Research And Implementation Of Deep Learning-Based Automatic Question-and-Answer System For Mental And Physical Health

Automated question answering system has been widely concerned in the field of natural language processing,and has been fully applied in many professional fields.With the increase of information entities and entity relations in the open knowledge graph,it provides complete data support for the development of question answering system based on knowledge graph.However,for the knowledge map of specific professional fields,the open knowledge map usually lacks the professional knowledge and information of specific professional fields,which has certain limitations.In the past research,researchers have introduced a lot of external knowledge and information into the question and answer algorithm.The knowledge information used in these models is mostly unstructured text data,but there are still many problems.First,they often rely heavily on the quality of the unstructured textual data.Second,they usually look at the ternary components of the knowledge map from a global perspective and lack a holistic approach.Finally,they tend to ignore the meaning of entity relationship in the question answering system.In this paper,the processes of medical knowledge extraction,medical knowledge fusion and medical knowledge storage are firstly realized,in which the bi-lstm-crf network is used for knowledge extraction,and a variety of data preprocessing methods such as entity alignment are used to construct the mental and physical health knowledge map based on the medical data collected from different sources.In order to improve the robustness of the model and the content of the answer,the common knowledge map and common knowledge data set are used as the supplement of the data and corpus.Later,we proposed a global conversation model based on the mental and physical health knowledge map,namely the GKCM(global knowledge conversational model)model,which embedded the medical knowledge triad through multiple multi-attention mechanism,feature extraction of 1v1 convolution kernel and residual connection.Model can not only obtain specific triples and its adjoining triples entities and entity relationship,also by way of iterative method was developed to obtain knowledge global knowledge of key information,theoretically allows the model to obtain the global perspective of medical knowledge system,in order to better to hold the semantic information of user problems and promote contact judgment model reasoning ability.On the other hand,it also strengthens the ability of the question and answer model to explore the path of entity and relation when making inferences.Finally,we built an automatic question-and-answer system based on GKCM network model.The development language is python,the frontend is built with Bootstrap framework and jQuery,the back-end is built with Flask framework,the database is built with Neo4 j graphical database,and the data interaction between the front and back ends is realized through Ajax,finally realizing the functions of question-and-answer interaction between users and the system.