non-discrimination au travail en France. Il montre en particulier que cette question ne semble pas encore appréhendée comme un « fait naturel » par tous les acteurs et interroge sa capacité à répondre au problème des discriminations au travail en contexte de crise économique et sociale. / With the breakthrough of the Anglo-saxon approach to manage a diverse workforce in this globalize world – best known as “Diversity management” in the USA – non-discrimination within the workplace became a major social issue and a growing concern for French companies in the early 2000s. So as to face the need to better understand the societal and managerial peculiarities of this phenomenon within the French context, this thesis explores the process of translating the issue of non-discrimination at work in managerial practices within companies. To understand and illustrate this process, this thesis provides an analysis of mixed data and an original case study: the French Diversity Label. To collect data and get access to the field, it was chosen to be immersed in a network of companies (FACE) with a 3-year research contract (CIFRE) and to complete immersion with a survey that implied two categories of participants: the designers of the Diversity Label and labelled companies. An integrative analysis grid combining neo-institutionalist theories and scientific literature on management instrumentation is also used. These methodological and theoretical choices allow us to meet the objective of this doctoral work at four levels. Considering non-discrimination at work as an organizational field, this thesis promotes a better understanding of the contextual characteristics regarding the issue of non-discrimination at work in France. In the light of theories of institutional change, it also offers a better understanding of the process by which the diversity approach enabled companies to consider and focus on the discrimination issue in their organisational practices. This thesis provides also an understanding of how these companies integrate and take ownership of the diversity management approach in practise. And finally, a first assessment of the effects of diversity management practices within companies is developed throughout the analysis of the Diversity Label in regards with appropriative perspectives of management instruments approach. In conclusion, this doctoral study questions the institutionalization of non-discrimination at work in France. It shows that this question seems not yet seen as a "natural fact" by all actors and questioned its ability to address the problem of discrimination in the workplace, in a context where a socioeconomic crisis is still going on.

Link label prediction is the problem of predicting the missing labels or signs of all the unlabeled edges in a network. For signed networks, these labels can either be positive or negative. In recent years, different algorithms have been proposed such as using regression, trust propagation and matrix factorization. These approaches have tried to solve the problem of link label prediction by using ideas from social theories, where most of them predict a single missing label given that labels of other edges are known. However, in most real-world social graphs, the number of labeled edges is usually less than that of unlabeled edges. Therefore, predicting a single edge label at a time would require multiple runs and is more computationally demanding. In this thesis, we look at link label prediction problem on a signed citation network with missing edge labels. Our citation network consists of papers from three major machine learning and data mining conferences together with their references, and edges showing the relationship between them. An edge in our network is labeled either positive (dataset relevant) if the reference is based on the dataset used in the paper or negative otherwise. We present three approaches to predict the missing labels. The first approach converts the label prediction problem into a standard classification problem. We then, generate a set of features for each edge and then adopt Support Vector Machines in solving the classification problem. For the second approach, we formalize the graph such that the edges are represented as nodes with links showing similarities between them. We then adopt a label propagation method to propagate the labels on known nodes to those with unknown labels. In the third approach, we adopt a PageRank approach where we rank the nodes according to the number of incoming positive and negative edges, after which we set a threshold. Based on the ranks, we can infer an edge would be positive if it goes a node above the threshold. Experimental results on our citation network corroborate the efficacy of these approaches. With each edge having a label, we also performed additional network analysis where we extracted a subnetwork of the dataset relevant edges and nodes in our citation network, and then detected different communities from this extracted sub-network. To understand the different detected communities, we performed a case study on several dataset communities. The study shows a relationship between the major topic areas in a dataset community and the data sources in the community.