The automatic analysis of emotions is a well-established area in the natural language processing ( NLP ) research field. It has shown valuable and relevant applications in a wide array of domains such as health and well-being, empathetic conversational agents, author profiling, consumer analysis, and security. Most emotion analysis research till now has focused on sources such as news documents and product reviews. In these cases, the NLP task is the classification into predefined closed-set emotion categories (e.g. happy, sad), or alternatively labels (positive, negative). A deep and fine-grained emotion analysis would require explanations of the trigger events that may have led to a user state. This type of analysis is still in its infancy. In this work, we introduce the concept of Emotion Carriers (EC) as the speech or text segments that may include persons, objects, events, or actions that manifest and explain the emotions felt by the narrator during the recollection. In order to investigate this emotion concept, we analyze Personal Narratives (PN) - recollection of events, facts, or thoughts from one’s own experience, - which are rich in emotional information and are less explored in emotion analysis research. PNs are widely used in psychotherapy and thus also in mental well-being applications. The use of PNs in psychotherapy is rooted in the association between mood and recollection of episodic memories. We find that ECs capture implicit emotion information through entities and events whereas the valence prediction relies on explicit emotion words such as happy, cried, and angry. The cues for identifying the ECs and their valence are different and complementary. We propose fine-grained emotion analysis using valence and ECs. We collect and annotate spoken and written PNs, propose text-based and speech-based annotation schemes for valence and EC from PNs, conduct annotation experiments, and train systems for the automatic identification of ECs and their valence.

Yes / While research on factors driving corporate investment decisions has blossomed, knowledge related to the Chief Executive Officer’s (CEO’s) market sentiment on investment decision outcomes is lacking. In this study, we extend the existing corporate finance literature by examining the underexplored issue of how CEOs’ market sentiment drives firms’ investment decisions. Capitalising on a large sample of US firms for the period 2004-2014, we uncovered some crucial observations. First, we found empirical support for our theoretical contention that market sentiment drives corporate investment decisions. Second, we established that, while financial flexibility induces managers to overinvest, the expectation of future profitability leads firms to underinvest during high sentiment periods. In addition, we uncovered that the 2007/08 financial crisis significantly impacted firm behaviour and realigned managerial decision-making. Thus, the sentiment-investment relationship is more pronounced during the crisis and the post-crisis periods. Our results are robust after accounting for the possibility of endogeneity and using alternative measures of both CEOs’ market sentiment and firm investment.

Depuis une trentaine d' années, les nouveaux enseignants sont confrontés à des conditions d'insertion difficiles, notamment en raison de la précarité professionnelle. Ce contexte peut affecter de différentes manières le développement et la mobilisation des compétences, processus qui sont en relation avec le développement du sentiment d'efficacité personnelle (SEP) face à l'enseignement. Le SEP étant considéré comme une ressource essentielle pour agir avec compétence, s'engager dans les tâches et persévérer dans la profession enseignante, nous avons consacré notre recherche doctorale à ce sujet. Pour la méthodologie, la recherche a fait appel à des entrevues semi-structurées auprès de 15 enseignantes et enseignants, à une liste de contrôle permettant aux participants de cibler les sources de leur SEP, et enfin à la technique des incidents critiques pour faire évoquer des expériences mettant en relief le déploiement du SEP en contexte de pratique. Les résultats de la recherche montrent que le SEP se développe premièrement, au cours de la formation initiale grâce à la formation théorique et surtout aux stages. Deuxièmement, durant la phase d'insertion professionnelle par la mobilisation réussie des compétences acquises en formation initiale, par l'expérience d'enseignement, les conditions d'insertion facilitantes, la formation continue et l'apport des différents acteurs du milieu scolaire et universitaire, notamment à travers le soutien social, la rétroaction, la collaboration, la reconnaissance et la confiance témoignées. Les résultats indiquent également que ceux et celles qui se sentent efficaces surmontent mieux les difficultés de l'insertion, prennent davantage d'initiatives, innovent dans leur pratiques et obtiennent de meilleures performances dans les tâches attribuées. Notre étude contribue à apporter une nouvelle perspective dans la compréhension du SEP basée sur de véritables expériences d'enseignants débutants du secondaire, fait découvrir les possibilités sur lesquelles les programmes de formation et les mesures d' insertion pourraient s'appuyer et enfin, permet d'enrichir la problématique de l'insertion en enseignement.

Sentiment analysis aims to use automated tools to detect subjective information such as opinions, attitudes, and feelings expressed in text, and has received a rapid growth of interest in natural language processing in recent years. Probabilistic topic models, on the other hand, are capable of discovering hidden thematic structure in large archives of documents, and have been an active research area in the field of information retrieval. The work in this thesis focuses on developing topic models for automatic sentiment analysis of web data, by combining the ideas from both research domains. One noticeable issue of most previous work in sentiment analysis is that the trained classifier is domain dependent, and the labelled corpora required for training could be difficult to acquire in real world applications. Another issue is that the dependencies between sentiment/subjectivity and topics are not taken into consideration. The main contribution of this thesis is therefore the introduction of three probabilistic topic models, which address the above concerns by modelling sentiment/subjectivity and topic simultaneously. The first model is called the joint sentiment-topic (JST) model based on latent Dirichlet allocation (LDA), which detects sentiment and topic simultaneously from text. Unlike supervised approaches to sentiment classification which often fail to produce satisfactory performance when applied to new domains, the weakly-supervised nature of JST makes it highly portable to other domains, where the only supervision information required is a domain-independent sentiment lexicon. Apart from document-level sentiment classification results, JST can also extract sentiment-bearing topics automatically, which is a distinct feature compared to the existing sentiment analysis approaches. The second model is a dynamic version of JST called the dynamic joint sentiment-topic (dJST) model. dJST respects the ordering of documents, and allows the analysis of topic and sentiment evolution of document archives that are collected over a long time span. By accounting for the historical dependencies of documents from the past epochs in the generative process, dJST gives a richer posterior topical structure than JST, and can better respond to the permutations of topic prominence. We also derive online inference procedures based on a stochastic EM algorithm for efficiently updating the model parameters. The third model is called the subjectivity detection LDA (subjLDA) model for sentence-level subjectivity detection. Two sets of latent variables were introduced in subjLDA. One is the subjectivity label for each sentence; another is the sentiment label for each word token. By viewing the subjectivity detection problem as weakly-supervised generative model learning, subjLDA significantly outperforms the baseline and is comparable to the supervised approach which relies on much larger amounts of data for training. These models have been evaluated on real world datasets, demonstrating that joint sentiment topic modelling is indeed an important and useful research area with much to offer in the way of good results.

Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal.

Les médias sociaux ont changé notre manière de communiquer entre individus, au sein des organisations et des communautés. La disponibilité de ces données sociales ouvre de nouvelles opportunités pour comprendre et influencer le comportement des utilisateurs. De ce fait, la fouille des médias sociaux connait un intérêt croissant dans divers milieux scientifiques et économiques. Dans cette thèse, nous nous intéressons spécifiquement aux utilisateurs de ces réseaux et cherchons à les caractériser selon deux axes : (i) leur expertise et leur réputation et (ii) les sentiments qu’ils expriment.De manière classique, les données sociales sont souvent fouillées selon leur structure en réseau. Cependant, le contenu textuel des messages échangés peut faire émerger des connaissances complémentaires qui ne peuvent être connues via la seule analyse de la structure. Jusqu’à récemment, la majorité des travaux concernant l’analyse du contenu textuel était proposée pour l’Anglais. L’originalité de cette thèse est de développer des méthodes et des ressources basées sur le contenu pour la fouille des réseaux sociaux pour la langue Française.Dans le premier axe, nous proposons d'abord d’identifier l'expertise des utilisateurs. Pour cela, nous avons utilisé des forums qui recrutent des experts en santé pour apprendre des modèles de classification qui servent à identifier les messages postés par les experts dans n’importe quel autre forum. Nous démontrons que les modèles appris sur des forums appropriés peuvent être utilisés efficacement sur d’autres forums. Puis, dans un second temps, nous nous intéressons à la réputation des utilisateurs dans ces forums. L’idée est de rechercher les expressions de confiance et de méfiance exprimées dans les messages, de rechercher les destinataires de ces messages et d’utiliser ces informations pour en déduire la réputation des utilisateurs. Nous proposons une nouvelle mesure de réputation qui permet de pondérer le score de chaque réponse selon la réputation de son auteur. Des évaluations automatiques et manuelles ont démontré l’efficacité de l’approche.Dans le deuxième axe, nous nous sommes focalisés sur l’extraction de sentiments (polarité et émotion). Pour cela, dans un premier temps, nous avons commencé par construire un lexique de sentiments et d’émotions pour le Français que nous appelons FEEL (French Expanded Emotion Lexicon). Ce lexique est construit de manière semi-automatique en traduisant et en étendant son homologue Anglais NRC EmoLex. Nous avons ensuite comparé FEEL avec les lexiques Français de la littérature sur des benchmarks de référence. Les résultats ont montré que FEEL permet d’améliorer la classification des textes Français selon leurs polarités et émotions. Dans un deuxième temps, nous avons proposé d’évaluer de manière assez exhaustive différentes méthodes et ressources pour la classification de sentiments en Français. Les expérimentations menées ont permis de déterminer les caractéristiques utiles dans la classification de sentiments pour différents types de textes. Les systèmes appris se sont montrés particulièrement efficaces sur des benchmarks de référence. De manière générale, ces travaux ont ouvert des perspectives prometteuses sur diverses tâches d’analyse des réseaux sociaux pour la langue française incluant: (i) combiner plusieurs sources pour transférer la connaissance sur les utilisateurs des réseaux sociaux; (ii) la fouille des réseaux sociaux en utilisant les images, les vidéos, les géolocalisations, etc. et (iii) l'analyse multilingues de sentiment. / Social Media has changed the way we communicate between individuals, within organizations and communities. The availability of these social data opens new opportunities to understand and influence the user behavior. Therefore, Social Media Mining is experiencing a growing interest in various scientific and economic circles. In this thesis, we are specifically interested in the users of these networks whom we try to characterize in two ways: (i) their expertise and their reputations and (ii) the sentiments they express.Conventionally, social data is often mined according to its network structure. However, the textual content of the exchanged messages may reveal additional knowledge that can not be known through the analysis of the structure. Until recently, the majority of work done for the analysis of the textual content was proposed for English. The originality of this thesis is to develop methods and resources based on the textual content of the messages for French Social Media Mining.In the first axis, we initially suggest to predict the user expertise. For this, we used forums that recruit health experts to learn classification models that serve to identify messages posted by experts in any other health forum. We demonstrate that models learned on appropriate forums can be used effectively on other forums. Then, in a second step, we focus on the user reputation in these forums. The idea is to seek expressions of trust and distrust expressed in the textual content of the exchanged messages, to search the recipients of these messages and use this information to deduce users' reputation. We propose a new reputation measure that weighs the score of each response by the reputation of its author. Automatic and manual evaluations have demonstrated the effectiveness of the proposed approach.In the second axis, we focus on the extraction of sentiments (emotions and polarity). For this, we started by building a French lexicon of sentiments and emotions that we call FEEL (French Expanded Emotions Lexicon). This lexicon is built semi-automatically by translating and expanding its English counterpart NRC EmoLex. We then compare FEEL with existing French lexicons from literature on reference benchmarks. The results show that FEEL improves the classification of French texts according to their polarities and emotions. Finally, we propose to evaluate different features, methods and resources for the classification of sentiments in French. The conducted experiments have identified useful features and methods in the classification of sentiments for different types of texts. The learned systems have been particularly efficient on reference benchmarks.Generally, this work opens promising perspectives on various analytical tasks of Social Media Mining including: (i) combining multiple sources in mining Social Media users; (ii) multi-modal Social Media Mining using not just text but also image, videos, location, etc. and (iii) multilingual sentiment analysis.

Aspect-based sentiment analysis is the field of study which extracts and interpret the sentiment, usually classified as positive or negative, towards some target or aspect in an opinionated text. This doctoral dissertation details an empirical study of techniques and methods for aspect extraction in aspect-based sentiment analysis with the focus on Portuguese. Three different approaches were explored: frequency-based, relation-based and machine learning. In each one, this work shows a comparative study between a Portuguese and an English corpora and the differences found in applying the approaches. In addition, richer linguistic knowledge is also explored by using syntatic dependencies and semantic roles, leading to better results. This work lead to the establishment of new benchmarks for the aspect extraction in Portuguese. / A análise do sentimento orientada a aspectos é o campo de estudo que extrai e interpreta o sentimento, geralmente classificado como positivo ou negativo, em direção a algum alvo ou aspecto em um texto de opinião. Esta tese de doutorado detalha um estudo empírico de técnicas e métodos para extração de aspectos em análises de sentimentos baseadas em aspectos com foco na língua Portuguesa. Foram exploradas três diferentes abordagens: métodos baseados na frequências, métodos baseados na relação e métodos de aprendizagem de máquina. Em cada abordagem, este trabalho mostra um estudo comparativo entre um córpus para o Português e outro para o Inglês e as diferenças encontradas na aplicação destas abordagens. Além disso, o conhecimento linguístico mais rico também é explorado pelo uso de dependências sintáticas e papéis semânticos, levando a melhores resultados. Este trabalho resultou no estabelecimento de novos padrões de avaliação para a extração de aspectos em Português.

O sucesso da mineração de opiniões para processar automaticamente grandes quantidades de conteúdo opinativo disponíveis na Internet tem sido demonstrado como uma solução de baixa latência e mais barata para a análise de opinião pública. No presente trabalho foi investigado se é possível prever variações de intenção de voto com base em séries temporais de sentimento extraídas de comentários de notícias, utilizando três eleições brasileiras como estudo de caso. As contribuições deste estudo de caso são: a) a comparação de duas abordagens para a mineração de opiniões em conteúdo gerado por usuários em português do Brasil; b) a proposta de dois tipos de atributos discriminantes para representar o sentimento em relação a candidatos políticos a serem usados para a previsão, c) uma abordagem para prever variações de intenção de voto que é adequada para cenários de dados esparsos. Foram desenvolvidos experimentos para avaliar a influência dos atributos discriminantes propostos em relação a acurácia da previsão, e suas respectivas preparações. Os resultados mostraram uma acurácia de 70% na previsão de variações de intenção de voto positivas e negativas. Estas contribuições são importantes passos em direção a um framework que é capaz de combinar opiniões de diversas fontes para encontrar a representatividade de uma população alvo, de modo que se possa obter previsões mais confiáveis. / The success of opinion mining for automatically processing vast amounts of opinionated content available on the Internet has been demonstrated as a less expensive and lower latency solution for gathering public opinion. In this work, we investigate whether it is possible to predict variations in vote intention based on sentiment time series extracted from news comments, using three Brazilian elections as case study. The contributions of this case study are: a) the comparison of two approaches for opinion mining in user-generated content in Brazilian Portuguese; b) the proposition of two types of features to represent sentiment behavior towards political candidates that can be used for prediction, c) an approach to predict polls vote intention variations that is adequate for scenarios of sparse data. We developed experiments to assess the influence on the forecasting accuracy of the proposed features, and their respective preparation. Our results display an accuracy of 70% in predicting positive and negative variations. These are important contributions towards a more general framework that is able to blend opinions from several different sources to find representativeness of the target population, and make more reliable predictions.

A partir d’une intervention auprès de médecins du travail, cette thèse explore les conditions de développement des sentiments au travail. Elle stabilise une conceptualisation de l’affectivité distinguant affect, émotion et sentiment qui pose des rapports entre l’affect lié aux conflits de l’activité (conflits liés à la conception de l’activité comme triade vivante sujet/objet/autrui et conflits liés aux rapports entre le déjà vécu et le vivant) et les sentiments et émotions qui en sont les instruments de réalisation. Le sentiment est défini comme l’instrument de réalisation de l’affect détaché de l’événement affectif et relié à l’activité de pensée. Une analyse multimodale de nos matériaux s’attache à repérer des indices de l’affect dans trois modalités étudiées (regard, voix, mot) et des indices de développement de la pensée (développement des significations des mots et des objets de discours). Nous concluons que l’intervention en clinique de l’activité par l’exposition de l’activité qu’elle autorise et la production de débats sur les critères du travail bien fait qu’elle organise peut provoquer des affects et contribuer au développement du sentiment du travail bien fait. / Based on an intervention with occupational health physicians, this thesis explores the developmental conditions of sentiments at work. The thesis seek to stabilize a conceptualisation of affectivity distinguishing affect, emotion and sentiment, and lays the relationships between affect, which is related to conflicts of activity (conflicts related to activity as a living triad subject/object/others and conflicts related to relationships between the « already lived » and the « living »), and sentiments and emotions, which constitute its instruments of realization. Sentiment is defined as instrument of affect realization, detached of affective event, and related to thinking activity.Multimodal analysis of research materials allows the identification of affect indices, based on three studied modalities (gaze, voice, word), and development of thinking indices (development of signification of word, and discourse objects).We conclude that intervention in clinic of activity, by exposing activity and producing disputations on quality of work criteria, can cause affects and contributes to develop the sentiment of « well-done-work ».

A partir d’une intervention auprès de médecins du travail, cette thèse explore les conditions de développement des sentiments au travail. Elle stabilise une conceptualisation de l’affectivité distinguant affect, émotion et sentiment qui pose des rapports entre l’affect lié aux conflits de l’activité (conflits liés à la conception de l’activité comme triade vivante sujet/objet/autrui et conflits liés aux rapports entre le déjà vécu et le vivant) et les sentiments et émotions qui en sont les instruments de réalisation. Le sentiment est défini comme l’instrument de réalisation de l’affect détaché de l’événement affectif et relié à l’activité de pensée. Une analyse multimodale de nos matériaux s’attache à repérer des indices de l’affect dans trois modalités étudiées (regard, voix, mot) et des indices de développement de la pensée (développement des significations des mots et des objets de discours). Nous concluons que l’intervention en clinique de l’activité par l’exposition de l’activité qu’elle autorise et la production de débats sur les critères du travail bien fait qu’elle organise peut provoquer des affects et contribuer au développement du sentiment du travail bien fait. / Based on an intervention with occupational health physicians, this thesis explores the developmental conditions of sentiments at work. The thesis seek to stabilize a conceptualisation of affectivity distinguishing affect, emotion and sentiment, and lays the relationships between affect, which is related to conflicts of activity (conflicts related to activity as a living triad subject/object/others and conflicts related to relationships between the « already lived » and the « living »), and sentiments and emotions, which constitute its instruments of realization. Sentiment is defined as instrument of affect realization, detached of affective event, and related to thinking activity.Multimodal analysis of research materials allows the identification of affect indices, based on three studied modalities (gaze, voice, word), and development of thinking indices (development of signification of word, and discourse objects).We conclude that intervention in clinic of activity, by exposing activity and producing disputations on quality of work criteria, can cause affects and contributes to develop the sentiment of « well-done-work ».