Natural language understanding in argumentative dialogue systems

Thumbnail Image

Files

NLU_in_ADS.pdf (760.32 KB)

Date

2020-01-09

Authors

Journal Title

Journal ISSN

Volume Title

Publication Type

Abschlussarbeit (Master; Diplom)

DOI

http://dx.doi.org/10.18725/OPARU-23716

Published in

Abstract

This thesis presents various techniques to implement the natural language understanding in argumentative dialog systems. We examine various models like the ones which exploit the linguistic properties of the English language, models which rely on the vector representation of words, with different similarity measurement techniques. In order to structure the user responses, we formalize the user interaction as an argument game. And we explore chat bot designs in order to understand the user intents in the game. The models are tested for the data obtained from Wikipedia pages. We also collect the real user responses and evaluate the model. The output indicates that, the training data, model architecture and the similarity measurement techniques play a significant role in the performance. In addition, the models based on linguistic properties perform better than the ones based on vector representation.

Description

Faculties

Fakultät für Ingenieurwissenschaften, Informatik und Psychologie

Institutions

Institut für Nachrichtentechnik
Institut für Organisation und Management von Informationssystemen

Citation

DFG Project uulm

Keywords

Natural language understanding, WordNet, Word embedding, Word2Vec, GloVe, BERT, Semantic similarity measures, RASA, Intent classification, Dialogue game for argumentation, Utterance mapping, Künstliche Intelligenz, Digitale Sprachverarbeitung, Natürliche Sprache, Automatische Sprachanalyse, Artificial intelligence, Natural language processing (Computer science), Computational linguistics, Semantics; Data processing, Automatic speech recognition, DDC 004 / Data processing & computer science