Research On Knowledge Behavior Of Online Health Community Users Based On Knowledge Extraction

With the rapid development of online health communities(OHCs),more and more users use Q&A OHCs(such as: Patientlikeme,Medhelp,Asking and Answer,Seek Medical Advice,etc.)to conduct Q&A counsultation with experts or peers for health assistance to better understand their condition.The massive Q&A data(information on diseases,symptoms,tests,drugs,and drug effects)accumulated on OHC provides conditions for studing knowledge extraction based on massive data,and also lays the foundation for analyzing the life cycle of disease medication health management,at the same time,knowledge extraction can also lay the foundation for behavioral research to better extract correlation feature variables.OHC plays an important role in alleviating the excessive strain and uneven distribution of medical resources,alleviating user anxiety and improving user trust.Generally,after a user asks a question,he/she faces the problem of choosing the most satisfying reply across multiple replies.At the same time,how to identify useful information from the large amount of Q&A data is of great significance for community knowledge management,search engine within the community of recommending the useful information.This article summarizes the literature on OHCs,information extraction and knowledge construction,dual-process theory and information adoption,and information usefulness,and summarizes the shortcomings of previous studies and future work: 1)The relationship extraction in the biomedical field in previous studies mainly includes: Research on chemical-induced diseases and drug side effects is based on the corpus of electronic medical records,discharge abstracts,or medical literature abstracts.Words and sentences in these relationship extraction corpuses are relatively structured texts.However,few scholars have studied how to extract the relationship of diseases diagnosis and disease medication management from these word colloquialism and less structured texts in OHC.and there is almost no research on the relationship extraction based on deep learning on Q&A corpous of OHC;2)The previous research of health management of the life cycle was mostly based on contracted personal doctors or electronic medical records of residents,with the phenomenon of the more users in OHC,and the more users' data has been accumulated over time.The massive users' data provides a possibility of analyzing health management based on life cycle in disease medication on OHC;the existing knowledge bases are rarely constructed based on OHC,therefore,based on the results of relationship extraction and the knowledge of disease encyclopedia,a knowledge map based on OHC can be constructed to complement existing knowledge bases,and at the same time,it can be better extract feature variables for behavior research in future;3)Current user adoption research in Q&A communities mainly focuses on open communities such as Wikipedia,Baidu knows,and online knowledge communities.However,there is little or no research on how users adopt satisfactory health information from physicians' online replies.4)Existing researches on identifying the usefulness information are mostly based on intuition to find the usefulness characteristics,and there is little research of designing feature variables from the theory of information systems,and even more,proposing a complete theoretical framework to solve the research problems of identifying the information usefulness.Conducting the relation extraction from massive OHC texts,as well as studing the user's knowledge behavior,have become important research directions and are of great significance.In general,our research completes the work of the following four aspects:(1)Aiming at the huge amount of doctor-patient Q&A data,we study the relationship extraction among diseases,symptoms and tests in OHC.By training word embedding vectors in the medical health field,we use the Bi-LSTM + CRF technology for entity recognition of disease,symptom,and tests of Q&A data,and construct a bidirectional gate recurrent neural network(2ATT-Bi GRU)relationship extraction(classification)model based on based on character-level and sentence-level attention mechanism,which is used to extract the relationship among diseases,symptoms and tests.(2)Based on the massive amount of doctor-patient Q&A data,we perform entity recognition among diseases,drugs,and drug effects,and extract the relationship among diseases,drugs,and drug effects based on the result of entity recognition.Due to the importance of the life cycle health management of disease medications for disease control and prevention,based on the result of the relationship extraction among disease,drugs and drug effects,we use the time series data of disease medications,from 1927 users who asked questions more than 5 times in the Q&A data,and use the results of disease medication relationship extraction to study the evolution of disease-drug use in accordance with the time series of user questions.The results can assist the disease medication health management based on users' life cycle.Based on the result of relationship extraction,we study the knowledge map construction technology,and build a knowledge map framework based on OHC,by carrying out the relationship extraction among diseases,symptoms,tests and drugs in the disease encyclopedia,and integrating the previous result of the triple relationship extraction about disease diagnosis and disease medication management,a knowledge map is constructed based on the data from OHC,which can complement and improve the existing knowledge base.Knowledge construction can better lay the foundation for extracting feature variables for user behavior research in future,making user behavior research more accurate and scientific.(3)In order to improving user satisfaction and enhancing trust in doctor-patient OHC,we study the knowledge behaviors influencing factors of users adopting a most satisfactory doctor reply in OHC.Based on the dual-process theory of knowledge adoption,we develop a conceptual model,using text analysis technology,extract variables from the two aspects of the argument quality and the source credibility,finally,we analyze the empirical results about what factors affect the user's knowledge adoption behavior,and which type of doctors' replies are most satisfactory for users.(4)Although there is a wealth of information in OHC,it is difficult for users to directly identify the most useful information from complex and massive data.As each online health platform is looking for a mechanism to help users find relevant and useful information to meet user needs,our research begins with the users' adoption and likes behaviors in Q&A OHC to study the information usefulness of doctors replies.Starting from the thinking model of designing science,taking the theory of knowledge adoption as the research's core theory,we propose meta-requirements from the central and peripheral routes,conduct meta-design,and propose the design hypothesis.Four machine learning methods are used to identify the information usefulness of the doctors' replies in OHC,compared with the current popular technology of deep learning and the classic research models of previous research,the final results confirm the advantages of our research framework.Compared with previous studies,the innovation of our research is reflected in:(1)Comprehensive use of various methods and new ideas for data analysis: we comprehensively use of various research methods such as text analysis,deep learning,knowledge mining,knowledge map construction,econometric analysis;and use the theory of information adoption theory(KAM),elaboration likelihood model(ELM),health management theory of disease prevention and care;and knowledge management theory.We analyze the participants in OHC from a behavioral perspective,study the decision-making behavior of users' adoption of doctors' replies in OHC from behavior change and knowledge acquisition,and the identify the information usefulness of doctors' replies based on users' adoption and likes behavior.(2)The existing disease-related relationship extraction is mostly based on electronic medical records and biomedical literature abstracts.The biggest advantage of our research is that it proposes a GRU network architecture combining characterlevel and sentence-level attention mechanisms(2ATT-Bi GRU)for doctor-patient Q&A corpus in OHC,to extract the multiple relationships related to disease;to train domainrelated word embedding vectors through large-scale data sets,and bidirectional GRU networks(It does not require manual design features,and obtain important contextual grammar and semantic features)and the structure of attention mechanism make our model surpasses the existing classic models in the relationship extraction of diseaserelated,and shows excellent results.After using the above relationship extraction among diseases,drugs,and drug effects,we propose to analyze the time series data of disease medications for users in OHC,and select time series data of the cardiovascular disease users who have asked questions more than 5 times,combine with user portraits(age,gender,disease characteristics,etc.),analyze the user's life cycle of medication health management,and obtain the information about the progress and evolution of disease medications and drug effects from the analysis results.The research can assist health managers to develop personalized health management solutions for users,in addition,many user medication cases can be shared with doctors to assist doctors in diagnosis and provide support for evidence-based medicine,expanding the scope of data application based on life cycle health management.(3)In the study of the users' knowledge adoption behavior in OHC,we use the large-scale recent corpus of the most popular doctor-patient OHC;use the dual-process theory to extend the user's knowledge behavior research to the Q&A OHC(previous research is all in the open community);and we use text mining technology to comprehensively extract the variables of the argument quality and the source credibility from the doctor-patient data;at the same time,we investigate the moderating variable of users involvement in the association between the factors of argument quality or source credibility on user adoption decisions;based on empirical results,our research provides a better understanding of users' knowledge adoption behavior in OHC using large-scale data sets,and deeply analyses the implications for platforms,doctors and users.(4)Starting from the thinking model of designing science,and we take the theory of knowledge adoption as the core theory,and propose a research framework for identifying the information usefulness of doctors' replies.Our study proposes a design process based on a knowledge adoption conceptual model,specifies meta-requirements for information usefulness,and builds a meta-design based on a conceptual model.The use of core theories from the social sciences provides theoretical support for the design process.The key step is to apply the core theories to design science.In particular,how to use conceptual models for meta-design is another challenge.Our research shows the calculation dimensional characteristics from the application of the knowledge adoption model to the information usefulness meta-design.They represent the information usefulness from the knowledge adoption theory.Finally,it shows how the usefulness framework proposed in our research surpasses previous intuition-based research.In addition,our research expands the theoretically driven design research of information systems.