Indiana University-Purdue University Indianapolis (IUPUI) / Textual emotion classification is a task in affective AI that branches from sentiment analysis and focuses on identifying emotions expressed in a given text excerpt. It has a wide variety of applications that improve human-computer interactions, particularly to empower computers to understand subjective human language better. Significant research has been done on this task, but very little of that research leverages one of the most emotion-bearing symbols we have used in modern communication: Emojis. In this thesis, we propose several transformer-based models for emotion classification that processes emojis as input tokens and leverages pretrained models and uses them , a model that processes Emojis as textual inputs and leverages DeepMoji to generate affective feature vectors used as reference when aggregating different modalities of text encoding. To evaluate ReferEmo, we experimented on the SemEval 2018 and GoEmotions datasets, two benchmark datasets for emotion classification, and achieved competitive performance compared to state-of-the-art models tested on these datasets. Notably, our model performs better on the underrepresented classes of each dataset.

No description available.

Estudos atuais na área de Interação Humano-Computador evidenciam a importância de se considerar aspectos emocionais na interação com sistemas computacionais. Acredita-se que ao permitir agentes artificiais identificar emoções de usuários, em uma interação humano-computador, torna-se possível induzir e despertar emoções a fim de estimulá-los em suas atividades. Um dos grandes desafios dos pesquisadores em Interação humano-computador é prover sistemas capazes de reconhecer, interpretar e reagir de modo inteligente e sensível às emoções do usuário, para atender aos requisitos do maior número possível de indivíduos; um dos caminhos que se apresenta é o desenvolvimento de sistemas flexíveis. O principal objetivo de se promover essa interação emotiva é contribuir para o aumento da coerência, consistência e credibilidade das reações e respostas computacionais providas durante a interação humana via interface humano-computador. Nesse contexto, surge a oportunidade de explorar sistemas computacionais capazes de identificar e inferir o estado emocional do usuário em tempo de execução. Este projeto tem como objetivo desenvolver e avaliar um modelo que possa: i.) identificar o estado emocional do usuário; ii.) prover um mecanismo de persuasão com vistas a mudar o estado emocional do usuário (com um estudo de caso em player de música) e; iii.) explorar a abordagem flexível na persuasão (de acordo com o estado emocional particular de cada usuário) através de mecanismos persuasivos que poderão variar entre um player de música, jogos e/ou vídeos. Assim, ao longo do estudo, o modelo baseado em Comitê de Classificação se mostrou eficiente na identificação das emoções básicas (alegria, aversão, medo, neutro, raiva, surpresa e tristeza) com média de acurácia superior a 80% e, ainda, observou-se a satisfação dos usuários mediante a aplicação do modelo com o player de música. / Current studies in the field of Human-Computer Interaction highlight the relevance of emotional aspects while interacting with computers systems. It is believed that allowing intelligent agents to identify users emotions, they can induce and awaken emotions in order to stimulate them while interacting with computers. A major challenge for researchers in human-computer interaction is to provide systems capable of recognizing, interpreting and reacting intelligently and sensitively to the emotions of the user, to meet the requirements of the largest possible number of individuals. One of the ways presented in this project is the development of flexible systems to meet a large number of emotions/behaviors. The main objective of promoting this emotional interaction is to contribute to increasing the coherence, consistency and credibility of reactions and computational responses provided during human interaction via human-computer interface. In this context, the opportunity arises to explore computational systems able to identify and infer the emotional state of the user at runtime. This project aims to develop and evaluate a model that can: i.) identify the emotional state of the user/developer; ii.) provide a mechanism of persuasion in order to change the emotional state of the user (with a case study in music player) and; iii.) explore the flexible approach in persuasion (according to the particular emotional state of each user) through persuasive mechanisms that may vary from a music player, games and/or videos. Thus, throughout the study, the Rating Committee based model is efficient for identification of basic emotions (happiness, disgust, fear, neutral, anger, surprise and sadness) with average accuracy higher than 80% and also noted himself to user satisfaction by applying the model to the music player.

Estudos atuais na área de Interação Humano-Computador evidenciam a importância de se considerar aspectos emocionais na interação com sistemas computacionais. Acredita-se que ao permitir agentes artificiais identificar emoções de usuários, em uma interação humano-computador, torna-se possível induzir e despertar emoções a fim de estimulá-los em suas atividades. Um dos grandes desafios dos pesquisadores em Interação humano-computador é prover sistemas capazes de reconhecer, interpretar e reagir de modo inteligente e sensível às emoções do usuário, para atender aos requisitos do maior número possível de indivíduos; um dos caminhos que se apresenta é o desenvolvimento de sistemas flexíveis. O principal objetivo de se promover essa interação emotiva é contribuir para o aumento da coerência, consistência e credibilidade das reações e respostas computacionais providas durante a interação humana via interface humano-computador. Nesse contexto, surge a oportunidade de explorar sistemas computacionais capazes de identificar e inferir o estado emocional do usuário em tempo de execução. Este projeto tem como objetivo desenvolver e avaliar um modelo que possa: i.) identificar o estado emocional do usuário; ii.) prover um mecanismo de persuasão com vistas a mudar o estado emocional do usuário (com um estudo de caso em player de música) e; iii.) explorar a abordagem flexível na persuasão (de acordo com o estado emocional particular de cada usuário) através de mecanismos persuasivos que poderão variar entre um player de música, jogos e/ou vídeos. Assim, ao longo do estudo, o modelo baseado em Comitê de Classificação se mostrou eficiente na identificação das emoções básicas (alegria, aversão, medo, neutro, raiva, surpresa e tristeza) com média de acurácia superior a 80% e, ainda, observou-se a satisfação dos usuários mediante a aplicação do modelo com o player de música. / Current studies in the field of Human-Computer Interaction highlight the relevance of emotional aspects while interacting with computers systems. It is believed that allowing intelligent agents to identify users emotions, they can induce and awaken emotions in order to stimulate them while interacting with computers. A major challenge for researchers in human-computer interaction is to provide systems capable of recognizing, interpreting and reacting intelligently and sensitively to the emotions of the user, to meet the requirements of the largest possible number of individuals. One of the ways presented in this project is the development of flexible systems to meet a large number of emotions/behaviors. The main objective of promoting this emotional interaction is to contribute to increasing the coherence, consistency and credibility of reactions and computational responses provided during human interaction via human-computer interface. In this context, the opportunity arises to explore computational systems able to identify and infer the emotional state of the user at runtime. This project aims to develop and evaluate a model that can: i.) identify the emotional state of the user/developer; ii.) provide a mechanism of persuasion in order to change the emotional state of the user (with a case study in music player) and; iii.) explore the flexible approach in persuasion (according to the particular emotional state of each user) through persuasive mechanisms that may vary from a music player, games and/or videos. Thus, throughout the study, the Rating Committee based model is efficient for identification of basic emotions (happiness, disgust, fear, neutral, anger, surprise and sadness) with average accuracy higher than 80% and also noted himself to user satisfaction by applying the model to the music player.

This work deals with the problem of text mining which is becoming more popular due to exponential growth of the data in electronic form. The work explores contemporary methods and their improvement using optimization methods, as well as the problem of text data understanding in general. The work addresses the problem in three ways: using traditional methods and their optimizations, using Big Data in train phase and abstraction through the minimization of language-dependent parts, and introduction of the new method based on the deep learning which is closer to how human reads and understands text data. The main aim of the dissertation was to propose a method for machine understanding of unstructured text data. The method was experimentally verified by classification of text data on 5 different languages – Czech, English, German, Spanish and Chinese. This demonstrates possible application to different languages families. Validation on the Yelp evaluation database achieve accuracy higher by 0.5% than current methods.

This thesis is focused on EEG processing and emotion classification within two-dimensional emotion space. First part consists of theoretical research about emotional responses of human subjects on sound, image and video stimuli. Emotions are examined from aspect of physiology and psychology. Furthermore technical overview of measurement, analysis and emotion classification within two-dimensional emotional space is discussed. Based on gathered knowledge measurement setup with audiovisual stimuli was designed and measured with two independent instruments – EGI GES400MR in laboratory conditions and Emotiv EPOC device in non-laboratory conditions. Signals were processed and emotions were classified based on chosen features. Performance of classifiers in multiple feature selection setups was evaluated.

<div>Textual emotion classification is a task in affective AI that branches from sentiment analysis and focuses on identifying emotions expressed in a given text excerpt. </div><div>It has a wide variety of applications that improve human-computer interactions, particularly to empower computers to understand subjective human language better. </div><div>Significant research has been done on this task, but very little of that research leverages one of the most emotion-bearing symbols we have used in modern communication: Emojis.</div><div>In this thesis, we propose several transformer-based models for emotion classification that processes emojis as input tokens and leverages pretrained models and uses them</div><div>, a model that processes Emojis as textual inputs and leverages DeepMoji to generate affective feature vectors used as reference when aggregating different modalities of text encoding. </div><div>To evaluate ReferEmo, we experimented on the SemEval 2018 and GoEmotions datasets, two benchmark datasets for emotion classification, and achieved competitive performance compared to state-of-the-art models tested on these datasets. Notably, our model performs better on the underrepresented classes of each dataset.</div>

Civil infrastructure serves as the driving force behind the evolution of a safe, sustainable, and efficient environment. However, the way information about civil infrastructure has been communicated to the public has been insufficient. Since every human is intrinsically different, designing, and dispersing information about civil infrastructure that accommodates everyone, while also being direct and concise has been a challenge for policymakers and other federal, state, local, and tribal civil engineering stakeholders. Though there has been a plethora of research conducted on message design and communication in other disciplines, little research has been done in the US that focuses on designing more accessible, actionable civil infrastructure messages. The objective of this research was to investigate how to improve the accessibility of civil infrastructure messages and communication infrastructure to enhance the public's ability to make daily infrastructure decisions. This research study utilized quantitative and qualitative methods to analyze and discuss various ways that civil infrastructure messages can be improved. Results from this study are based on the exploration of three different ways in which civil infrastructure messaging can be improved: policy, transportation/roadway safety, and emergency response. Data sources include eight publicly accessible energy policies from 1978-2022, a publicly available dataset of more than 75 thousand WEAs, and a dataset retrieved from Shealy et al. (2020), which collected data on 300 Virginia drivers in both rural and urban areas. A descriptive policy analysis and Flesch-Kincaid readability test were conducted to historically analyze energy policies and understand their accessibility impacts for research question 1; a brain activation network analysis was conducted and nodal network measures (i.e., network density, degree centrality) were used to investigate the cognitive response Virginia drivers had for various types of non-traditional traffic safety messages for research question 2; and sentiment analysis, emotion detection analysis, as well as a two-phased qualitative coding analysis (i.e., in-vivo coding, focused coding) were conducted to investigate how WEAs can be better designed to increase public attention and engagement for research question 3. The findings from this study demonstrate how emotional content that is present in tweets authored by community members affected by the natural disaster event can be incorporated into the WEA template. The findings from research question 1 identified potential issues with accessibility and energy policy. Also, the findings from this study describe the content included in the parallel documents that federal agencies use to communicate the most important information of a policy. The findings from research question 2 demonstrate that while the various types of non-traditional traffic safety messages produced variances in cognitive response, messages that included negative emotional content or statistics should be further explored on their impact on evoking safer driving behaviors. The findings from research question 3 reported on how emotional content could be incorporated into the template design of WEAs. The implications from this dissertation provide valuable insights for policymakers, civil engineers, transportation engineers, and emergency response stakeholders and the conclusions set the stage for future research to improve the design of more accessible civil infrastructure messages. / Doctor of Philosophy / Civil infrastructure messages are used daily, but improper design can make them difficult to understand or to continue to use over long periods of time. Also, every human is different and interprets information about civil infrastructure, which adds a level of difficulty to designing effective civil infrastructure messages. Though there has been a lot of research on the effectiveness of civil infrastructure, little research has used a human-centered design approach to improve civil infrastructure messages. This study analyzes three different ways to improve civil infrastructure messages: policy, traffic safety, and emergency response. We used publicly available energy policies from 1978-2022, data collected by co-authors from Shealy et al. (2020) to analyze the cognitive response of 300 Virginia drivers to various types of non-traditional traffic safety messages, a publicly available dataset of more than 75 thousand Wireless Emergency Alerts sent by FEMA, and a publicly available data set of more than 9.1 thousand tweets about Hurricane Harvey. To analyze this data, this research study utilized various methods to understand how easy policies are to read, to understand how the brains of Virginia drivers respond to different types of non-traditional traffic safety messages and to identify the differences between tweets and WEAs. Results from this study suggest that parallel documents should be published alongside energy policies to help the public understand the main points of the policy, establish a readability metric to use for all energy policies, continue to investigate non-traditional traffic safety messages that included negative emotional content or statistics, measure the brain activation and observe long-term driving behaviors, use more negative emotional content in templated WEAs, and use social media data to better design templated WEAs. The findings reported from this study can be beneficial for various types of civil infrastructure stakeholders such as policymakers, utilities, US State Departments of Transportation, FEMA, alerting officials, and the public to further explore ways in which the language of civil infrastructure messages can be improved to address accessibility issues with energy policy, traffic safety, and emergency response to the public.

Táto diplomová práca sa venuje automatickému rozpoznávaniu ludí v obrazových dátach s využitím konvolučných neurónových sieti na určenie polohy tváre a následnej analýze získaných dát. Výsledkom analýzy tváre je určenie pohlavia, emócie a veku osoby. Práca obsahuje popis použitých architektúr konvolučných sietí pre každú podúlohu. Sieť na odhad veku má natrénované nové váhy, ktoré sú vzápätí zmrazené a majú do svojej architektúry vložené LSTM vrstvy. Tieto vrstvy sú samostatne dotrénované a testované na novom datasete vytvorenom pre tento účel. Výsledky testov ukazujú zlepšenie predikcie veku. Riešenie pre rýchlu, robustnú a modulárnu detekciu tváre a ďalších ludských rysov z jedného obrazu alebo videa je prezentované ako kombinácia prepojených konvolučných sietí. Tieto sú implementované v podobe skriptu a následne vysvetlené. Ich rýchlosť je dostatočná pre ďalšie dodatočné analýzy tváre na živých obrazových dátach.