| In the era of big data,the discovery of medical literature knowledge has always been the focus of researchers.With the emergence of massive medical data,a large amount of biomedical information,such as diseases,genes and drugs,is presented in the tumor research literature in an unstructured way.The complex diversity of data has proposed new research challenge for knowledge discovery.At present,there is little research on the fusion of textual words embedding representation technology and topic level knowledge to explore the medical literature.Therefore,the study proposed a knowledge discovery model of tumor literature based on word embedding representation.The tumor literature is used as the research object,and the word embedding representation model(Topic Word Embedding,TWE)is used to perform word embedding representation of word vector and topic vector.Based on the neural network(Siamese Network),combined with text and domain knowledge topics for similarity calculation,the cluster knowledge analysis is used to realize the knowledge discovery of tumor literature.On the one hand,the study explored the semantic information representation of medical literature at the topic level through the keyword embedding model,and the effect of the topic embedding representation method in the knowledge discovery of medical literature topics.The TWE word embedding representation method combined with the topic model considers the context information of the words.Meanwhile,the method considers the topic information corresponding to the words,making the semantic information richer and the words and topics well-learned.The results showed that compared with the traditional topic model-based representation method,the TWE word embedding representation method has a contour coefficient of 0.58 in topic clustering,which can better find the potential associations between topics.On the other hand,the researches showed that unstructured medical information can effectively help to realize the knowledge discovery in the medical field.Considering the method of calculating the semantic similarity of topics based on subject words,the study explored the effectiveness of the deep learning method in the discovery of subject knowledge,combing the words embedding representation and domain knowledge topic semantics.In this study,the TWE keyword embedding model is used to represent the text and the topic vector.The TWE keyword embedding representation is combined with the domain knowledge topic as input.The twin neural network is used to train the data and adjust the parameters,F value is0.94,which is higher than other text representation methods.Finally,through comparative analysis and cluster analysis,the diseases and related genes are analyzed from the perspectives of ultra,high,medium and low frequencies,and it is found that the current hot research and the target genes that may become research hotspots in the future.It proves the effectiveness of the proposed deep learning method of the combination of word embedding representation and domain knowledge topic semantics.Finally,in the aspect of application,based on the above proposed model and empirical research results,this study builds a multi-source heterogeneous data topic knowledge discovery platform based,employing using tumor journal literature and scientific report literature as data sources. |
