02 Mars 2021
14:00 Master's Defense Fully distance
Theme
Context-based intention classification in natural language dialog flow processing
Student
Jeanfranco David Farfan Escobedo
Advisor / Teacher
Julio Cesar dos Reis
Brief summary
Classifying intentions in conversations is a challenging task, specifically if messages are collected from real data. Typically, text messages contain a variety of grammatical errors, outliers, language slang, etc. Most existing intention classification methods use only the current statement (sentence) to predict the intentions of a given flow of conversations. In fact, they do not consider the role of the context (one or a few previous statements) in the flow of dialogue for this task. This Master's Dissertation studies approaches to investigate the role of contextual information for the problem of classification of intentions. Our intention classification method is based on a convolutional neural network that obtains vector representations of BERT to perform an accurate classification of speech acts with context. Experimental results in a Brazilian Portuguese corpus of real data collected and anonymized reveal the relevance in addressing the context for the classification of intention in conversation flows.
Examination Board
Headlines:
Julio Cesar dos Reis IC / UNICAMP
Rodrigo Bonacin CTI Renato Archer
Helio Pedrini IC / UNICAMP
Substitutes:
Esther Luna Colombini IC / UNICAMP
Ana Maria Monteiro UNIFACCAMP