Dissertations / Theses on the topic 'Twitter sentiment polarity classification'

Questo lavoro di tesi si pone l'obiettivo di fornire un'ampia panoramica sull'attuale stato dell'arte della ricerca sulla sentiment analysis mostrando le metodologie, le tecniche e le applicazioni realizzate negli ultimi anni e di presentare le implementazioni concrete (ed i risultati ottenuti) di due diversi sistemi per la sentiment polarity classification di tweet per la lingua italiana. Il primo sistema (FICLIT+CS@Unibo System) utilizza un approccio basato sull'orientamento semantico tramite la realizzazione e l'utilizzo di un lessico annotato e la propagazione della polarità lungo alberi sintattici mentre il secondo utilizza algoritmi stocastico/statistici di machine learning per la creazione di un modello generalizzato per la classificazione del sentimento a partire da un training set annotato.

Questo lavoro di tesi ha portato alla realizzazione di un sistema di polarity classification per Twitter in lingua italiana. Dato un insieme di keyword, l'obiettivo posto era quello di effettuare la ricerca ed il recupero di tweet inerenti, analizzare i risultati definendo la polarità di ogni tweet e mostrarli graficamente all'utente; ponendo particolare attenzione alla qualità dell'analisi dei tweet ed affidando il recupero e la visualizzazione grafica a sistemi esistenti. Gli studi e gli approfondimenti effettuati hanno portato alla realizzazione di un sistema di classificazione supervisionato. Il primo passo del sistema implementato consiste in un preprocessing che sfrutta le caratteristiche intrinsiche di Twitter: emoticons, emoji, hashtag, ecc. Una volta terminato il preprocessing, i tweet sono stati rappresentati vettorialmente utilizzando il metodo Paragraph Vector, nello specifico l'implementazione Doc2Vec presente nella libreria Gensim. La classificazione avviene utilizzando due Convolutional Neural Network (CNN), la prima determina se un tweet è positivo o no e la seconda agisce nello stesso modo, ma determinando se è negativo o meno. In questo modo i tweet, mediante la combinazione dei risultati di entrambi i classificatori, vengono divisi in quattro categorie: positivo, negativo, neutro e misto La valutazione del sistema è stata effettuata utilizzando i tweet di addestramento e test, forniti nella campagna di valutazione EVALITA 2016. L'idea implementata è innovativa, dato che non è mai stato presentato ad EVALITA un sistema che unisse il risultato di un modello Doc2Vec con un classificatore CNN. Il modello implementato si sarebbe classificato in seconda posizione, dimostrando le sue ottime prestazioni.

Sentiment analysis is a field within the area of natural language processing that studies the sentiment of human written text. Within sentiment analysis, sentiment classification is a research area that has been of growing interest since the advent of digital social-media platforms, concerned with the classification of the subjective information in text data. Many studies have been conducted on sentiment classification, producing numerous of openly available tools and resources that further advance research, though almost exclusively for the English language. There are very few openly available Swedish resources that aid research, and sentiment classification research in non-English languages most often use English resources one way or another. The lack of non-English resources impedes research in other languages and there is very little research on sentiment classification using Swedish resources. This thesis addresses the lack of knowledge in this area by designing and implementing a sentiment classifier using Swedish resources, in order to evaluate how methods and best practices commonly used in English research transfer to Swedish. The results in this thesis indicate that Swedish resources can be used in the construction of internationally competitive sentiment classifiers and that methods commonly used in English research for pre- processing text data may not be optimal for the Swedish language.

The social micro-blog site Twitter grows in user base each day and has become an attractive platform for companies, politicians, marketeers, and others wishing to share information and/or opinions. With a growing user market for Twitter, more and more systems and research are released for taking advantage of its informal nature and doing opinion mining and sentiment analysis. This master thesis describes a system for doing Sentiment Analysis on Twitter data and experiments with grid searches on various combinations of machine learning algorithms, features and preprocessing methods to achieve so. The classification system is fairly domain independent and performs better than baseline. This system is designed to be fast enough to classify big amounts of data and tweets in a stream, and provides an application program interface (API) to easily transfer data to applications or end users. Three visualisation applications are implemented, showing how to use the API and providing examples of how sentiment data can be used.The main contributions are: C1: A literary study of the state-of-the-art for Twitter Sentiment Analysis.C2: The implementation of a general system architecture for doing Twitter Sentiment Analysis. C3: A comparison of different machine learning algorithms for the task of identifying sentiments in short messages in a fairly semi-independent domain.C4: Implementations of a set of visualisation applications, showing how to use data from the generic system and providing examples of how to present sentiment analysis data.

Language models can be applied to a diverse set of tasks with great results, but training a language model can unfortunately be a costly task, both in time and money. By transferring knowledge from one domain to another, the costly training only has to be performed once, thus opening the door for more applications. Most current research is carried out with English as the language of choice, thus limiting the amount of available already trained language models in other languages. This thesis explores how the amount of data available for training a language model effects the performance on a Twitter sentiment classification task, and was carried out using Swedish as the language of choice. The Swedish Wikipedia was used as a source for pre-training the language models which were then transferred over to a domain consisting of Swedish tweets. Several models were trained using different amounts of data from these two domains in order to compare the performance of these models. The results of the model evaluation shows that transferring knowledge from the Swedish Wikipedia to tweets yield little to no improvements, while unsupervised fine-tuning on tweets give raise to large improvements in performance.
Språkmodeller kan appliceras på en mängd olika uppgifter med bra resultat, men att träna en språkmodell kan dessvärre vara kostsamt både tids- och pengamässigt. Genom att överföra information från en domän till en annan behöver denna kostsamma träningsprocess bara genomföras en gång, och ger således lättare tillgång till dessa modeller. Dagens forskning genomförs främst med engelska som språk vilket således begränsar mängden av färdigtränade modeller på andra språk. Denna rapport utforskar hur mängden data tillgänglig för träning av språkmodeller påverkar resultatet i ett problem gällande attitydanalys av tweets, och utfördes med svenska som språk. Svenska Wikipedia användes för att först träna språkmodellerna som sedan överfördes till en domän bestående av tweets på svenska. Ett flertal språkmodeller tränades med olika mängd data från dessa två domäner för att sedan kunna jämföra deras prestanda. Resultaten visar att överföring av kunskap från Wikipedia till tweets knappt gav upphov till någon förbättring, medan oövervakad träning på tweets förbättrade resultaten markant.

Sentiment analysis on social media is an important part of today's need for information gathering. Different machine learning techniques have been used in recent years, and usage of an emoticon heuristic to automatically annotate training sets has been a popular approach. As emojis are becoming more popular to use in text-based communication this thesis investigates the feasibility of an emoji training heuristic for multi-class sentiment analysis using a Multinomial Naive Bayes Classifier. Training sets consisting of 4000 to 400 000 tweets were used to train the classifier using various configurations of N-grams. The results show that an emoji heuristic performs well compared to emoticon- or hashtag-based heuristics. However, classifier confusion is highly dependent on class selection and emoji representations when multi-class sentiment analysis is performed.
Sentimentanalys är ett problem av stor vikt på sociala medier. Ett flertal olika maskininlärningstekniker har använts på senare år och att använda en träningsmängd som är automatiskt annoterad med hälp av en heuristik baserad på så kallade emoticons har varit ett populärt angreppssätt. Användningen av så kallade emojis i textbaserad kommunikation har ökat på sistone. I linje med denna utveckling så ämnar studien att undersöka om det är hållbart med användning av en heuristik baserad på emojis för flerklassig sentimentanalys. Detta undersöks med hjälp av en Multinomial Naive Bayes-klassificerare som tränas med mängder av storlek 4000 till 400 000 (stycken tweets) och olika variationer av N-gram. Resultatet visar att en emojibaserad heuristik fungerar bra jämfört med en som är baserad på hashtags eller emoticons. Dock så har val av klasser och emojirepresentationer en stor påverkan på förvirringen hos klassificeraren.

This work is concerned with the automatic understanding of evaluative text. We investigate sentence level opinion polarity prediction by assigning lexical polarities and deriving sentence polarity from these with the use of contextual valence shifters. A methodology for iterative failure analysis is developed and used to refine our lexicon and identify new contextual shifters. Algorithms are presented that employ these new shifters to improve sentence polarity prediction accuracy beyond that of a state-of-the-art existing algorithm in the domain of consumer product reviews. We then apply the best configuration of our algorithm to the domain of movie reviews.

We classify messages posted to social media network Twitter based on the sentiment and topic of the messages. We use the results of the classification to sometimes generate responses that are sent to the original user and their network on Twitter using natural language processing. A network of users who post science related content is used as the sources of data. The classifications of the dataset show worse results than others have achieved for sentiment analysis of content on Twitter, possibly due to the data sets that were used.

Submitted by Gabriela Araujo (gabriela.denise@unifesp.br) on 2017-09-12T21:02:52Z No. of bitstreams: 1 DISSERTAÇÃO - GABRIELA DENISE DE ARAUJO.pdf: 1482312 bytes, checksum: 96da3bfe95afe2bd4424ada9c8c7b89a (MD5)
Approved for entry into archive by Diogo Misoguti (diogo.misoguti@gmail.com) on 2017-09-20T14:18:49Z (GMT) No. of bitstreams: 1 DISSERTAÇÃO - GABRIELA DENISE DE ARAUJO.pdf: 1482312 bytes, checksum: 96da3bfe95afe2bd4424ada9c8c7b89a (MD5)
Made available in DSpace on 2017-09-20T14:18:49Z (GMT). No. of bitstreams: 1 DISSERTAÇÃO - GABRIELA DENISE DE ARAUJO.pdf: 1482312 bytes, checksum: 96da3bfe95afe2bd4424ada9c8c7b89a (MD5) Previous issue date: 2014-07-31
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)
Objetivo: Construir um método de classificação de sentimento, aqui denominado Sentiment Descriptor Indexing (SDI) ou Indexador de Descritores Sentimentais, para ser aplicado em mensagens do Twitter em português brasileiro relacionadas a temas de saúde possibilitando oferecer uma análise de sentimento com caracterização de aspectos da popularidade e repercussão dos temas. Métodos: A primeira etapa considerou a construção do algoritmo SDI que se baseia na coocorrência de termos do Twitter com descritores do vocabulário ANEW-BR. Emoticons e tratamento de negação foram incorporados no SDI. Na segunda etapa foi realizada uma avaliação do desempenho do algoritmo SDI para mensagens sobre o tema “câncer” de um pe-ríodo de três semanas. As mensagens foram classificadas por voluntários como sa-úde ou não saúde, e positiva, negativa ou neutra e em paralelo pelo SDI. As classifi-cações foram pareadas gerando uma avaliação de desempenho. Também foram geradas análise de sentimento e nuvem de termos. Na terceira etapa foi realizado um experimento de análise de sentimento para os temas “câncer” e “diabetes” em um período de seis meses, com análises de repercussão e popularidade. Resulta-dos: As classificações humana e SDI concordaram na classificação majoritária posi-tiva. Os valores de precisão e revocação resultaram 0,68 e 0,67 respectivamente, gerando melhor desempenho com f0,5-measure 0,68. No experimento coletou-se um total de 25.230 mensagens sobre o tema "câncer" com classificação de sentimento positiva (71%). Pela nuvem de palavras foi possível observar que celebridades, insti-tutos, hospitais, campanhas de saúde e tipos de câncer são assuntos populares so-bre o tema. Para o tema "diabetes" 3.328 mensagens foram coletadas com classifi-cação de sentimento positiva (78%). Para este tema as palavras mais frequentes, indicadas na nuvem de palavras, estavam relacionadas a alimentos e doenças como obesidade e hipertensão. Conclusão: Os resultados obtidos na etapa de avaliação do classificador SDI mostrou que o SDI teve um bom desempenho na tarefa de clas-sificar mensagens do Twitter sobre saúde comparada a classificação realizada por humanos. Entretanto, o tema escolhido retornou mensagens difíceis de serem rotu-ladas até mesmo pelos humanos, gerando discordâncias nas classificações. As con-tribuições deste trabalho visam suprir a falta de métodos de análise de sentimentos para a língua portuguesa brasileira bem como incentivar sua aplicação na melhoria de outras atividades em processamento de linguagem natural.
Objective: Build a sentiment classification method, named Sentiment Descriptor In-dexing (SDI), to be applied in Twitter’s messages in brazilian portuguese related to health topics, providing sentiment analysis with characterization of aspects of the popularity and impact of issues. Methods: The first step regarded the SDI algorithm construction that it is based on the cooccurence of Twitter's terms with descriptors of ANEW-BR vocabulary. Emoticons and deny treatment were embedded in the SDI. In the second step, an evaluation was performed in the algorithm SDI for messages related the topic "cancer" collected in a period of three weeks. The messages were classified by volunteers in topic about health or not health, and positive, negative or neutral and in parallel by the SDI. The ratings were paired generating a performance evaluation, sentiment analysis and cloud of terms. In the third step an experiment of sentiment analysis was performed for the topics "cancer" and "diabetes" in a period of six months, with analysis of impact and popularity. Results: The human and SDI classifications agreed in positive majority classification. The values of precision and recall resulted 0.68 and 0.67 respectively, the best performance was in f0,5-measure 0,68. In experiment, it was collected a total of 25,230 messages on "cancer" and the sentiment classification of these messages was positive (71%). Through the cloud of words was possible to observe that celebrities, institutes, hospitals, health campaigns and types of cancers are popular subjects on the topic. For the topic "diabetes", 3,328 messages were collected and the sentimental classification was positive (78%). For this topic the most frequent words, given the cloud of words were related to food and diseases such as obesity and hypertension. Conclusions: The results obtained in the evaluation step showed that the SDI had a good performance in the task of classifying Twitter’s messages about health topics compared the classification performed by humans. However, the topic chosen brought messages difficult to be labeled even by humans, causing disagreements in the classifications among them. The contributions of this work aims to meet the lack of sentiment analysis methods for the brazilian portuguese language and encourage its application in improving oth-er activities in natural language processing.

Il problema relativo alla predizione, la ricerca di pattern predittivi all‘interno dei dati, è stato studiato ampiamente. Molte metodologie robuste ed efficienti sono state sviluppate, procedimenti che si basano sull‘analisi di informazioni numeriche strutturate. Quella testuale, d‘altro canto, è una tipologia di informazione fortemente destrutturata. Quindi, una immediata conclusione, porterebbe a pensare che per l‘analisi predittiva su dati testuali sia necessario sviluppare metodi completamente diversi da quelli ben noti dalle tecniche di data mining. Un problema di predizione può essere risolto utilizzando invece gli stessi metodi : dati testuali e documenti possono essere trasformati in valori numerici, considerando per esempio l‘assenza o la presenza di termini, rendendo di fatto possibile una utilizzazione efficiente delle tecniche già sviluppate. Il text mining abilita la congiunzione di concetti da campi di applicazione estremamente eterogenei. Con l‘immensa quantità di dati testuali presenti, basti pensare, sul World Wide Web, ed in continua crescita a causa dell‘utilizzo pervasivo di smartphones e computers, i campi di applicazione delle analisi di tipo testuale divengono innumerevoli. L‘avvento e la diffusione dei social networks e della pratica di micro blogging abilita le persone alla condivisione di opinioni e stati d‘animo, creando un corpus testuale di dimensioni incalcolabili aggiornato giornalmente. Le nuove tecniche di Sentiment Analysis, o Opinion Mining, si occupano di analizzare lo stato emotivo o la tipologia di opinione espressa all‘interno di un documento testuale. Esse sono discipline attraverso le quali, per esempio, estrarre indicatori dello stato d‘animo di un individuo, oppure di un insieme di individui, creando una rappresentazione dello stato emotivo sociale. L‘andamento dello stato emotivo sociale può condizionare macroscopicamente l‘evolvere di eventi globali? Studi in campo di Economia e Finanza Comportamentale assicurano un legame fra stato emotivo, capacità nel prendere decisioni ed indicatori economici. Grazie alle tecniche disponibili ed alla mole di dati testuali continuamente aggiornati riguardanti lo stato d‘animo di milioni di individui diviene possibile analizzare tali correlazioni. In questo studio viene costruito un sistema per la previsione delle variazioni di indici di borsa, basandosi su dati testuali estratti dalla piattaforma di microblogging Twitter, sotto forma di tweets pubblici; tale sistema include tecniche di miglioramento della previsione basate sullo studio di similarità dei testi, categorizzandone il contributo effettivo alla previsione.

I sistemi di Social Media Monitoring hanno l'obiettivo di analizzare dati provenienti da social media come social network, forum e blog (detti User-Generated Content) per trarre un quadro generale delle opinioni degli utenti a proposito di un particolare argomento. Il progetto di tesi si pone l'obiettivo di progettare e creare un prototipo per un sistema di Social Media Monitoring concentrato in particolare sull'analisi di contenuti provenienti da Twitter.

Social media today is a dominant communication tool, which structures not only our social interactions but also filter the information users are getting displayed. The big social media platforms use our interaction data to analyse our behaviour and sell the data for commercial interest. But not only the pure interaction data is valuable for these platforms. Also hidden information, which can be derived from our interactive networks, about our social structures, social classifications and social status are gathered and monetised. This research attempts on the one hand to uncover some of these methods used by social media platforms, and on the other hand, also wants to show how useful these new methods can be for research on social phenomena. Therefore, this study goes beyond the confining limits of traditional sociology, where either qualitative or quantitative methods are applied. Following the idea of Critical Realism, the positivist and constructivist methods are applied in combination in order to provide thick accounts of the studied material. In this study, varying socioeconomic classification systems (like the Sinus-Milieu models) are investigated and evaluated against the background of Bourdieu’s ideas on cultural and social forms of capital. The present study uses a mixed method approach (Social Network Analysis and Sentiment Analysis) to analyse quantitative data from Twitter conversations which were collected during the Austrian EU Election 2019. In conclusion, one could say that the overall purpose of this study is to demonstrate the usefulness of Critical Realism for social media research, since this approach can create a thicker account of the studied material than other, more traditional methods. This undertaking is demonstrated by the findings of the study. These findings are the building of specific sub-clusters of EU candidates which are not related to the same political background and traditional demographics but whose relation can be detected and described using Bourdieu’s concepts of social and cultural capital. As a mean for gathering empirical data, Twitter turned out to be a useful and accessible tool for this study.

O sentimento em opiniões terceiras sempre foi e continua a despertar interesse e extrema preocupação por parte dos gestores ou tomadores de decisões. Terceiros podem ser indivíduos, produtos, entidades empresariais, instituições de pesquisa e órgãos governamentais etc. dado que força de expressões e ideias controversas podem causar grandes celeumas. Desta forma o feedback emocional pode ser propulsor de mudanças, no sentido de proporcionar a busca contínua de melhorias por um lado ou determinar por outro o insucesso da entidade. Logo a análise de sentimentos é uma ferramenta indispensável no apoio ao processo de tomada de decisões. No processo de análise de sentimentos focamos na classificação de polaridade de opiniões em textos ou documentos em termos positiva ou negativa. Uma gama de Aplicações e recursos actualmente está alinhada à Língua Inglesa, o que mostra a existência de poucas experiências noutras Línguas. O objectivo é desenvolver um protótipo como ferramenta de auxílio a análise de opiniões em textos, auxiliado pelas técnicas de Aprendizado em Automática e Processamento em Linguagem Natural. Na realização de experiências aplicamos a Classificação Supervisionado a dois Corpus nomeadamente Review Movie e SentiCorpus-pt, que contém textos com opiniões sobre diversos filmes e políticos Portugueses participantes de um debate eleitoral respectivamente. A metodologia aplicada baseia-se na classificação de padrões linguísticos tais como PosTag, Chunking e outras formas simples de negação. Para a melhorar a classificação determinamos a orientação semântica das palavras, desde os seus recursos léxicos através do SentiWordnet Sense, que é uma ferramenta em Inglês que faz a tradução de qualquer língua para o Inglês, antes de extracção de polaridade. A nossa abordagem é avaliar os dois corpora. Tarefa que é auxiliada pelo casamento de termos linguísticos com vista a melhorar a performance. O classificador utiliza para tal o modelo Balsa de Palavras como linha de base; ABSTRACT: The feeling for third parties opinions has always been and it continues to arouse interest and extreme concern by managers or decision makers, third parties can be individuals, products, business organizations, research institutions and government institutions etc.. Knowing that expressions force and controversial ideas can cause major embarrassment. ln this way, the emotional feedback can be change propeller, in order to provide continuous search for improvements on one side or to determine on the other hands the failure of the entity. So, the sentiment analysis is an indispensable tool to support the process of decision making. ln the process of sentiment analysis we have focused on the classification of polarity of opinions in texts or documents in positive or negative terms. A range of applications and resources nowadays are aligned with English Language, which demonstrates the existence of few experiences in other languages. The aim is to develop a prototype as a supporting tool to analise texts opinions, supported by learning techniques on Machine and Natural Language Processing. When conducting experiments, we have applied the Supervised Classification into two Corpus namely Review Movie and SentiCorpus-pt, which contains texts with opinions of different films and Portuguesa politician’s participants in electoral debate respectively. The applied methodology is based on the classification of linguistic patterns such as Pos-tag, Chunking is other simple form of denial. To improve the classification and determine the semantic orientation of words, from the lexicon resources through SentiWordnet Sense, which is a tool in English that makes the translation from any language to English before identifying the polarity. Our approach is to evaluate the two corpus. Task that is supported by the marriage of linguistic terms in order to improve the performance. For such, the classifier uses the bags word model as the baseline.

碩士
國立臺灣大學
資訊工程學研究所
100
The researches of sentiment analysis aim at exploring the emotional state of writers. The analysis highly depends on the application domains. Analyzing sentiments of the articles in different domains may have different results. In this study, we focus on corpora from three different domains, then examine the polarity degrees of vocabularies in these three domains, and propose methods to capture sentiment differences. Finally, we apply the results to sentiment classification with supervised SVM learning. The experiments show that the proposed methods can effectively improve the sentiment classification performance.

Due to the advances in location-acquisition techniques, citizen geospatial data has emerged with opportunity for research, development, innovation, and business. A variety of research has been developed to study society and citizens through exploring patterns from geospatial data. In this thesis, we investigate patterns of population and human sentiments using GPS trajectory data and geo-tagged tweets. Kernel density estimation and emerging hot spot analysis are first used to demonstrate population distribution across space and time. Then a flow extraction model is proposed based on density difference for human movement detection and visualization. Case studies with volleyball game in West Lafayette and traffics in Puerto Rico verify the effectiveness of this method. Flow maps are capable of tracking clustering behaviors and direction maps drawn upon the orientation of vectors can precisely identify location of events. This thesis also analyzes patterns of human sentiments. Polarity of tweets is represented by a numeric value based on linguistics rules. Sentiments of four US college cities are analyzed according to its distribution on citizen, time, and space. The research result suggests that social media can be used to understand patterns of public sentiment and well-being.