| The development of the Internet and big data technology has modified medical treatment approaches.This is embodied in the continuous emergence of online medical services with unique features.Then,due to the trans-time-and-space service advantages of the online medical platforms,clinician-patient communication and sharing of medical information can effectively reduce the medical treatment cost and resolve the difficulty of receiving high-quality medical service caused by the unequal distribution of medical resources.Meanwhile,the rapid development of online medical services provides development opportunities to the medical and health insurance industry,which is now in the transformation stage of development mode.Online medical services employ business intelligent technology to change the service mode,risk control,and revenue channels of traditional medical insurance and promote the formation of a new health service chain of "Internet+medical treatment+ insurance".At present,the complete closed-loop new health service chain has compensated for the deficiencies of many commercial health insurances at the income end and expenditure end,partially solving the intractable claim settlement problem,making a breakthrough in the scale of the medical insurance market by reducing premiums,and providing differentiated services and attracting online medical users.However,it will encounter significant challenges in future development.For example,it is not evident how commercial cancer insurance can achieve sustainable development under this service chain.The reason is that the disease risks are too roughly and unfairly graded by insurance companies,making it challenging to embody the difference in disease risk between different online medical users.The users’ insurance buying is related to the risk degree of disease and users’ preferences.The insurance companies pay no attention to the preferences and attitudes of online medical users.Accordingly,the insurance companies have not performed in-depth communication with online medical platforms or obtained users’ disease risk through the data on the platforms.As a result,they cannot determine the cancer insurance rate compatible with the disease risk and cannot realize the pricing based on each specific situation.Besides,they cannot obtain users’ preferences for cancer insurance from the data on online medical platforms and cannot realize the matching of personalized cancer insurance,considering personalized disease risk acquisition and precise pricing.Therefore,from the perspective of risk and pricing,this dissertation investigates how to integrate users’ disease risk on online medical platforms with the pricing of cancer insurance.From the perspective of preference and matching,this dissertation explores how to match the cancer insurance under risk pricing with user preferred cancer insurance on online medical platforms.Firstly,the records of online medical entities formed by online medical data are utilized to predict users’ disease risk and complete the precise pricing based on risk.Secondly,the users’ preference for cancer risk perception is obtained to ensure users’ preference for the perception of cancer insurance and complete the matching of cancer insurance based on pricing and preference.The main innovations of this work are given as follows:(1)The user’s disease risk is predicted.Firstly,the medical entity time-sequential characteristic graph(METCG)is formed by integrating the sequential relationship of online medical entity records(OMERs)with the graph’s expression form.The optimized TEApriori algorithm extracts the METCG based on the time sequence graph.According to the graph reconstruction theory,the portraits of online medical users are reconstructed through the METCG.The reconstruction coefficient is obtained to predict disease risk based on the METCG and derive the online medical users’ disease condition at the current stage.Secondly,the METCG-based collaborative neighborhood prediction method is proposed.The disease risk portrait of online medical users is obtained using the similarity between online medical users and online medical case archive users.Moreover,the future disease risk probability and disease time of online medical users are predicted through the typical disease relationship of similar users in the portrait.(2)The preference of users’ cancer risk perception is obtained.Firstly,the LDA and Twitter-LDA models are extended to propose a topic model PQDR-LDA based on the correlation between users’ questions and doctor answers to extract the disease themes from text data with sparse and dense medical entities,as well as vague and clear text semantics.Besides,a user preference access method based on contextual cognitive behavior is proposed to extract the user preferences from the mined online inquiry texts under the disease theme.In this way,more accurate users’ cognitive preferences are obtained from the context of multi-dimensional textual space and multi-dimensional disease space.Finally,the similarity relation between all online inquiry tests and early cancer texts is employed to derive the cancer risk perceived by users.The cancer risk preference perceived by users can be obtained more accurately by integrating the cognitive preferences of all online inquiry texts and the cancer risks perceived by users under various disease themes.(3)The perceived preference for cancer insurance is mined.Firstly,the whitening cosine similarity measure is adopted to design the weight for the cancer information between two domains.The weight is introduced into the MMD measure to establish a measuring method of maximum distribution weighted mean difference.Accordingly,a disease feature migration classification method of cancer information under joint distribution is proposed based on the joint distribution adjustment idea to obtain the tagged cancer risk samples.Secondly,a joint tensor decomposition method(CPFT-JTF),which is based on the migration of cross-domain preference features,is proposed to construct users’ preference features of cancer features in cross-domain cancer risk samples,thereby forming the ternary relationship of user-sample-feature,and organizing through the tensor.Furthermore,the cross-domain migration is realized by jointly decomposing the preference feature tensor of cancer features in the cancer risk samples of the two domains.Specifically,the perception and attention tensors in the preference features are jointly decomposed,and the score of users’ preference prediction is obtained in the cancer insurance domain.(4)The precise pricing based on risk is performed.Firstly,the Phase-Type distribution model is constructed and applied to cancer insurance to calculate the net premium.The construction of the Phase-Type distribution retains the easy processing characteristic of the exponential distribution and partially overcomes the limitation of no aftereffect.Thus,it can better match the actuarial scenario of cancer insurance.By increasing the number of states,the dependence of the conversion rate on age and duration among the three states is realized,and the calculating of the time-homogeneous Markov model maintains its simplicity to establish the actuarial model.Secondly,the disease risk probability of the insured is obtained from the prediction methods of disease risk,and the disease risk combination coefficient is constructed based on the disease risk probability to obtain the adjustment ratio of cancer insurance premium.Then,the pure insurance premium calculated by the cancer insurance actuarial model is optimized and adjusted under Phase-Type distribution,finally achieving the premise pricing of cancer insurance.(5)The precise matching based on preferences is realized.Firstly,a cancer insurance tagging system of multi-dimensional classification is established.A screening method for cancer insurance cases is proposed based on tag matching to screen the cancer insurance cases under users’ preference insurance.Through the construction of cancer theme tag generation of active learning,this dissertation designs the screening process of cancer insurance cases under the same user from three aspects,including classification of classifiers,case uncertainty selection strategy,and iteration termination strategy.Secondly,the cancer event model is established to accurately calculate the similarity between the initially-screened cancer insurance cases under the same user,and the method of extracting the elements of cancer events is proposed.Accordingly,combined with the knowledge in the cancer domain,a method for calculating the similarity between cancer prevention events under the same user is proposed.Moreover,cancer insurance cases’ similarity is measured,based on the similarity of cancer prevention events,thereby obtaining the precise matching of cancer insurance under the same user.66 figures,44 tables,302 references... |
PDF Full Text Request