Publication: Generación de narrativa basada en criterios de calidad aprendidos automáticamente
Loading...
Official URL
Full text at PDC
Publication Date
2016
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En la sociedad actual, tenemos en alta estima a aquellas personas que demuestran tener un alto grado de creatividad, ya que implica la correcta aplicación de habilidades cognitivas que comúnmente consideramos reflejo de inteligencia. Con este proyecto, intentamos arrojar algo de luz sobre la creatividad computacional, concentrándonos en el ámbito de la generación automática de historias. Veremos los diferentes paradigmas existentes para la generación automática de historias, así como los sistemas previamente implementados, que nos han servido de referencia para completar el nuestro propio. Expondremos los detalles de nuestro sistema, un primer prototipo de un generador de historias capaz de valorar sus propios resultados, de forma que pueda filtrarlos para presentar los que considera mejores, además de poder cambiar completamente el contexto y el contenido de la historia de forma sencilla. Para ello, nos hemos basado en la generación de historias basada en gramáticas formales, con filtrado basado en un modelo generado previamente mediante aprendizaje máquina. Dicho modelo es generado a partir de las valoraciones de usuarios reales a historias generadas por el sistema para, posteriormente, analizar qué elementos de esas historias son los que desencadenan dicha valoración. De esta forma, estamos estudiando qué elementos hacen que una historia sea interesante para una persona, lo cual es especialmente interesante debido a la falta de consenso en este ámbito. A la hora de narrar las historias, hemos utilizado un enfoque basado en plantillas predefinidas por simplicidad, ya que la generación de lenguaje natural queda fuera del ámbito de este proyecto. Con todo esto, hemos conseguido implementar un generador de historias básico capaz, no sólo de generar un número muy elevado de historias diferentes, si no de valorar cuáles de esas historias son interesantes.
In nowadays society, we hold in high regard those people who demonstrate a high level of creativity, since it implies a correct application of cognitive skills that are commonly considered intelligent. With this project, we’re trying to shed some light over computational creativity, focusing on the automatic story generation field. We’ll have a look at the different existing paradigms for automatic story generation, as well as the systems previously implemented, that have served as reference for ours. We’ll expose the details of our system, a first prototype of a storyteller capable of assessing it’s own results, as well as being able to easily change the context and content of the stories generated. For that, we relied on story generation based on formal grammars, with interest level based filtering. This model is generated from the appreciations of the stories by real users to, later on, analyze which elements of the stories trigger these appreciations. This way, we’re studying which elements make a story interesting to a real person, being that specially interesting due to the lack of consensus in that field. When telling the stories, we used a template based approach for the sake of simplicity, because natural language generation is out of the scope of this project. With all this, we’ve implemented a basic story generator capable of, not only generating a high number of different stories, but able to assess which stories are the most interesting.
In nowadays society, we hold in high regard those people who demonstrate a high level of creativity, since it implies a correct application of cognitive skills that are commonly considered intelligent. With this project, we’re trying to shed some light over computational creativity, focusing on the automatic story generation field. We’ll have a look at the different existing paradigms for automatic story generation, as well as the systems previously implemented, that have served as reference for ours. We’ll expose the details of our system, a first prototype of a storyteller capable of assessing it’s own results, as well as being able to easily change the context and content of the stories generated. For that, we relied on story generation based on formal grammars, with interest level based filtering. This model is generated from the appreciations of the stories by real users to, later on, analyze which elements of the stories trigger these appreciations. This way, we’re studying which elements make a story interesting to a real person, being that specially interesting due to the lack of consensus in that field. When telling the stories, we used a template based approach for the sake of simplicity, because natural language generation is out of the scope of this project. With all this, we’ve implemented a basic story generator capable of, not only generating a high number of different stories, but able to assess which stories are the most interesting.
Description
Máster en Ingeniería Informática, Facultad de Informática, Departamento de Ingeniería del Software e Inteligencia Artificial, curso 2015-2016