Qi-Dashboard: visualização de informação de debates online
A carregar...
Data
Autores
Título da revista
ISSN da revista
Título do volume
Editora
Resumo
Ao longo dos últimos anos a grande quantidade de dados presente no nosso quotidiano
tem vindo a moldar a forma como trabalhamos e vivemos. Muitas disciplinas científicas
tiveram de evoluir rapidamente, e inclusive, novas disciplinas emergiram para responder
aos novos problemas do ecossistema da informação, tais como a difusão de noticias falsas,
o discurso de ódio, entre outros. O processamento estatístico de linguagem natural dentro
da Inteligência Artificial, e a visualização de informações dentro da Ciência dos Dados,
são dois exemplos de tais disciplinas, e de facto, as duas disciplinas centrais do presente
trabalho. Para esta dissertação foi implementada uma solução inicial para o problema de
extrair e interpretar informação que descreve comportamentos de grupo em situação de
debate online em contexto educativo.
A solução concreta é um dashboard visual que apresenta indicadores baseados em duas
dimensões: (1) o fluxo de mudanças de tópico ao longo das conversas, a qual é indicativa
da coerência do grupo; (2) a distribuição de interações por interveniente, a qual mostra a
media em que os intervenientes participaram de forma equilibrada ou se a conversa foi
dominada por um subconjunto, relativamente pequeno, de participantes. O cumprir os
objetivos desta dissertação significou, por um lado, um estudo das teorias e fundamentos
da visualização de dados, e, por outro, também aprender os métodos e paradigmas do
que é conhecido como ‘text as data’. Assim, demonstrou-se que é possível transformar
textos não estruturados, provenientes de interações de grupos online, em visualizações,
que capturam conhecimento útil no diagnóstico de potenciais problemas relativos à saúde
da conversa em redes sociais.
Palavras-chave: Dashboards visuais, representação vetorial de texto, modelação
de tópicos
Over the last few years, the large amount of data present in our daily lives has shaped how we work and live in society. As a result, many scientific disciplines had to evolve rapidly. Even new fields emerged to respond to the unique problems of the information ecosystem, such as the dissemination of fake news and hate speech, among others. Natural language statistical processing within Artificial Intelligence, and information visualisation within Data Science, are two examples of such disciplines, and indeed, the two major fields of the present work. For this dissertation, an initial solution was implemented to the problem of extracting and interpreting information that describes group behaviours in an online debate situation in an educational context. The concrete solution is a visual dashboard that shows indicators based on two dimensions. The first is the flow of topic changes throughout conversations, which is indicative of the group’s coherence. The second is the distribution of interactions per participant, which shows the extent to which they interacted in a balanced way or whether the conversation was dominated by a relatively small subset of participants. Fulfilling the objectives of this dissertation required a deep study of the theories and fundamentals of data visualisation. Also, it demanded learning the methods and paradigms of what is known as ‘text as data’. It is possible to transform unstructured texts originating from online group interactions into visualisations that capture helpful knowledge in diagnosing potential problems related to the health of conversation in social networks. Keywords: Visual dashboards, topic modelling, sentence embeddings
Over the last few years, the large amount of data present in our daily lives has shaped how we work and live in society. As a result, many scientific disciplines had to evolve rapidly. Even new fields emerged to respond to the unique problems of the information ecosystem, such as the dissemination of fake news and hate speech, among others. Natural language statistical processing within Artificial Intelligence, and information visualisation within Data Science, are two examples of such disciplines, and indeed, the two major fields of the present work. For this dissertation, an initial solution was implemented to the problem of extracting and interpreting information that describes group behaviours in an online debate situation in an educational context. The concrete solution is a visual dashboard that shows indicators based on two dimensions. The first is the flow of topic changes throughout conversations, which is indicative of the group’s coherence. The second is the distribution of interactions per participant, which shows the extent to which they interacted in a balanced way or whether the conversation was dominated by a relatively small subset of participants. Fulfilling the objectives of this dissertation required a deep study of the theories and fundamentals of data visualisation. Also, it demanded learning the methods and paradigms of what is known as ‘text as data’. It is possible to transform unstructured texts originating from online group interactions into visualisations that capture helpful knowledge in diagnosing potential problems related to the health of conversation in social networks. Keywords: Visual dashboards, topic modelling, sentence embeddings
Descrição
Orientação: Manuel Arturo Marques Pita
Palavras-chave
MESTRADO EM ENGENHARIA INFORMÁTICA E SISTEMAS DE INFORMAÇÃO, ENGENHARIA INFORMÁTICA, GESTÃO DA INFORMAÇÃO, SISTEMAS DE INFORMAÇÃO, COMPUTER ENGINEERING, INFORMATION MANAGEMENT, INFORMATION SYSTEMS, REPRESENTAÇÃO VETORIAL DE TEXTO, MODELAÇÃO DE TÓPICOS, DASHBOARDS, SENTENCE EMBEDDINGS, TOPIC MODELLING