Research On Prediction And Prediction Of Health Information Adoption Based On Integrated Learning

In 2021,the number of Internet medical users will be 710 million,which increases 7.25%comparing to last year.By the influence of the epidemic,the influence of online communities is gradually increasing because the online health consultation is highly convenient and low-cost.Online communities such as ‘xydy.com’ platform and ‘120ask.com’ platform have attracted a large number of consultation users and become the important platform for public to search health information.On the other hand,the quality of health information in various online communities is untrustworthy,and the distorted and inferior information seriously affects users’ information adoption,which even causes users’ health anxiety.Based on the massive data of the adoption behavior in online healthy communities,it is important to explore the factors that influence users’ adoption of health information.The research will improve the efficiency of communication between doctors and patients according to users’ health information demand and improve the health literacy of users.At present,it is interesting and important to extract value from massive online data.However,the existing health information adoption research mainly adopts questionnaires,interviews and experiments to analyze the influence of information characteristics on users’ willingness to adopt health information.Seldom research has explored the information value from massive data of online healthy communities.According to the research results of the factors of health information adoption intention and the information adoption in general Q & A communities,we crawl Q& A data from online healthy communities.We extract the influencing factors of user information adoption behavior and build a model based on the idea of ensemble learning to improve the accuracy of information adoption prediction.First,we design a crawler program to obtain 12,149Q&A data and 1,045,796 historical text data from ‘xywy.com’ platform.We construct an index system including text structure,online social communication record,professional authority.Then we select machine learning methods such as RF and Xgboost to evaluate the influence of each factor through feature engineering,compare the influence of different factors combinations through classified prediction and extract the factors combinations with the highest prediction accuracy.Further more we analyze the evolution track of each factor’s influence.Secondly,based on ensemble learning(Xgboost,lightgbm,random forest,etc.)and non-ensemble learning(support vector machines,decision trees,MLP,etc.),we construct a stacking model.Then we compare the prediction accuracy between single prediction model and stacking ensemble strategy with different model combination with ROC curve,accuracy,F1 and Recall.The results show that the Stacking integrated model can effectively improve the accuracy of classification prediction.Further,the portability of the platform is verified by the data from ‘120ask.com’platform.Thirdly,based on the results of the factors affecting the information adoption behavior in online healthy communities,we suggest that the medical staff in the healthy community should choose simple and popular language to improve the efficiency of communication with patients.At the same time it is important for the medical staff to take part in more interaction with users to improve their influence.And the online communities can use the integrated stacking model to improve the predict accuracy of doctor’s question and answer adoption to improve the efficiency of accurate push to inquiring users.