Relationship Discovery And Hotspot Analysis On Diabetes And Obesity Using Representation Model

With the improvement of people's living standard and the unhealthy life style,obesity has been a prevalent disease.Because of the correlation on gene,diabetes mellitus has a close relationship with obesity,these two diseases not only cause damage to human's health,but also place a huge economic burden on society.They have become one of the most serious public health challenges in the world.To help researchers quickly and effectively reveal the close relationships between diabetes mellitus,obesity and other diseases from a large amount of literature and help to search more effective treatments for these diseases,we propose a novel model based on latent Dirichlet allocation topic model and representative model,called diabetes mellitus and obesity relationship discovery model based on representative learning.We downloaded more than 337,000 pieces of diabetes mellitus and obesity related literature published from 2007 to 2016 as a corpus,and applied our model to discover the relationships between diabetes mellitus,obesity and other diseases.To analyze the results of our experiments and clearly show the relationships between diseases,we used two visualization tools,namely word cloud analysis figure and segment chord diagram.From the literature collection of recent 10 years,we got the experiment results as following: 26 diseases have relationships with diabetes mellitus,such as depression,Alzheimer's disease,hypertension,retinopathy,tumor and so on.17 diseases have relationships with obesity,such as asthma,gastric disease,heart disease and so on.There are 15 diseases having relationships with both of diabetes mellitus and obesity,such as depression,anxiety,tuberculosis,hypothalamic disease,adrenal disease,respiratory disease,lung disease,hepatitis,liver diseases,schizophrenia,heart disease,OSAS,myocardial infarction,cardiovascular disease and hypertension.To show the credibility of our discoveries and estimate our model,we found the proofs from clinical reports and research literature which are not included in the training data.Fortunately,all the relationships between diseases are verified.Finally,we analyzed the research hotspots in the last 10 years with the help of visualization tools,and predict that tnf(tumor necrosis factor),tumor,hypertension,inflammation,adolescent obesity/diabetes and cell biology pathogenetic mechanism may be the hot research topics related to diabetes and obesity in the near future according to latent trend and current research status.We believe that our experiment results can help biomedical researchers better focus their attention and adjust the direction of their work.