Question understanding in question answering systems via deep learning

نوع: Type: thesis

مقطع: Segment: masters

عنوان: Title: Question understanding in question answering systems via deep learning

ارائه دهنده: Provider: Majid Kamrannejad

اساتید راهنما: Supervisors: Mir Hossein Dezfoulian (Ph.D)

اساتید مشاور: Advisory Professors: Muharram Mansoorizadeh (Ph.D)

اساتید ممتحن یا داور: Examining professors or referees: Mohammad Nassiri (Ph.D) - Mehdi Sakhaei nia (Ph.D)

زمان و تاریخ ارائه: Time and date of presentation: Wednesday, February 19, 2020, 5 PM

مکان ارائه: Place of presentation: Amphitheater Department of Computer Engineering

چکیده: Abstract: Question Answering is one branches of information retrieval and natural language processing. Generally, the Q&A system is a computer program that can extract answers from the natural language documents. One area of question answering is machine reading comprehension. Machine reading comprehension is the ability to read text by machine and answer questions posed by the text. So, the system must, like humans, be able to understand the words of the text and the relationships between them correctly. Given the newness of this field, most of the proposed approaches are based on deep learning. The proposed method is also based on deep learning. Initially, the data is preprocessed and its contexts and questions are tokenize into words and characters. Then, words are embedded with pre-trained Glove and characters embedded with random initialization. Then encoding characters with Bi-GRU network and context words and questions corresponding to concatted. Using the Bi-GRU network, to encoding the context and question. Then context-to-question attention is obtained by the attention mechanism. The output of the attention mechanism is encoded by the residual learning network that this network has the layer of self-attention. The network output is used to estimate the response and the predicted response is evaluated. In this study, the SQuAD dataset has been used and tried to present a method that offers acceptable performance, accuracy with increased speed. The results indicated that the proposed method achieved the proper development of speed and accuracy than previous models.

فایل: ّFile: Download فایل