logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-06192017-100155


Tipo di tesi
Tesi di laurea magistrale
Autore
MADOTTO, ANDREA
Indirizzo email
andreamad8@gmail.com
URN
etd-06192017-100155
Titolo
Question Dependent Recurrent Entity Network for Question Answering
Dipartimento
INFORMATICA
Corso di studi
INFORMATICA
Relatori
relatore Prof. Attardi, Giuseppe
Parole chiave
  • NLP QuestionAnswering
Data inizio appello
21/07/2017
Consultabilità
Completa
Riassunto
Question Answering is a task which requires building models that are able to automatically reply to questions given by humans. In the recent years, growing interest has been shown in tasks that require reasoning abilities in order to answer questions.

Thus, in this study, we propose an extensive literature review of the state-of-the-art, which is followed by the introduction of two models to accomplish different Question Answering tasks, which are: Community Question Answering, Reasoning Question Answering and Reading Comprehension.

Specially, we analysed and improved a novel model called Recurrent Entity Network, which follows the Memory Network framework, and it has shown a great potential in solving reasoning tasks. We named our model Question Dependent Recurrent Entity Network since our main contribution is to include the question into the memorization process.

Then we proposed a novel model called ThRee Embedding Recurrent Neural Network which has been used for the Community Question Answering task. In this case, we tried to embed information of a Dependency Parser in order to create an enhanced representation of inputs sentences.

All of our models have been validated by using both synthetic and real datasets. For the best of our knowledge, we achieved a new state-of-the-art in the Reasoning Question Answering task, and very promising results in Reading Comprehension and Community Question Answering ones.
File