Research On Knowledge Base Question Answer Based On Meta Learning

With the continuous development of Internet technology and the gradual improvement of information extraction and knowledge processing technologies,many large-scale open source knowledge bases have appeared on the Internet,which provide rich knowledge resources for automatic question answering systems,which greatly promotes development of knowledge base questions and answers progress.At the same time,users have higher and higher requirements for search time and quality of data searched by search engines.Traditional search engine technology is now difficult to meet the increasing query needs of users,so how to build a solution that can accurately and efficiently answer user questions The automatic question answering system has become an important research project in the industry.Due to the vigorous development of open source knowledge bases,knowledge base-based question answering systems have shown strong advantages in both Chinese and English question and answer fields.Unlike traditional search engines,knowledge base-based question answering systems do not return a series of web page links,but to push users the precise answers that are queried in the knowledge base,greatly improve the user experience of the search engine,and bring economic benefits to the enterprise.Therefore,the knowledge base question answering system has received a lot of attention,attracting more and more scientific researchers to invest in related research work.However,there are also many challenges in the question and answer process of the knowledge base.First of all,the ability of humans to generate data and the demand for rapid acquisition of accurate data continue to increase.However,accurate and sufficient training data in the question and answer of the knowledge base are often not easy to obtain.Secondly,how does the machine use the small amount of data currently collected to quickly identify the target entity in the natural language question.Finally,how to predict the question sentence as a certain semantic relationship classification in the knowledge base based on a small amount of accurate annotation data.The deep learning method has achieved great success in the field of natural language in the field of mathematics,and has been integrated into all stages of automatic question and answer.This article uses deep learning-related technologies to alleviate the above-mentioned challenges in automatic question answering systems.In response to the above challenges,a neural network model based on meta-learning is proposed in the relationship prediction stage.The meta-model parameters related to the classification task are first trained on a data space that does not intersect the data set,and then migrated to WebQuestions that predict a small sample size.Then fine-tune the model parameters on the model of the data set.In order to quickly match the subject entity in the question with the entity in the knowledge base and improve the accuracy of entity linking,this paper uses a calculation method that combines literal similarity and semantic similarity to sort the similarity to construct an inverted index,and then establish a path index based on the knowledge base to search for answers.The proposed method is analyzed and verified on the popular WebQuestions data set with a small sample size,and the accuracy of the question answering system on the test set has been improved.