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.

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.

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.

<p>When people write a review about a business, they write and rate it based on their personal experience of the business. Sentiment analysis is a natural language processing technique that determines the sentiment of text, including reviews. However, unlike computers, the personal experience of humans emphasizes their preferences and observations that they deem important while ignoring other components that may not be as important to them personally. Traditional sentiment analysis does not consider such preferences. To utilize these human preferences in sentiment analysis, this paper explores various methods of weighting aspects in an attempt to improve sentiment analysis accuracy. Two types of methods are considered. The first method applies human preference by assigning weights to aspects in calculating overall sentiment analysis. The second method uses the results of the first method to improve the accuracy of traditional supervised sentiment analysis. The results show that the methods have high accuracy when people have strong opinions, but the weights of the aspects do not significantly improve the accuracy.</p>

A análise de sentimentos é um campo de estudo com recente popularização devido ao crescimento da Internet e do conteúdo que é gerado por seus usuários, principalmente nas redes sociais, nas quais as pessoas publicam suas opiniões em uma linguagem coloquial e em muitos casos utilizando de artifícios gráficos para tornar ainda mais sucintos seus diálogos. Esse cenário é observado no Twitter, uma ferramenta de comunicação que pode facilmente ser usada como fonte de informação para várias ferramentas automáticas de inferência de sentimentos. Esforços de pesquisas têm sido direcionados para tratar o problema de análise de sentimentos em redes sociais sob o ponto de vista de um problema de classificação, com pouco consenso sobre qual é o classificador com melhor poder preditivo, bem como qual é a configuração fornecida pela engenharia de atributos que melhor representa os textos. Outro problema é que em um cenário supervisionado, para a etapa de treinamento do modelo de classificação, é imprescindível se dispor de exemplos rotulados, uma tarefa árdua e que demanda esforço humano em grande parte das aplicações. Esta tese tem por objetivo investigar o uso de agregadores de classificadores (classifier ensembles), explorando a diversidade e a potencialidade de várias abordagens supervisionadas quando estas atuam em conjunto, além de um estudo detalhado da fase que antecede a escolha do classificador, a qual é conhecida como engenharia de atributos. Além destes aspectos, um estudo mostrando que o aprendizado não supervisionado pode fornecer restrições complementares úteis para melhorar a capacidade de generalização de classificadores de sentimento é realizado, fornecendo evidências de que ganhos já observados em outras áreas do conhecimento também podem ser obtidos no domínio em questão. A partir dos promissores resultados experimentais obtidos no cenário de aprendizado supervisionado, alavancados pelo uso de técnicas não supervisionadas, um algoritmo existente, denominado de C3E (Consensus between Classification and Clustering Ensembles) foi adaptado e estendido para o cenário semissupervisionado. Este algoritmo refina a classificação de sentimentos a partir de informações adicionais providas pelo agrupamento em um procedimento de autotreinamento (self-training). Tal abordagem apresenta resultados promissores e competitivos com abordagens que representam o estado da arte em outros domínios. / Sentiment analysis is a field of study that shows recent popularization due to the growth of Internet and the content that is generated by its users. More recently, social networks have emerged, where people post their opinions in colloquial and compact language. This is what happens in Twitter, a communication tool that can easily be used as a source of information for various automatic tools of sentiment inference. Research efforts have been directed to deal with the problem of sentiment analysis in social networks from the point of view of a classification problem, where there is no consensus about what is the best classifier, and what is the best configuration provided by the feature engineering process. Another problem is that in a supervised setting, for the training stage of the classification model, we need labeled examples, which are hard to get in the most of applications. The objective of this thesis is to investigate the use of classifier ensembles, exploring the diversity and the potential of various supervised approaches when these work together, as well as to provide a study about the phase that precedes the choice of the classifier, which is known as feature engineering. In addition to these aspects, a study showing that unsupervised learning techniques can provide useful and additional constraints to improve the ability of generalization of the classifiers is also carried out. Based on the promising results got in supervised learning settings, an existing algorithm called C3E (Consensus between Classification and Clustering Ensembles) was adapted and extended for the semi-supervised setting. This algorithm refines the sentiment classification from additional information provided by clusters of data, in a self-training procedure. This approach shows promising results when compared with state of the art algorithms.

A análise de sentimentos é um campo de estudo com recente popularização devido ao crescimento da Internet e do conteúdo que é gerado por seus usuários, principalmente nas redes sociais, nas quais as pessoas publicam suas opiniões em uma linguagem coloquial e em muitos casos utilizando de artifícios gráficos para tornar ainda mais sucintos seus diálogos. Esse cenário é observado no Twitter, uma ferramenta de comunicação que pode facilmente ser usada como fonte de informação para várias ferramentas automáticas de inferência de sentimentos. Esforços de pesquisas têm sido direcionados para tratar o problema de análise de sentimentos em redes sociais sob o ponto de vista de um problema de classificação, com pouco consenso sobre qual é o classificador com melhor poder preditivo, bem como qual é a configuração fornecida pela engenharia de atributos que melhor representa os textos. Outro problema é que em um cenário supervisionado, para a etapa de treinamento do modelo de classificação, é imprescindível se dispor de exemplos rotulados, uma tarefa árdua e que demanda esforço humano em grande parte das aplicações. Esta tese tem por objetivo investigar o uso de agregadores de classificadores (classifier ensembles), explorando a diversidade e a potencialidade de várias abordagens supervisionadas quando estas atuam em conjunto, além de um estudo detalhado da fase que antecede a escolha do classificador, a qual é conhecida como engenharia de atributos. Além destes aspectos, um estudo mostrando que o aprendizado não supervisionado pode fornecer restrições complementares úteis para melhorar a capacidade de generalização de classificadores de sentimento é realizado, fornecendo evidências de que ganhos já observados em outras áreas do conhecimento também podem ser obtidos no domínio em questão. A partir dos promissores resultados experimentais obtidos no cenário de aprendizado supervisionado, alavancados pelo uso de técnicas não supervisionadas, um algoritmo existente, denominado de C3E (Consensus between Classification and Clustering Ensembles) foi adaptado e estendido para o cenário semissupervisionado. Este algoritmo refina a classificação de sentimentos a partir de informações adicionais providas pelo agrupamento em um procedimento de autotreinamento (self-training). Tal abordagem apresenta resultados promissores e competitivos com abordagens que representam o estado da arte em outros domínios. / Sentiment analysis is a field of study that shows recent popularization due to the growth of Internet and the content that is generated by its users. More recently, social networks have emerged, where people post their opinions in colloquial and compact language. This is what happens in Twitter, a communication tool that can easily be used as a source of information for various automatic tools of sentiment inference. Research efforts have been directed to deal with the problem of sentiment analysis in social networks from the point of view of a classification problem, where there is no consensus about what is the best classifier, and what is the best configuration provided by the feature engineering process. Another problem is that in a supervised setting, for the training stage of the classification model, we need labeled examples, which are hard to get in the most of applications. The objective of this thesis is to investigate the use of classifier ensembles, exploring the diversity and the potential of various supervised approaches when these work together, as well as to provide a study about the phase that precedes the choice of the classifier, which is known as feature engineering. In addition to these aspects, a study showing that unsupervised learning techniques can provide useful and additional constraints to improve the ability of generalization of the classifiers is also carried out. Based on the promising results got in supervised learning settings, an existing algorithm called C3E (Consensus between Classification and Clustering Ensembles) was adapted and extended for the semi-supervised setting. This algorithm refines the sentiment classification from additional information provided by clusters of data, in a self-training procedure. This approach shows promising results when compared with state of the art algorithms.

The opinion of customers about companies are very important as this can influence a company’s profit. Companies often get customer feedback via surveys or other official methods in order to improve their services. However, some customers feel threatened when their opinions are publicly asked and thus prefer to voice their opinion on the internet where they take comfort in anonymity. This form of customer feedback is difficult to monitor as the information can be found anywhere on the internet and new information is generated at an astonishing rate. Currently there are companies such as Brandseye and Brand.Com that provide online reputation management services. These services have various shortcomings such as cost and is incapable of accessing historical data. Companies are also not allowed to purchase these software and can only use the software on a subscription basis. The design proposed in this document will be able to scan any number of user defined websites and save all the information found on the websites in a series of index files, which can be queried for occurrences of user defined keywords at any time. Additionally, the software will also be able to scan Twitter and Facebook for any number of user defined keywords and save any occurrences of the keywords to a database. After scanning the internet, the results will be passed through a similarity filter, which will filter out insignificant results as well as any duplicates that might be present. Once passed through the filter the remaining results will be analysed by a sentiment analysis tool which will determine whether the sentence in which the keyword occurs is positive or negative. The analysed results will determine the overall reputation of the keyword that was used. The proposed design has several advantages over current systems: - By using the modular design several tasks can execute at the same time without influencingeach other. For example; information can be extracted from the internet while existing resultsare being analysed. - By providing the keywords and websites that the system will use the user will have full controlover the online reputation management process. - By saving all the information contained in a website the user will be able to take historicalinformation into account to determine how the keywords reputation changes over time. Savingthe information will also allow the user to search for any keyword without rescanning theinternet. The proposed system was tested and successfully used to determine the online reputation of many user defined keywords. / MIng (Computer and Electronic Engineering), North-West University, Potchefstroom Campus, 2015

The opinion of customers about companies are very important as this can influence a company’s profit. Companies often get customer feedback via surveys or other official methods in order to improve their services. However, some customers feel threatened when their opinions are publicly asked and thus prefer to voice their opinion on the internet where they take comfort in anonymity. This form of customer feedback is difficult to monitor as the information can be found anywhere on the internet and new information is generated at an astonishing rate. Currently there are companies such as Brandseye and Brand.Com that provide online reputation management services. These services have various shortcomings such as cost and is incapable of accessing historical data. Companies are also not allowed to purchase these software and can only use the software on a subscription basis. The design proposed in this document will be able to scan any number of user defined websites and save all the information found on the websites in a series of index files, which can be queried for occurrences of user defined keywords at any time. Additionally, the software will also be able to scan Twitter and Facebook for any number of user defined keywords and save any occurrences of the keywords to a database. After scanning the internet, the results will be passed through a similarity filter, which will filter out insignificant results as well as any duplicates that might be present. Once passed through the filter the remaining results will be analysed by a sentiment analysis tool which will determine whether the sentence in which the keyword occurs is positive or negative. The analysed results will determine the overall reputation of the keyword that was used. The proposed design has several advantages over current systems: - By using the modular design several tasks can execute at the same time without influencingeach other. For example; information can be extracted from the internet while existing resultsare being analysed. - By providing the keywords and websites that the system will use the user will have full controlover the online reputation management process. - By saving all the information contained in a website the user will be able to take historicalinformation into account to determine how the keywords reputation changes over time. Savingthe information will also allow the user to search for any keyword without rescanning theinternet. The proposed system was tested and successfully used to determine the online reputation of many user defined keywords. / MIng (Computer and Electronic Engineering), North-West University, Potchefstroom Campus, 2015

Η χρήση του Twitter από ολοένα και περισσότερους ανθρώπους έχει ως συνέπεια την παραγωγή μεγάλου όγκου «υποκειμενικών» δεδομένων. Η ανάγκη για εξεύρεση τυχόν πολύτιμης κρυμμένης πληροφορίας σε αυτά τα δεδομένα, έδωσε ώθηση στην ανάπτυξη ενός νέου πεδίου έρευνας, του Sentiment Analysis, που έχει ως αντικείμενο τον εντοπισμό του συναισθήματος ενός χρήστη (ή μιας ομάδας χρηστών) ως προς κάποιο θέμα. Οι παραδοσιακοί αλγόριθμοι και μέθοδοι εντοπισμού συναισθήματος στηρίζονται στην λεκτική ανάλυση φράσεων ή προτάσεων σε «επίσημα» κείμενα και καλούνται word based approaches. Ωστόσο, το μικρό μέγεθος των κειμένων του Twitter, σε συνδυασμό με την χαλαρότητα της χρησιμοποιούμενης γλώσσας (από πλευράς χρηστών), δεν επιτρέπει την αποτελεσματική χρήση αυτών των τεχνικών. Για τον λόγο αυτό, προτιμάται η χρήση τεχνικών που βασίζονται σε χαρακτήρες (αντί για λέξεις) και καλούνται character based approaches. Στόχος της διπλωματικής εργασίας είναι η εφαρμογή της character based μεθόδου στην ανάλυση tweets πολιτικού περιεχομένου. Συγκεκριμένα, χρησιμοποιήθηκαν δεδομένα από την πολιτική σκηνή των Η.Π.Α., με σκοπό να εντοπιστεί η προτίμηση ενός χρήστη ως προς το Ρεπουμπλικανικό ή το Δημοκρατικό κόμμα μέσω σχετικών tweets. Για την ανάλυση χρησιμοποιήθηκε επιβλεπόμενη μάθηση με την βοήθεια του Naive Bayes ταξινομητή. Αρχικά, συλλέχθηκε ένα σύνολο από 7904 tweets, προερχόμενα από τους επίσημους λογαριασμούς Twitter 48 γερουσιαστών. Το σύνολο αυτό χωρίσθηκε σε δυο επιμέρους σύνολα, το σύνολο εκπαίδευσης και το σύνολο ελέγχου, ελέγχοντας για κάθε μια από τις δυο μεθόδους ανάλυσης (την word based και character based μέθοδο) την ακρίβεια της ταξινόμησης. Από τα πειράματα πρόεκυψε πως η character based μέθοδος ταξινομεί τα tweets με μεγαλύτερη ακρίβεια. Στην συνέχεια συλλέξαμε δυο νέα σύνολα έλεγχου, ένα από τον επίσημο λογαριασμό Twitter του Ρεπουμπλικανικού κόμματος και ένα από τον επίσημο λογαριασμό Twitter του Δημοκρατικού κόμματος. Αυτή την φορά, ως σύνολο εκπαίδευσης χρησιμοποιήθηκε ολόκληρο το αρχικό σύνολο από τα tweets των γερουσιαστών και ελέγχθηκε η ακρίβεια ταξινόμησης για την character based μέθοδο στα δυο νέα σύνολα ελέγχου. Αν και στην περίπτωση του Democratic Twitter account τα αποτελέσματα μπορούν να χαρακτηριστούν ως «ικανοποιητικά», μιας και η ακρίβεια της ταξινόμησης πλησίασε το 80%, για την περίπτωση του Republican Twitter account κάτι τέτοιο δεν ισχύει. Για το λόγο αυτό, προχωρήσαμε σε μια πιο διεξοδική μελέτη της δομής και του περιεχομένου αυτών tweets. Από την ανάλυση προέκυψαν ορισμένα ενδιαφέροντα αποτελέσματα για την προέλευση των χαμηλών ποσοστών στην ακρίβεια ταξινόμησης. Συγκεκριμένα, πρόεκυψε πως στην πλειοψηφία των tweets που έγιναν από τους Ρεπουμπλικάνους γερουσιαστές, δεν περιέχονταν κάποια προσωπική τους άποψη. Ήταν απλά μια αναφορά σε κάποιο άρθρο ή video που είδαν στον διαδίκτυο. Άρα, η πλειοψηφία των tweets αυτών περιέχουν «αντικειμενική» αντί για «υποκειμενική» πληροφορία. Συνεπώς, δεν είναι δυνατόν να εξαχθούν τα χαρακτηριστικά εκείνα που θα βοηθήσουν στον εντοπισμό της πολικότητας των χρηστών. / As more people enter the “social web”, social media platforms are becoming an increasingly valuable source of subjective information. The large volume of social media content available requires automatic techniques in order to process and extract any valuable information. This need recently gave rise to the field of Sentiment Analysis, also known as Opinion Mining. The goal of sentiment analysis is to identify the position of a user (or a group of users – a crowd), with respect to a particular issue or topic. Existing sentiment analysis systems aim at extracting patterns mainly from formal documents with respect to a particular language (most techniques concern English). They either search for discriminative series of words or use dictionaries that assess the meaning and sentiment of specific words and phrases. The limited size of Twitter posts in conjunction with the non-standard vocabulary and shortened words (used by its users) inserts a great deal of noise, making word based approaches ineffective. For all of the above reasons, a new approach was recommended in the literature. This new approach is not based on the study of words but rather on the study of consecutive character sequences (namely character-based approaches). In this work, we demonstrate the superiority of the character based approach over the word based one in determining political sentiment. We argue that this approach can be used in order to efficiently determine the political preference (e.g. Republican or Democrat) of voters or to identify the importance that particular issues have on particular voters. This type of feedback can be useful in the organization of political campaigns or policies. We created a corpus consisting of 7904 tweets, collected from the Twitter accounts of 48 U.S. senators. This corpus was then separated into two sets, the training set and the test set, in order to measure for each method (word and character based) the accuracy of the classification. From the experiments it was found that the character based method classified the tweets with greater accuracy. In the next test, we used two new test sets, one from the official Twitter account of the Republican Party and one from the official Twitter account of the Democratic Party. The main difference, with respect to the previous test, was the use of the total set of tweets collected from the senators’ Twitter accounts as a training set and the use of the tweets from the official Twitter accounts of each party as a test set. Even though from the official Democrat Twitter account, 80% of the tweets were correctly classified as Democrat, for the official Republican Twitter account this is not the case (56.7% accuracy). This was found to be partly because the majority of the Republican account tweets were references to online articles or videos and not the personal opinions or views of the users. In other words, such tweets cannot be characterized as personal (subjective), in order to classify the respective user as leaning towards one party or the other, but rather should be considered as objective.

Ingeniero Civil Eléctrico / El análisis de sentimientos o sentiment analysis es el estudio por el cual se determina la opinión de las personas en Internet sobre algún tema en específico, prediciendo la polaridad de los usuarios (a favor, en contra, neutro, etc), abarcando temas que van desde productos, películas, servicios a intereses socio-culturales como elecciones, guerras, fútbol, etc. En el caso particular de esta memoria, se estudian los principales métodos usados en la literatura para realizar un análisis de sentimientos y se desarrolla un caso empleando parte de estas técnicas con sus respectivos resultados. La plataforma escogida fue Twitter, debido a su alto uso en Chile y el caso de estudio trata acerca de las elecciones presidenciales primarias realizadas en la Alianza por Chile entre los candidatos Andrés Allamand de Renovación Nacional (RN) y Pablo Longueira del partido Unión Demócrata Independiente (UDI). De esta forma, se busca predecir los resultados de las primarias, identificando la gente que está a favor de Allamand y la gente que apoya a Longueira. De igual manera, se busca identificar a los usuarios que están en contra de uno o ambos candidatos. Para predecir la opinión de los usuarios se diseñó un diccionario con palabras positivas y negativas con un puntaje asociado, de manera que al encontrar estos términos en los tweets se determina la polaridad del mensaje pudiendo ser positiva, neutra o negativa. El Algoritmo diseñado tiene un acierto cercano al 60% al ocupar las 3 categorías, mientras que si sólo se ocupa para determinar mensajes positivos y negativos la precisión llega a un 74%. Una vez catalogados los tweets se les asigna el puntaje a sus respectivos usuarios de manera de sumar estos valores a aquellas cuentas que tengan más de un tweet, para luego poder predecir el resultado de las elecciones por usuario. Finalmente, el algoritmo propuesto determina como ganador a Pablo Longueira (UDI) por sobre Andrés Allamand (RN) con un 53% de preferencia mientras que en las elecciones en urnas realizadas en Julio de 2013 en Chile el resultado fue de un 51% sobre 49% a favor de Longueira, lo cual da un error de un 2%, lo que implica que el análisis realizado fue capaz de predecir, con un cierto margen de error, lo que sucedió en las elecciones. Como trabajo futuro se plantea usar el diccionario y algoritmo diseñados para realizar un análisis de sentimientos en otro tema de interés y comprobar su efectividad para diferentes casos y plataformas.