Abstract
The purpose of automated question answering is to let the machine understand natural language questions and give accurate answers in the form of natural language. This technology requires the machine to store a large amount of background knowledge. In recent years, the rapid development of knowledge graph has made the knowledge based question answering (KBQA) more and more popular. Traditional styles of KBQA methods mainly include semantic parsing, information extraction and vector modeling. With the development of deep learning, KBQA with deep learning has gradually become the mainstream method. This paper introduces the application of deep learning in KBQA mainly from the following aspects: the development history of KBQA, KBQA methods using deep learning, common datasets used in KBQA, the comparison of various methods and the future trend.
Highlights
The question answering system refers to the technology that allows a machine to process the input questions from users and give corresponding answers
This paper introduces the application of deep learning in knowledge based question answering (KBQA) mainly from the following aspects: the development history of KBQA, KBQA methods using deep learning, common datasets used in KBQA, the comparison of various methods and the future trend
We review the history of the KBQA system, especially after deep learning has been added
Summary
The question answering system refers to the technology that allows a machine to process the input questions from users and give corresponding answers. The algorithm will train a model based on these data to make the mapping vector of the question and the corresponding answer as close as possible in the low-dimensional space. Methods of KBQA that use deep learning can be divided into two main paradigms: semantic parsing with deep learning and information extraction with deep learning The former mainly add various neural network models to the traditional semantic parsing framework to improve specific components, such as feature extraction, relationship recognition and similarity calculation. The latter mainly uses deep neural networks when mapping questions and answers to low-dimensional spaces and calculating their similarity.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have