CN112992367B

CN112992367B - Smart medical interaction method based on big data and smart medical cloud computing system

Info

Publication number: CN112992367B
Application number: CN202110309022.6A
Authority: CN
Inventors: 崔剑虹
Original assignee: Micro Pulse Technology Co Ltd
Current assignee: Micro Pulse Technology Co., Ltd
Priority date: 2021-03-23
Filing date: 2021-03-23
Publication date: 2021-09-28
Anticipated expiration: 2041-03-23
Also published as: CN112992367A; CN113506639A; CN113506638A

Abstract

The embodiment of the disclosure provides a big data-based intelligent medical interaction method and an intelligent medical cloud computing system, which are used for determining the correlation degree of each medical service knowledge point for each problem to be responded based on the knowledge point code distribution of each medical service knowledge point and the problem code distribution of each problem to be responded, acquiring the knowledge point heat corresponding to each medical service knowledge point, calculating the referability degree of each medical service knowledge point for the problem set to be responded based on the correlation degree and the knowledge point heat, and selecting a target knowledge point from the medical service knowledge points to be associated to the problem set to be responded based on the referability degree. Therefore, the method and the device can recommend the appropriate knowledge points to the problem set to be responded, simultaneously can consider the reference value of the problem set to be responded, and further can select the more matched knowledge points to perform information consultation interaction on the problem set to be responded, so that the accuracy of recommending the knowledge points is improved, and the knowledge rate of the knowledge points can be improved.

Description

Smart medical interaction method based on big data and smart medical cloud computing system

Technical Field

The disclosure relates to the technical field of smart medical treatment based on big data technology, and exemplarily relates to a smart medical treatment interaction method based on big data and a smart medical treatment cloud computing system.

Background

The smart medical service is an important component of a smart city, and is a medical system which comprehensively applies technologies such as a medical internet of things, data fusion transmission and exchange, cloud computing, a metropolitan area network and the like, fuses medical infrastructure and IT infrastructure through an information technology, takes a medical cloud data center as a core, spans space-time limitation of an original medical system, and carries out intelligent decision on the basis, so that the optimization of medical service is realized.

The intelligent medical treatment integrates an individual, an apparatus and a mechanism into a whole, closely associates a patient, a medical staff, an insurance company, a researcher and the like, realizes business cooperation, and increases the triple benefits of society, mechanism and individuals. Meanwhile, medical services such as remote registration, online consultation and online payment are pushed to hands of each person through technologies such as mobile communication and mobile internet, and the problem of difficulty in seeing a doctor is solved.

Based on this, in the related art, for different intelligent medical terminals (for example, terminals disposed in medical service institutions) disposed all over, for the online consultation process, multiple times of interaction with the user are required, and medical response information of the medical service can be accurately pushed for the user according to the latest medical service entry, so that the user can continue to perform consultation navigation according to the required information. However, in the related art, the knowledge rate of the accuracy of knowledge point recommendation is low.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, an object of the present disclosure is to provide a smart medical interaction method and a smart medical cloud computing system based on big data.

In a first aspect, the present disclosure provides a smart medical interaction method based on big data, which is applied to a smart medical cloud computing system, where the smart medical cloud computing system is in communication connection with a plurality of smart medical mobile terminals, and the method includes:

acquiring medical response information of a target medical service initiated by the intelligent medical mobile terminal, and acquiring an online consultation theme initiated by the intelligent medical mobile terminal aiming at any medical response entry in the medical response information;

acquiring medical service knowledge points related to the online consultation subject, and determining the correlation degree of each medical service knowledge point for each problem to be responded based on the knowledge point code distribution of each medical service knowledge point and the problem code distribution of each problem to be responded in the problem set to be responded;

acquiring the heat of the knowledge points corresponding to the medical service knowledge points, and calculating the referenceable degree of the medical service knowledge points for the problem set to be responded based on the correlation degree and the heat of the knowledge points;

and selecting a target knowledge point from the medical service knowledge points to be associated to the set of questions to be answered based on the referential degree so as to interact with an online consultation subject initiated by the intelligent medical mobile terminal.

In a second aspect, the disclosed embodiment further provides a big data-based smart medical interaction system, which includes a smart medical cloud computing system and a plurality of smart medical mobile terminals in communication connection with the smart medical cloud computing system;

the smart medical cloud computing system is used for:

According to any one of the aspects, in the embodiments provided by the present disclosure, based on the knowledge point code distribution of each medical service knowledge point and the problem code distribution of each problem to be responded in the problem set to be responded, the degree of correlation of each medical service knowledge point for each problem to be responded is determined, the heat of the knowledge point corresponding to each medical service knowledge point is obtained, based on the degree of correlation and the heat of the knowledge point, the referability of each medical service knowledge point for the problem set to be responded is calculated, and based on the referability, a target knowledge point is selected from the medical service knowledge points and associated to the problem set to be responded. Therefore, the method and the device can recommend the appropriate knowledge points to the problem set to be responded, simultaneously can consider the reference value of the problem set to be responded, and further can select the more matched knowledge points to perform information consultation interaction on the problem set to be responded, so that the accuracy of recommending the knowledge points is improved, and the knowledge rate of the knowledge points can be improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings that need to be called in the embodiments are briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present disclosure, and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.

Fig. 1 is a schematic view of an application scenario of a big data-based intelligent medical interaction system according to an embodiment of the present disclosure;

FIG. 2 is a schematic flowchart illustrating a big data-based intelligent medical interaction method according to an embodiment of the present disclosure;

FIG. 3 is a functional block diagram of a big data based intelligent medical interactive device according to an embodiment of the present disclosure;

fig. 4 is a schematic block diagram of a smart medical cloud computing system for implementing the big data-based smart medical interaction method according to the embodiment of the present disclosure.

Detailed Description

The present disclosure is described in detail below with reference to the drawings, and the specific operation methods in the method embodiments can also be applied to the device embodiments or the system embodiments.

Fig. 1 is an explanatory diagram of a big data-based intelligent medical interactive system 10 provided by an embodiment of the present disclosure. The big data based smart medical interaction system 10 may include a smart medical cloud computing system 100 and a smart medical mobile terminal 200 communicatively connected to the smart medical cloud computing system 100. The big data based intelligent medical interactive system 10 shown in fig. 1 is only one possible example, and in other possible embodiments, the big data based intelligent medical interactive system 10 may also include only at least some of the components shown in fig. 1 or may also include other components.

In a possible design approach, the smart medical cloud computing system 100 and the smart medical mobile terminal 200 in the big-data based smart medical interactive system 10 may cooperatively perform the big-data based smart medical interactive method described in the following method embodiments, and the detailed description of the method embodiments may be referred to in the following steps of the smart medical cloud computing system 100 and the smart medical mobile terminal 200.

In order to solve the technical problem in the foregoing background, fig. 2 is a flowchart illustrating a big data-based smart medical interaction method according to an embodiment of the present disclosure, which may be executed by the smart medical cloud computing system 100 shown in fig. 1, and the big data-based smart medical interaction method is described in detail below.

Step S110, acquiring medical response information of the target medical service initiated by the smart medical mobile terminal 200, and acquiring an online consultation topic initiated by the smart medical mobile terminal 200 for any one medical response entry in the medical response information.

In one embodiment, the medical response information may refer to entry information associated with the target medical service, and the smart medical mobile terminal 200 may initiate an online consultation topic for any one of the medical response entries in the medical response information.

Step S120, acquiring medical service knowledge points related to the online consultation subject, and determining the correlation degree of each medical service knowledge point for each to-be-responded question based on the knowledge point code distribution of each medical service knowledge point and the question code distribution of each to-be-responded question in the to-be-responded question set.

In one embodiment, the medical service knowledge points related to the online consultation topic may be set according to actual design requirements, for example, if the online consultation topic is a diabetes consultation topic, the medical service knowledge points may include the medical service knowledge points related to the diabetes consultation topic, such as diabetes cause, diabetes caution items, and the like.

In one embodiment, the knowledge point code distribution may include a knowledge point range, a knowledge point type, a knowledge point applicable condition, a knowledge point coverage, a knowledge point query number, a knowledge point response number, and the like, and the dimensionality of the knowledge point code distribution may be selected according to actual design requirements, so that the acquired knowledge point code distribution is more comprehensive.

In one embodiment, each healthcare knowledge point may have a plurality of knowledge point code distributions, which may include: a second knowledge point code distribution that is continuously updated with the service location update, and a third knowledge point code distribution that is not continuously updated with the service location update. The second knowledge point code distribution may include knowledge point coverage, number of knowledge point queries, number of knowledge point responses, and the like. The third knowledge point coding distribution may include knowledge point ranges, knowledge point types, knowledge point applicable conditions, and the like.

In an embodiment, the set of questions to be answered may include one or more questions to be answered, one set of questions to be answered may correspond to one advisory services page, and each question to be answered in the set of questions to be answered may generate an advisory message in the advisory services page corresponding to the set of questions to be answered.

In one embodiment, the question code distribution may include a to-answer question base code distribution and a to-answer question specific code distribution. The basic code distribution of the questions to be answered can comprise the categories, labels, grades and the like of the questions to be answered. The special code distribution of the questions to be answered can comprise the extension question directions of the questions to be answered and the like.

In one embodiment, each question to be answered may have a plurality of question-coding distributions, which may include: the problem code distribution comprises a first problem code distribution used for describing the problem response strategy to be responded and the service position updating relation, a second problem code distribution continuously updated along with the service position updating, and a third problem code distribution which is not continuously updated along with the service position updating.

In an embodiment, distribution association parameters between the problem code distribution of each to-be-responded problem and the knowledge point code distribution of each medical service knowledge point can be calculated, and the distribution association parameters are input and mapped through a Sigmoid function, so that the correlation degree of each medical service knowledge point for each to-be-responded problem is obtained. The Sigmoid function can map the distribution correlation parameters to zero to one, so as to obtain the correlation degree corresponding to the distribution correlation parameters, and the correlation degree is used as the correlation degree of each medical service knowledge point for each problem to be responded.

In one embodiment, the distribution correlation parameter may be, but is not limited to, an Euclidean distance, a Manhattan distance, a Chebyshev distance, a Minkowski distance, a normalized Euclidean distance, a Mahalanobis distance, an included cosine, a Hamming distance, a Jacard distance, a correlation distance, or an information entropy, etc.

In one embodiment, before calculating the distribution association parameters between the question code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point, the question code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point may be mapped to the same code distribution grid, so as to more conveniently calculate the degree of correlation between the question code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point.

In an embodiment, a problem code distribution range corresponding to problem code distribution of each problem to be answered and a knowledge point code distribution range corresponding to knowledge point code distribution of each medical service knowledge point may be obtained, the problem code distribution of each problem to be answered is input into a kernel function whose kernel function processing range is the problem code distribution range for data conversion, and the knowledge point code distribution of each medical service knowledge point is input into a kernel function whose kernel function processing range is the knowledge point code distribution range for data conversion, so that the problem code distribution of each problem to be answered and the knowledge point code distribution of each medical service knowledge point are converted into the same code distribution grid.

In an embodiment, before mapping the problem code distribution of each to-be-answered problem and the knowledge point code distribution of each medical service knowledge point to the same code distribution grid, multiple problem code distributions of each to-be-answered problem can be fused, and multiple knowledge point code distributions of each medical service knowledge point can be fused, so that the correlation degree between the problem code distribution of each to-be-answered problem and the knowledge point code distribution of each medical service knowledge point can be calculated more conveniently.

In one embodiment, conditional random field calculations may be performed on every two first problem code distributions of each problem to be answered to obtain conditional random field features that can represent labeling or analysis sequence information in the first problem code distributions. The nonlinear feature fitting processing can be carried out on every two third problem code distributions of each problem to be responded to obtain the problem nonlinear feature fitting feature, and the problem nonlinear feature fitting feature can embody the relation between every two third problem code distributions of each problem to be responded to. The second problem code distribution, the conditional random field features and the problem nonlinear feature fitting features of all the problems to be responded can be fused, so that the obtained fusion information can reflect deeper information contained in the problem code distribution.

In an embodiment, the same kind of problem code distribution of each to-be-answered problem may be fused, and then fused with different kinds of problem code distributions of each to-be-answered problem, for example, the second problem code distribution, the conditional random field feature, and the problem nonlinear feature fitting feature of each to-be-answered problem may be fused, and then the fused second problem code distribution, the fused conditional random field feature, and the fused problem nonlinear feature fitting feature of each to-be-answered problem may be fused.

In an embodiment, nonlinear feature fitting processing may be performed on every two third knowledge point code distributions of each medical service knowledge point to obtain a knowledge point nonlinear feature fitting feature, and the knowledge point nonlinear feature fitting feature can embody a relationship between every two third knowledge point code distributions of each knowledge point. And fusing the second knowledge point code distribution of each medical service knowledge point and the nonlinear feature fitting feature of the knowledge point, so that the obtained fusion information can reflect deeper information contained in the knowledge point code distribution.

In one embodiment, the same kind of knowledge point code distribution of the medical service knowledge points may be fused and then fused with different kinds of knowledge point code distributions of the medical service knowledge points, for example, the second knowledge point code distribution, the conditional random field feature and the knowledge point nonlinear feature fitting feature of each medical service knowledge point may be fused, and then the second knowledge point code distribution, the fused conditional random field feature and the fused knowledge point nonlinear feature fitting feature of each medical service knowledge point may be fused.

Step S130, acquiring the heat of the knowledge points corresponding to the medical service knowledge points, and calculating the referenceable degree of the medical service knowledge points for the problem set to be responded based on the correlation degree and the heat of the knowledge points.

In an embodiment, the heat degree of the knowledge point corresponding to each medical service knowledge point is flexibly set according to the specific situation of the knowledge point, and is an initial reference degree provided by a user after each medical service knowledge point is received, and is used for representing the audience probability of the medical service knowledge point, for example, if the click rate or consultation rate of the medical service knowledge point is higher, the heat degree of the knowledge point is higher, and the setting rule of the heat degree of the specific knowledge point may refer to the heat degree rule in the prior art, which is not specifically limited here.

In an embodiment, a product between the heat of the knowledge point corresponding to each medical service knowledge point and the correlation degree of each medical service knowledge point for each problem to be responded can be calculated to obtain an initial reference degree of each medical service knowledge point for each problem to be responded, and the initial reference degree of each medical service knowledge point for each problem to be responded is weighted to obtain a referenceable degree of each medical service knowledge point for a set of problems to be responded.

In one embodiment, a weighted sum of the degree of correlation and the heat of the knowledge points may be calculated to obtain a referenceable degree of each healthcare knowledge point for the set of questions to be answered.

And step S140, selecting a target knowledge point from the medical service knowledge points to be associated with the problem set to be responded based on the referential degree so as to interact with an online consultation theme initiated by the intelligent medical mobile terminal.

In an embodiment, a set number of medical service knowledge points may be selected from the medical service knowledge points as target knowledge points in an order from a large referential degree to a small referential degree, and the target knowledge points are associated with a set of questions to be answered. In one embodiment, the candidate list may be generated in descending order of the referenceability, and a set number of the medical service knowledge points ranked first may be selected from the candidate list. In one embodiment, the number of target knowledge points may be multiple.

In one embodiment, a set number of medical service knowledge points may be sent to a client where a problem set administrator to be answered is located, and the problem set administrator to be answered selects a target knowledge point from the set number of medical service knowledge points and forwards the target knowledge point to the problem set to be answered. In one embodiment, the target knowledge point may be pushed to the set of questions to be answered by the platform system where the set of questions to be answered is located.

In the foregoing embodiment, based on the knowledge point code distribution of each medical service knowledge point and the problem code distribution of each to-be-responded problem in the to-be-responded problem set, the degree of correlation of each medical service knowledge point with respect to each to-be-responded problem is determined, the heat degree of the knowledge point corresponding to each medical service knowledge point is obtained, based on the degree of correlation and the heat degree of the knowledge point, the referrability degree of each medical service knowledge point with respect to the to-be-responded problem set is calculated, based on the referrability degree, the target knowledge point is selected from the medical service knowledge points and associated with the to-be-responded problem set, while the appropriate knowledge point is recommended to the to-be-responded problem set, the reference degree value of the to-be-responded problem set can be considered, and further the more matched knowledge point can be selected to perform information consultation and interaction with the to-be-responded problem set, so that not only the accuracy of recommendation of the knowledge points is improved, besides, the knowledge rate of the knowledge points can be improved, and in addition, the problem to be responded obtains a reference value in the knowledge points of the problem set to be responded, so that the matching degree of the problem to be responded in the problem set to be responded can be improved.

In an embodiment, in the foregoing step S120, determining a degree of relevance of each medical service knowledge point to each question to be answered in the set of questions to be answered based on the knowledge point code distribution of each medical service knowledge point and the question code distribution of each question to be answered, may include: and inputting the knowledge point code distribution of each medical service knowledge point and the problem code distribution of each problem to be responded in the problem set to be responded into the machine learning network, and acquiring the correlation degree of each medical service knowledge point output by the machine learning network aiming at each problem to be responded.

In one embodiment, the knowledge point code distribution of each medical service knowledge point can be input to a first machine learning unit of the machine learning network, and the question code distribution of each question to be answered is input to a second machine learning unit of the machine learning network, the first machine learning unit and the second machine learning unit are connected to a global learning unit, and the correlation degree of each medical service knowledge point output by the global learning unit for each question to be answered is obtained.

In one embodiment, the first machine learning unit may include: a first nonlinear feature fitting subunit and a first fusion subunit connected to the output of the first nonlinear feature fitting subunit. The first nonlinear feature fitting subunit is used for performing nonlinear feature fitting processing on third knowledge point coding distribution in the knowledge point coding distribution, and the first fusion subunit is used for fusing second knowledge point coding distribution in the knowledge point coding distribution and nonlinear feature fitting features of the knowledge points output by the first nonlinear feature fitting subunit.

In one embodiment, the second machine learning unit may include: the device comprises a second nonlinear feature fitting subunit, a conditional random field subunit and a second fusion subunit, wherein the input end of the second fusion subunit is connected with the output end of the second nonlinear feature fitting subunit and the output end of the conditional random field subunit. The second nonlinear feature fitting subunit is used for performing nonlinear feature fitting processing on third problem code distribution in the problem code distribution, the conditional random field subunit is used for performing conditional random field calculation on the first problem code distribution in the problem code distribution, and the second fusion subunit is used for fusing the second problem code distribution in the problem code distribution, the problem nonlinear feature fitting features output by the second nonlinear feature fitting subunit and the conditional random field features output by the conditional random field subunit.

In one embodiment, the global learning unit may calculate a degree of correlation between the knowledge point fusion feature output by the first fusion subunit and the question-to-answer fusion feature output by the second fusion subunit.

For example, assume that the machine learning network a may include a first machine learning unit a1, a second machine learning unit a2, and a global learning unit A3 connecting the first machine learning unit a1 and the second machine learning unit a 2.

In one embodiment, the first machine learning unit a1 may include a first discrete coding subunit a11, a first nonlinear characteristic fitting subunit a12 connected to the output of the first discrete coding subunit a11, a first continuous coding subunit a13, a first merging subunit a14 connected to the output of the first nonlinear characteristic fitting subunit a12 and the output of the first continuous coding subunit a13, a first spatial transformation module a15 connected to the output of the first merging subunit a14, and a first regularization module a16 connected to the output of the first spatial transformation module a 15. The first discrete encoding subunit a11 is configured to perform encoding processing on discrete knowledge point encoding distributions in knowledge point encoding distributions of the medical service knowledge points to obtain discrete knowledge point encoding distribution vectors; the first nonlinear feature fitting subunit a12 is configured to perform nonlinear feature fitting on every two discrete knowledge point coding distribution vectors in the discrete knowledge point coding distribution vectors of each medical service knowledge point; the first continuous coding subunit a13 is configured to perform coding processing on the second knowledge point coding distribution in the knowledge point coding distribution of each medical service knowledge point to obtain a second knowledge point coding distribution vector; the first fusion subunit a14 is configured to fuse the second knowledge point coding distribution vector with the knowledge point nonlinear feature fitting feature vector output by the first nonlinear feature fitting subunit a 12; the first spatial conversion module A15 converts the knowledge point fusion eigenvectors output by the first fusion subunit A14 into a target coding distribution grid; the first regularization module a16 is configured to regularize the knowledge point fusion feature vectors in the target coding distribution grid.

In one embodiment, the second machine learning unit a2 may include: the time-series characteristic coding subunit A21, a conditional random field subunit A22 connected to the output terminal of the time-series characteristic coding subunit A21, a second discrete coding subunit A23, a second nonlinear characteristic fitting subunit A24 connected to the output terminal of the second discrete coding subunit A23, a second continuous coding subunit A25, a second fusion subunit A26 connected to the output terminal of the conditional random field subunit A22, the output terminal of the second nonlinear characteristic fitting subunit A24 and the output terminal of the second continuous coding subunit A25, a second spatial transform module A27 connected to the output terminal of the second fusion characteristic module A26, and a second regularization module A28 connected to the output terminal of the second spatial transform module A27. The time sequence characteristic coding subunit a21 is configured to perform coding processing on a first problem code distribution in the problem code distributions of each to-be-responded problem to obtain a first problem code distribution vector; the conditional random field subunit a22 is configured to perform conditional random field calculation on every two first problem code distribution vectors in the first problem code distribution vectors of each to-be-answered problem to obtain conditional random field feature vectors; the second discrete encoding subunit a23 is configured to perform encoding processing on the discrete problem code distribution in the problem code distribution of each to-be-responded problem to obtain a discrete problem code distribution vector; the second nonlinear feature fitting subunit a24 is configured to perform nonlinear feature fitting on every two discrete problem code distribution vectors in the discrete problem code distribution vectors of each to-be-answered problem; the second continuous coding subunit a25 is configured to perform coding processing on a second problem code distribution in the problem code distributions of the problems to be answered to obtain a second problem code distribution vector; the second fusion subunit A26 is configured to fuse the second problem coding distribution vector, the conditional random field feature vector, and the problem nonlinear feature fitting feature vector output by the second nonlinear feature fitting subunit A24; the second space conversion module A27 converts the problem fusion eigenvectors to be answered output by the second fusion subunit into a target coding distribution grid; the second regularization module a28 is configured to perform regularization on the fusion feature vectors of the to-be-answered questions in the target coding distribution grid.

In one embodiment, the first spatial transform module a15 may include multiple fully connected layers that enable dimensionality reduction of the knowledge point fused feature vectors while transforming them to the target coding distribution grid. The second space transformation module a27 may include a plurality of fully connected layers, and may perform dimension reduction on the fusion feature vector of the to-be-answered question while transforming the fusion feature vector of the to-be-answered question to the target coding distribution grid, so that the global learning unit may calculate the degree of correlation between the fusion feature vector of the knowledge point after dimension reduction in the same coding distribution grid and the fusion feature vector of the to-be-answered question.

In an embodiment, the global learning unit a3 may be configured to calculate distribution association parameters between the problem fusion feature vectors to be answered in the target coding distribution grid after regularization of each problem to be answered and the knowledge point fusion feature vectors in the target coding distribution grid after regularization of each medical service knowledge point, and perform input mapping on the distribution association parameters through a Sigmoid function to obtain a degree of correlation of each medical service knowledge point for each problem to be answered.

In one embodiment, the structure of the machine learning network may be divided into three regions, an input region, a mapping region connected to the output of the input region, and an output region connected to the output of the mapping region. The input area may include a first discrete coding subunit a11, a first consecutive coding subunit a13, a timing feature coding subunit a21, a second discrete coding subunit a22, and a second consecutive coding subunit a 25. The mapping region may include a first nonlinear feature fitting subunit a12, a first fusion subunit a14, a first spatial conversion module a15, a first regularization module a16, a conditional random field subunit a22, a second nonlinear feature fitting subunit a24, a second fusion subunit a26, a second spatial conversion module a27, and a second regularization module a 28. The output region may include global learning unit a 3.

In one embodiment, the structure of the second non-linear feature fitting subunit may refer to the structure of the first non-linear feature fitting subunit.

In the embodiment, any two first problem coding distribution vectors in the first problem coding distribution vectors are directly connected through conditional random field calculation, and the conditional random field subunits are introduced for feature extraction of sequence information, so that the correlation between the two first problem coding distribution vectors can be captured, further, deeper characterization vectors can be learned, and the deep interdependent implicit information can be captured more easily.

For example, in one embodiment, the machine learning network may be pre-trained by: acquiring a problem code distribution sample sequence and a knowledge point code distribution sample sequence, and the correlation degree between each problem code distribution sample in the problem code distribution sample sequence and each knowledge point code distribution sample in the knowledge point code distribution sample sequence; inputting the problem code distribution samples into a first machine learning unit, inputting the knowledge point code distribution samples into a second machine learning unit, and obtaining classification confidence degrees between each problem code distribution sample in a problem code distribution sample sequence and each knowledge point code distribution sample in a knowledge point code distribution sample sequence output by a machine learning network; and comparing the classification confidence with the correlation degree, and if the classification confidence is inconsistent with the correlation degree, adjusting the machine learning network to ensure that the classification confidence is consistent with the correlation degree.

For example, in one embodiment, the problem code distribution samples and the knowledge point code distribution samples with the correlation degree reaching the target degree threshold value may be used as first samples, the problem code distribution samples and the knowledge point code distribution samples with the correlation degree not reaching the target degree threshold value may be used as second samples, and if the ratio of the number of the first samples to the number of the second samples does not reach the threshold value, the first samples are copied until the ratio of the number of the first samples to the number of the second samples reaches the threshold value.

For example, in an embodiment, if the ratio of the number of the first samples to the number of the second samples does not reach the threshold, a first distribution relation parameter between each first sample and any first sample may be calculated, and the first samples with the distribution relation parameters smaller than the first distribution relation parameter are randomly selected as the synthesized first samples; until the sum of the number of first samples and the number of synthesized first samples is compared to the number of second samples reaching a threshold.

In an embodiment, the specific implementation manner of acquiring the medical response information of the target medical service initiated by the intelligent medical mobile terminal in step S110 may be implemented through the following steps.

And step S111, constructing a medical service entry network.

In this embodiment, the medical service entry network includes elements and association attributes, the elements include medical service elements corresponding to the medical service and entry elements corresponding to the entries, the two elements having an interpretation relationship or an extended association attribute use the association attributes for association tagging, the interpretation relationship includes an attribute relationship between the medical service elements and the entry elements that have initiated an interpretation procedure, and the extended association attribute includes at least one of a jump relationship between the medical service elements and a topic hierarchy relationship between the entry elements.

For example, the smart medical cloud computing system may build a network of medical business terms based on medical business big data stored in a database. The medical service entry network is a large-scale network structure formed by medical service big data of a large number of medical services. Two types of elements can be included in the healthcare entry network: medical business elements and entry elements. For example, at least two medical service elements or at least two entry elements exist in the medical service entry network, for example, since the medical service entry network is constructed using a large amount of medical service data stored in the smart medical cloud computing system 100, a plurality of medical service elements and a plurality of entry elements exist in the medical service entry network. For example, the attribute relationship between the elements is represented by an association attribute, for example, when there is an interpretation relationship between the medical service and the entry, the medical service element and the entry element are labeled by the association attribute, and the association attribute represents the interpretation relationship between the two elements; for another example, when a jump relationship exists between the medical service and the medical service, two medical service elements are labeled by using an associated attribute, and the associated attribute represents the jump relationship between the two elements; for another example, for a medical service, different association degrees may be assigned to different terms associated with the medical service, and then an association degree may be assigned to an association attribute between a medical service element and a term element, where the association attribute indicates that there are an interpretation relationship between the two medical service elements and the association degree of the interpretation relationship. For example, in one embodiment, the association attribute may be a directed association attribute or an undirected association attribute. The directional association attribute indicates that the attribute relationship between the two elements may have directionality, for example, in the jump relationship between the medical service and the medical service, the medical service a refers to the medical service B, and the association attribute between the medical service element a of the medical service a and the medical service element B of the medical service B may be an active reference that points from the medical service a to the medical service B to emphasize the medical service a, or of course, the association attribute may also be a referenced relationship that points from the medical service B to the medical service a to emphasize the medical service B.

For example, assume that a medical service element a, a term element B, and an association attribute C between the elements are included in a medical service term network, and the association attribute C represents an attribute relationship between the elements. For example, one medical service occupies one medical service element in the medical service entry network, and one entry occupies one entry element in the medical service entry network, so that when ten thousand medical services and one thousand entries are included in the data for constructing the medical service entry network, ten thousand medical service elements and one thousand entry elements are correspondingly included in the medical service entry network.

Three relationships between elements are mentioned above: explaining the relationship, the jump relationship and the topic hierarchy relationship, and respectively introducing the three relationships below.

The interpretation relationship is an attribute relationship between the medical service (medical service element) and the entry (entry element). The interpretation relationship represents that a direct association relationship exists between the medical service and the entry, that is, an interpretation process occurs between the medical service and the entry, for example, the entry is added to the medical service, the entry is quoted by the medical service, a keyword of the entry is generated during construction of the medical service, related information of the entry is quoted by the medical service, and the like. For example, the interpretation relationship may be acquired by the smart medical mobile terminal 200, the smart medical mobile terminal 200 sends the acquired interpretation relationship to the smart medical cloud computing system 100 for storage, and the smart medical cloud computing system 100 constructs a medical service entry network according to the stored interpretation relationship of each medical service reported by each smart medical mobile terminal 200. For example, since the medical service has an interpretation relationship with the entry, the medical service element corresponding to the medical service has an interpretation relationship with the entry element corresponding to the entry.

A jump relationship is an attribute relationship between two medical services (medical service elements). The skip relationship represents that a direct association relationship exists between two medical services, for example, a service relationship exists between two medical services, a reference and referenced relationship exists between two medical services, a subscription and subscribed relationship exists between two medical services, a sharing and shared relationship exists between two medical services, two medical services (accounts) belong to the same facilitator, and the like. For example, the skip relation may also be an indirect association relation, for example, two medical services belong to the same service category, two medical services join the same correlation label together, two medical services have a service relation with another medical service, and the like. For example, the medical service mentioned in the present disclosure refers to a medical service that can be identified by the smart medical cloud computing system 100, that is, a medical service account, and if a plurality of medical service nodes exist in the same medical service, the same medical service is identified as a plurality of medical services by the smart medical cloud computing system. For example, since there is a jump relationship between two medical services, there is a jump relationship between the medical service element corresponding to the medical service and the medical service element corresponding to the other medical service.

The topic hierarchy is a weighted interpretation relationship. For example, the topic matching degree of the medical service to the entry at the topic level can be calculated according to the topic level of the entry, and the association degree of the medical service to the entry can be obtained according to the topic matching degree. This embodiment refers to the interpretation relationship weighted by the degree of association as the topic hierarchy relationship. That is, the topic hierarchy relationship is a topic matching degree relationship between the term and the frequent pattern item, for example, the term element having an interpretation relationship with the first medical service element includes a first term element, a second term element and a third term element, when the first term element is the center of the first medical service element, the topic matching degree of the first term element and the second term element is 3 units, and the topic matching degree of the first term element and the third term element is 1 unit, then the association degree of the first medical service element and the association attribute of the first term element is the greatest, the association degree of the first medical service element and the association attribute of the second term element is the next, and the association degree of the first medical service element and the association attribute of the third term element is the smallest. For example, the association attribute representing the topic hierarchical relationship is also a medical service element and a term element which are connected and have an interpretation relationship, but the association attribute may have an association degree, and the association degree represents the topic matching degree of the term element and the frequent pattern item of the medical service element, so the topic hierarchical relationship between the term elements can be represented according to the association degree on the association attribute.

For example, the smart medical cloud computing system 100 obtains topic associated data and extension data of a medical service entry, where the topic associated data of the medical service entry includes at least two medical services and entries related to the medical services, and the extension data includes at least one of medical service skip data and entry topic data; and constructing a medical service entry network according to the theme related data and the expansion data of the medical service entries.

For example, the above-mentioned medical service data is used to construct the medical service entry network, where the medical service data includes the subject related data and the extension data of the medical service entry. The topic association data of the medical service entry is used for recording an interpretation relationship between the medical service and the entry, for example, the topic association data of the medical service entry includes: medical service (medical service Identification (ID)), vocabulary entry (vocabulary entry Identification (ID)), and correspondence (interpretation relationship) between medical service and vocabulary entry. The extension data contains three possibilities: skipping data of medical services; entry topic data; medical service jump data and entry topic data. The medical service skipping data is used for recording the skipping relation between the medical service and the medical service, and the medical service skipping data comprises: the first medical service, the second medical service, and a correspondence relationship (jump relationship) between the first medical service and the second medical service. The term topic data is used for topic hierarchy information of the airborne term, for example, the term topic hierarchy data includes: the subject hierarchy of the entries and the entries, or the subject nodes of the entries and the entries.

For example, the data contained in the three data listed above includes data content that is only the basic data type required for implementing the method provided by the present disclosure, and based on the data types listed above, the subject associated data of the medical service entry may further include: the medical service updates the subject frequency of the entry, the medical service information and the like; the medical service jump data may further include: the type of the jump relationship (active jump, passive jump, etc.), the confidentiality of the service, the duration of the service relationship, the interaction frequency of the service, whether to refer specifically, the number of common service, etc.; the term topic data may further include: entry type, attributes, etc. of the entry.

For example, a medical service entry jumping network can be constructed based on the interpretation relationship and the jumping relationship, a medical service entry topic network can be constructed based on the interpretation relationship and the topic hierarchy relationship, and a medical service entry jumping network and a medical service entry topic network can be constructed based on the interpretation relationship, the jumping relationship and the topic hierarchy relationship. Different network contact characteristics between the medical service and the entry can be learned by learning based on different networks, and further medical service medical response information is provided from different layers.

Step S112, performing service learning based on the medical service entry network to obtain medical service learning characteristics of the medical service elements and entry learning characteristics of the entry elements.

And the service learning is used for deep learning and extracting network contact characteristics in the medical service entry network according to the association attributes among the elements.

For example, the embodiment may employ CNN to extract features in the medical service entry network, specifically feature propagation among elements, and for each element, may absorb and fuse features propagated by its associated element, and generate a new vector whose dimensions are unchanged by combining with its existing features. The characteristic propagation can be iterated for multiple layers, so that the connection information of one layer or even a high layer in a network structure is extracted; meanwhile, the initial attribute feature of the element can be used as the 0 th learning feature, so that the graph convolution network can effectively utilize the element attribute and the structure information of the network at the same time.

For example, the smart medical cloud computing system first generates 0 th learning feature (initial learning feature) of each element according to the attribute of each element itself, for example, generates an initial learning feature of a medical service element according to a medical service ID, and generates an initial learning feature of a term element according to a term ID. Or the intelligent medical cloud computing system can also generate initial learning features of the medical service elements according to the medical service ID and the medical service gender and generate initial learning features of the entry elements according to the entry ID and the entry type.

Then, the intelligent medical cloud computing system inputs the initial learning features of the element to the associated elements according to the connection relation between the element and the associated elements, and meanwhile, the element can obtain the initial learning features of the associated elements input by the associated elements. The element can learn the learning characteristics of the element according to the obtained initial learning characteristics of the associated element and the initial learning characteristics of the element to obtain the 1 st learning characteristic of the element. Thus, the 1 st learning feature of the element can learn the feature of the element related to the element. Repeating the step of business learning, the 2 nd learning feature of the element can be obtained according to the 1 st learning feature of the element and the 1 st learning feature of the associated element, and since the 1 st learning feature of the associated element has learned the feature of the associated element, the 2 nd learning feature of the element includes the feature of the associated element. Therefore, the element can continuously learn the characteristics of the remote element by iterative learning.

For example, as shown in fig. 4, taking three elements in the medical service entry network as an example, wherein element 1 and element 2 are labeled by the association attribute, element 2 and element 3 are labeled by the association attribute, element 1 has the initial learning feature 1, element 2 has the initial learning feature 2, and element 3 has the initial learning feature 3. First, feature propagation is performed, with element 1 sending initial learning features 1 to element 2, element 2 sending initial learning features 2 to element 1 and element 3, and element 3 sending initial learning features 3 to element 2. Then, feature learning is performed, the element 1 obtains a1 st learning feature 1 according to the obtained initial learning feature 2 and the initial learning feature 1 of itself, the element 2 obtains a1 st learning feature 2 according to the obtained initial learning feature 1 and the initial learning feature 3 and the initial learning feature 2 of itself, and the element 3 obtains a1 st learning feature 3 according to the obtained initial learning feature 2 and the initial learning feature 3 of itself. Then, feature propagation is performed again, element 1 inputs the 1 st learning feature 1 to element 2, element 2 inputs the 1 st learning feature 2 to element 1 and element 3, and element 3 inputs the 1 st learning feature 3 to element 2. Then, feature learning is performed again, the element 1 obtains a2 nd learning feature 1 according to the obtained 1 st learning feature 2 and the 1 st learning feature 1 of itself, the element 2 obtains a2 nd learning feature 2 according to the obtained 1 st learning feature 1 and the 1 st learning feature 3 and the 1 st learning feature 2 of itself, and the element 3 obtains a2 nd learning feature 3 according to the obtained 1 st learning feature 2 and the 1 st learning feature 3 of itself. Thus, element 1 can learn the feature of element 3 after two business studies, and similarly, element 3 can also learn the feature of element 1.

The feature propagation refers to a process in which each element inputs the learning feature of the element to the associated element, and the feature learning refers to a process in which each element updates the learning feature of the element according to the obtained learning feature. And the deeper learning characteristics can be obtained by carrying out the business learning in an iteration mode.

For example, when the association attribute in the medical service entry network is a directed association attribute, the feature propagation is propagated in one direction according to the direction indicated by the association attribute, for example, the association attribute of element 1 and element 2 points from element 1 to element 2, then during the feature propagation, element 1 will send the learning feature of this element to element 2, and element 2 will not send the learning feature of this element to element 1.

For example, the initial learning features of the healthcare business elements and the entry elements are generated from the element information of the elements. For example, the element information of the medical service element includes attribute information of the medical service, and for example, the element information includes at least one of a medical service ID, a medical service category, a medical service time, a medical service name, a medical service portrait, and a medical service unique code. The element information of the entry element includes attribute information of the entry, for example, the element information includes at least one of an entry ID and an entry type.

For example, in the following embodiments, the service learning for the medical service entry jumping network and the service learning for the medical service entry topic network may be introduced separately, and no matter in the medical service entry jumping network or the medical service entry topic network, the initial learning features of the same elements are the same, that is, the 0 th learning feature of the same element in the two networks is the same, and the bottom layer representation thereof is shared.

And S113, outputting medical response information of the target medical service according to the medical service element of the target medical service and at least two confidences of at least two entry elements.

In this embodiment, the confidence is determined according to the medical service learning features and the entry learning features.

For example, the confidence level is a combination of a medical service element and a term element corresponding to a confidence level. When the intelligent medical cloud computing system needs to determine medical response information of a medical service, the medical service elements are fixed as a target medical service, the term elements comprise all terms on a medical service term network, the confidence of each term (the rest terms except the term having an explanation relationship with the target medical service) in the target medical service and medical service term network is calculated, and the first terms with the highest confidence are used as the medical response information of the target medical service.

For example, after iterative business learning, each element may get multiple learning features. And calculating a confidence coefficient according to the learning characteristics of one medical service element and the learning characteristics of one entry element, wherein the confidence coefficient is used for representing the degree of relevance of the medical service to the entry. For example, a higher confidence level indicates a higher likelihood that the healthcare service will associate the entry in the future.

Therefore, to predict the medical response information that may be related to the target medical service, the confidence level of the medical service element of the target medical service and all the entry elements in the medical service entry network that do not have an interpretation relationship with the target medical service is calculated. And then, the terms which are not accessed by the medical services are sorted from high to low according to the confidence degrees, and the first few terms with the highest confidence degrees are used as medical response information of the target medical services to be output.

For example, in the case of medical response information, instead of calculating the confidence level between the term having an interpretation relationship with the medical service and the medical service, only the confidence level between the term having no interpretation relationship with the medical service and the medical service may be calculated, and then several terms are selected from the terms having no interpretation relationship and are associated with the medical service as medical response information.

In conclusion, the medical service entry network is constructed according to the interpretation relationship and the extended association attribute, and then the correlation relationship between the medical service and the entries is learned based on the medical service entry network, so that the medical information response is more accurately performed for the medical service. The method comprises the steps of performing service learning according to a medical service entry network, iteratively learning the association characteristics between the medical service and the entries, wherein the association characteristics not only comprise the service flow direction relation of the medical service, but also comprise the jump relation between the medical service and the medical service or the theme level attribute relation between the entries and the entries, and outputting the confidence that the medical service may be associated with a certain entry according to the association characteristics, so that the service response precision and reliability of medical service response are improved.

For example, an exemplary embodiment of constructing a medical service entry hop network based on an interpretation relationship and a hop relationship and performing feature propagation based on the medical service entry hop network is provided. Namely, the medical service entry network includes a medical service entry jump network, and the correlation attribute in the medical service entry jump network is used for the interpretation relationship and the jump relationship between the connection elements.

In a separate exemplary embodiment, on the basis of the foregoing exemplary embodiment, step S112 further includes step S1121, and step S113 further includes step S1131.

Step S1121, performing service learning based on the medical service entry jump network to obtain the medical service jump learning characteristic of the medical service element and the entry jump learning characteristic of the entry element.

For example, the extension data includes medical service jump data, and the medical service jump data includes jump relationships between medical services; the medical service entry network comprises a medical service entry jump network constructed according to the subject associated data of the medical service entry and the medical service jump data, and two elements with an interpretation relation or a jump relation in the medical service entry jump network are subjected to associated labeling by using associated attributes.

For example, the smart medical cloud computing system uses the association attribute to connect the medical service elements and the entry elements according to the interpretation relationship between the medical service elements and the entry elements, and uses the association attribute to connect two medical service elements with a jump relationship according to the jump relationship between the medical service elements, so as to obtain a medical service entry jump network.

For example, feature propagation includes two processes, feature input and feature acquisition.

First, feature propagation is performed. And calling the element to input the xth jump learning feature of the element to the associated element based on the medical service entry jump network, wherein the associated element is an element labeled by the associated attribute and the element, and the 0 th jump learning feature is an initial learning feature generated according to the element information of the element.

For example, the associated element is an element that inputs a learning feature to the present element, for example, if the associated attribute is a directed associated attribute, and element a points to element B through the directed associated attribute, element a is an associated element of element B, but element B is not an associated element of element a, because element B does not send the learning feature to element a.

Then, feature acquisition is performed. And the calling element obtains the x-th associated jump learning characteristic of the associated element input by the associated element.

For example, in the medical service entry skip network, for the medical service elements, the association elements include an associated medical service element and an associated entry element, and the xth associated skip learning feature obtained by the medical service element includes an xth associated medical service skip learning feature and an xth associated entry skip learning feature; for the entry element, the association element comprises an association medical service element, and the xth association skip learning feature obtained by the entry element comprises an xth association medical service skip learning feature;

for example, a feature learning update mode is given.

Performing feature learning according to the xth jump learning feature of the element and the xth associated jump learning feature of the associated element to obtain the xth +1 jump learning feature of the element; the medical service elements correspond to the (x +1) th medical service jump learning feature, and the entry elements correspond to the (x +1) th entry jump learning feature.

For example, in response to the element being a medical service association element, determining the sum of the xth jump learning feature, the xth associated medical service jump learning feature and the xth associated entry jump learning feature as the xth +1 jump feature vector; calculating the inner product of the (x +1) th jump feature vector and the (x +1) th jump weight; and calculating the (x +1) th jump sum of the (x +1) th jump vector inner product and the (x +1) th jump service feature, and substituting the (x +1) th jump sum into a Sigmoid function to obtain the (x +1) th medical service jump learning feature of the medical service element.

Determining the sum of the xth jump learning feature and the xth associated medical service jump learning feature as an xth +1 jump feature vector in response to the element being a vocabulary entry element; calculating the inner product of the 2(i +1) th jump feature vector of the (x +1) th jump feature vector and the (x +1) th jump weight; and calculating the 2(i +1) th jump sum of the 2(i +1) th jump vector inner product and the x +1 th jump service characteristic, and substituting the 2(i +1) th jump sum into a Sigmoid function to obtain the x +1 th entry jump learning characteristic of the entry element.

And repeating the steps to carry out service learning in an iteration mode, so that the Wth medical service skipping learning characteristic of the medical service element in the medical service entry skipping network is obtained, and the Wth entry skipping learning characteristic of the entry element in the medical service entry skipping network is obtained.

Step S1131, outputting medical response information of the target medical service according to the medical service element of the target medical service and at least two jump confidence coefficients of at least two vocabulary entry elements, wherein the jump confidence coefficients are determined according to the jump learning feature of the medical service and the jump learning feature of the vocabulary entry.

For example, a method of calculating hop confidence is presented.

The smart medical cloud computing system sequentially fuses W medical service skipping learning features of medical service elements to obtain medical service skipping fusion features; sequentially fusing the W entry jump learning characteristics of the entry elements to obtain entry jump fusion characteristics; obtaining a skipping confidence coefficient according to a vector inner product of the medical service skipping fusion feature and the entry skipping fusion feature; and determining the entries corresponding to the first K entry elements with the maximum skipping confidence coefficient of the medical service element of the target medical service in the at least two entry elements as the medical response information of the target medical service.

In conclusion, the correlation relationship between the medical service and the entry is further learned by adopting a graph convolution method according to the theme related data of the medical service entry and the medical service jump data, so that the medical information response is more accurately performed for the medical service. The medical service entry jump network is constructed by using the subject associated data of the medical service entry and the medical service jump data, the medical service entry jump network not only comprises the data of the medical service historical update entry, but also comprises the jump relation of the medical service, then, the business learning is carried out according to the medical business entry jumping network, the association characteristics between the medical business and the entries are iteratively learned, the related characteristics not only comprise the service flow relation of the medical service, but also comprise the jump relation between the medical service and the medical service, outputting a skipping confidence level that the medical service may be associated with a certain entry according to the association characteristics, and if the skipping confidence level is higher, indicating that the medical service is more relevant to the entry, the more the medical service tends to associate the entry, the entry is associated to the medical service, so that the service response precision and reliability of the medical service response are improved.

For example, an exemplary embodiment is provided for constructing a medical service entry topic network based on an interpretation relationship and a topic hierarchy relationship, and performing feature propagation based on the medical service entry topic network. Namely, the medical service entry network includes a medical service entry topic network, and two elements in the medical service entry topic network, which have an interpretation relationship or a topic hierarchical relationship, are associated and labeled by using an association attribute.

Another exemplary embodiment of the present disclosure is given below. On the basis of the foregoing exemplary embodiment, step S111 further includes step S1111 to step S1114, step S112 further includes step S1122, and step S113 further includes step S1132.

Step S1111, acquiring at least one frequent pattern item of the medical service by using a frequent pattern item mining algorithm, wherein the frequent pattern item is obtained according to the entry topic occurrence information of the entry having an interpretation relationship with the medical service.

For example, a network of medical services and associated terms may be constructed according to the interpretation relationship between the medical services and the terms, and the association attribute between elements in the network represents the interpretation relationship between the medical services and the terms. Based on this, the association attribute of the interpretation relation can be weighted according to the topic hierarchy relation of the vocabulary entry.

For example, each entry that is highly popular with medical services typically surrounds several frequent pattern terms. Therefore, according to the terms associated with the medical services in the topic association data of the medical service terms, the frequent pattern terms with higher heat of the medical services are found out, and then the topic matching degree of each term and the frequent pattern terms is respectively calculated, so that the association degree of the medical services to the terms in the topic matching degree can be obtained, for example, the greater the difference from the frequent pattern terms, the smaller the association degree of the medical services to the terms.

Therefore, the smart medical cloud computing system needs to find out the frequent pattern items with high popularity in the medical service from the plurality of entries having the interpretation relation with the medical service. For example, the topic association data of the medical service entries further includes the times or frequency of the medical service update entries, the smart medical cloud computing system sorts the entries according to the times or frequency of the medical service update entries from high to low, then takes the entry with the highest time or frequency as a first frequent pattern item, respectively computes the direct topic matching degrees of other entries and the frequent pattern item, and divides the entries with the direct topic matching degrees larger than the topic matching degree threshold value into the frequent pattern item partitions corresponding to the frequent pattern items. And finding out the entry with the highest frequency or frequency from the entries with the direct subject matching degree larger than the threshold value of the subject matching degree as a second frequent pattern item, respectively calculating the direct subject matching degrees of the remaining entries and the second frequent pattern item, and dividing the entry with the direct subject matching degree larger than the threshold value of the subject matching degree into the frequent pattern item partition corresponding to the second frequent pattern item. Therefore, the process of obtaining the frequent mode items is completed until all the entries are divided into the frequent mode item partition of one frequent mode item, and at least one frequent mode item with high medical service popularity can be obtained by using the method.

For example, the frequent pattern item with a high medical service popularity may be found according to other manners, for example, after the multiple entries are divided into multiple frequent pattern item regions according to the above method, the topic level average value of all entries in the frequent pattern item partition is obtained, and the entry corresponding to the topic level average value is used as the frequent pattern item of the frequent pattern item partition. For another example, the first few terms with the highest frequency degree of the medical service are taken as frequent pattern items, and then other terms are divided into frequent pattern item partitions of the frequent pattern items with the closest topic matching degree.

For example, a frequent pattern item partition refers to a partition interval in which the topic matching degree with the frequent pattern item is greater than a topic matching degree threshold. And the theme matching degree threshold is used for dividing the entries into frequent pattern item partitions corresponding to the frequent pattern items. For example, the threshold value of the degree of matching of the subject corresponding to different frequent pattern items may be the same or different. For example, a distribution diagram of terms is drawn according to the subject level information (subject level) of the terms, the terms having an interpretation relationship with the medical service are then used as a first frequent pattern item, the term with the largest number of medical service visits on the distribution diagram is used as a reference item, a frequent pattern item partition of the first frequent pattern item is found with a subject matching degree threshold as a range, and the terms located in the frequent pattern item partition belong to the frequent pattern item partition of the first frequent pattern item. And then, taking the entry with the highest medical service access frequency in the entries outside the frequent pattern entry partition of the first frequent pattern item as a second frequent pattern item, and finding out the frequent pattern entry partition of the second frequent pattern item by taking the second frequent pattern item as a reference item and taking a topic matching degree threshold as a range, wherein the entries positioned in the frequent pattern entry partition belong to the frequent pattern entry partition of the second frequent pattern item. Therefore, a plurality of frequent pattern items of the medical service can be found out on the distribution diagram according to the topic hierarchy information of the vocabulary entry.

Step S1112, divide the entries whose topic matching degrees with the frequent pattern items are greater than the threshold into the frequent pattern item partitions corresponding to the frequent pattern items.

For example, according to the explanation of step S1111, a plurality of entries corresponding to medical services may be recently divided into the frequent pattern item partition.

For example, when the topic matching degree of one entry topic is smaller than the threshold (topic matching degree threshold), the entry is divided into the frequent pattern item partition of the frequent pattern item with the closest topic matching degree, i.e. the division is performed according to the principle of closeness.

And S1113, calculating the topic matching degree of the entries in the frequent pattern item partition and the frequent pattern item, and determining the association degree of the medical service to the entries according to the topic matching degree.

And respectively calculating the direct subject matching degree of each entry and the frequent pattern item in the frequent pattern item partition according to the subject level information (subject level) of the entry, and then determining the association degree of the entry according to the direct subject matching degree, so that the entry with longer direct subject matching degree is associated with smaller degree, and the entry with shorter direct subject matching degree is associated with larger degree.

For example, the weight of the entry corresponding to the frequent pattern item in each frequent pattern item region is determined as 1, and the weights of the other entries are 1 divided by the direct topic matching degree, so that the farther the direct topic matching degree is, the smaller the correlation degree is, and the closer the direct topic matching degree is, the larger the correlation degree is.

Step S1114, a medical service entry topic network is constructed according to the interpretation relation between the medical service and the entry and the association degree of the medical service to the entry.

The association degree of the terms according to the medical service can be weighted for the interpretation relationship, that is, in the medical service term topic network, the association attributes between the medical service and the terms have weighted values, the weighted values of different association attributes are different, and the weighted values are the association degree of the medical service to the terms.

In the feature learning, the vector of the element may be updated according to the weight value of the associated attribute. For example, when the element obtains a first learning feature of a first related element and a second learning feature of a second related element during feature propagation, where the degree of association between the element and the first related element related attribute is 0.5 and the degree of association between the element and the second related element related attribute is 0.3, the first learning feature and the second learning feature are weighted by the degree of association, and feature learning is performed based on the learning feature of the element, the weighted first learning feature, and the weighted second learning feature. For example, the first learning feature after weighting is 0.5 × first learning feature, and the second learning feature after weighting is 0.3 × second learning feature.

Therefore, the topic hierarchy information of the entry can be introduced into the medical service entry network, and the characteristics of the medical service correlation on the topic hierarchy level can be extracted according to the topic hierarchy information of the entry, so that the medical service response information can be better sent to the medical service.

Step S1122, performing service learning based on the medical service entry topic network to obtain medical service topic learning characteristics of the medical service elements and entry topic learning characteristics of the entry elements.

For example, the extension data includes entry topic data including at least entry topic occurrence information of an entry; the medical service entry network comprises a medical service entry topic network, and the association attribute in the medical service entry topic network is used for connecting the access relation between the medical service element and the entry element and the topic hierarchy relation between the medical service element and the entry element.

First, feature propagation is performed. And calling the element to input the xth theme learning feature of the element to the associated element based on the medical service entry theme network, wherein the associated element is an element labeled by the associated attribute and the element, and the 0 th theme learning feature is an initial learning feature generated according to the element information of the element.

Then, feature acquisition is performed. And the calling element obtains the x-th associated topic learning characteristic of the associated element input by the associated element.

For example, in a medical service entry topic network, for a medical service element, an association element includes an association entry element, and then the xth association topic learning feature obtained by the medical service element includes an xth association entry topic learning feature; for the entry element, the associated element comprises an associated medical service element, and the xth associated subject learning feature obtained by the entry element comprises an xth associated medical service subject learning feature;

for example, a feature learning update mode is given.

Performing feature learning according to the xth theme learning feature of the element and the xth associated theme learning feature of the associated element to obtain the xth +1 theme learning feature of the element; the medical service elements correspond to the (x +1) th medical service theme learning features, and the entry elements correspond to the (x +1) th entry theme learning features.

For example, in response to the element being a medical service association element, calculating a first vector inner product of the association degree of the xth associated entry topic learning feature and the association element, and determining the sum of the xth topic learning feature and the first vector inner product as the xth topic feature vector; calculating the (x +1) th theme feature vector inner product of the (x +1) th theme feature vector and the (x +1) th theme weight; and calculating the sum of the (x +1) th theme vector inner product and the (x +1) th theme biased by the (x +1) th theme, and substituting the (x +1) th theme sum into a Sigmoid function to obtain the (x +1) th medical service theme learning characteristic of the medical service element.

Responding to the condition that the element is a vocabulary entry element, calculating a second vector inner product of the association degree of the xth associated medical service theme learning feature and the associated element, and determining the sum of the xth theme learning feature and the second vector inner product as an xth +1 theme feature vector; calculating the 2(i +1) th theme feature vector inner product of the (x +1) th theme feature vector and the (x +1) th theme weight; and calculating the 2(i +1) th theme sum of the 2(i +1) th theme vector inner product and the x +1 th theme offset, and substituting the 2(i +1) th theme sum into a Sigmoid function to obtain the x +1 th entry theme learning characteristic of the entry element.

And repeating the steps to carry out service learning in an iteration mode, so that the W-th medical service theme learning characteristic of the medical service element in the medical service vocabulary entry theme network is obtained, and the W-th vocabulary entry theme learning characteristic of the vocabulary entry element in the medical service vocabulary entry theme network is obtained.

Step S1132, outputting medical response information of the target medical service according to the medical service element of the target medical service and at least two topic confidence degrees of the at least two entry elements, where the topic confidence degree is determined according to the medical service topic learning feature and the entry topic learning feature.

For example, a method of calculating hop confidence is presented.

The intelligent medical cloud computing system sequentially fuses the W medical service theme learning features of the medical service elements to obtain medical service theme fusion features; sequentially fusing the W entry topic learning characteristics of the entry elements to obtain entry topic fusion characteristics; obtaining a topic confidence coefficient according to the vector inner product of the medical service topic fusion feature and the entry topic fusion feature; and determining the entries corresponding to the first K entry elements with the maximum topic confidence of the medical service element of the target medical service in the at least two entry elements as the medical response information of the target medical service.

In summary, the correlation relationship between the medical service and the entry is further learned by adopting a graph convolution method according to the topic associated data and the entry topic data of the medical service entry, so that the medical information response is more accurately performed for the medical service. The medical service entry topic network is constructed by using topic associated data and entry topic data of medical service entries, the medical service entry topic network not only contains data of medical service history updating entries, but also contains topic hierarchical relations of the entries and frequent pattern items of the medical service, then service learning is carried out according to the medical service entry topic network, association characteristics between the medical service and the entries are iteratively learned, the association characteristics not only contain service flow direction relations of the medical service, but also contain topic hierarchical relations of the entries and frequent pattern items, a topic confidence coefficient that the medical service may be associated with one entry is output according to the association characteristics, if the topic confidence coefficient is high, the relevance of the medical service to the entry is large, the medical service tends to associate the entry, the entry is associated with the medical service, the service response precision and reliability of medical service response are improved.

For example, an exemplary embodiment is provided for constructing a medical service entry jumping network and a medical service entry topic network based on an interpretation relationship, a jumping relationship and a topic hierarchy relationship, and performing feature propagation based on the medical service entry jumping network and the medical service entry topic network. Namely, the medical service entry network comprises a medical service entry skip network and a medical service entry topic network; two elements with interpretation relation or jump relation in the medical service entry jump network are associated and labeled by using the association attributes; and performing association labeling on two elements with interpretation relation or topic hierarchical relation in the medical service entry topic network by using the association attribute.

Another exemplary embodiment of the present disclosure is given below. For example, step S112 further includes steps S1121 through S1122, and step S113 further includes steps S1133 through S1136.

For example, the explanation of step S1121 may refer to step S1121 described previously.

For example, the explanation of step S1122 can refer to the aforementioned step S1122.

For example, in the present embodiment, the sequence executed in step S1121 and step S1122 is not limited, and the two steps may be executed simultaneously; the two steps can also be executed in sequence, and the sequence can also be any.

Step S1133, sequentially fusing the W medical service jump learning features of the medical service elements to obtain a medical service jump fusion feature; sequentially fusing the W entry jump learning characteristics of the entry elements to obtain entry jump fusion characteristics; and obtaining the skipping confidence coefficient according to the vector inner product of the medical service skipping fusion characteristic and the entry skipping fusion characteristic.

Step S1134, sequentially fusing the W medical service theme learning features of the medical service elements to obtain medical service theme fusion features; sequentially fusing the W entry topic learning characteristics of the entry elements to obtain entry topic fusion characteristics; and obtaining the topic confidence coefficient according to the vector inner product of the medical service topic fusion feature and the entry topic fusion feature.

For example, business learning can be performed in a medical business entry jumping network and a medical business entry topic network, respectively. That is, the same healthcare element has different learning features in the two networks (the 0 th learning feature is the same, and the other learning features are different). Therefore, the skipping confidence coefficient and the topic confidence coefficient can be respectively calculated according to the learning features in two different networks, and then the final confidence coefficient is obtained according to the skipping confidence coefficient and the topic confidence coefficient.

Step S1135, determining the weighted confidence of the jump and the confidence of the topic as the confidence.

Step S1136, determining, as the medical response information of the target medical service, a vocabulary entry corresponding to the first K vocabulary entry elements with the maximum confidence of the medical service element of the target medical service, from among the at least two vocabulary entry elements.

The intelligent medical cloud computing system sequences the entries corresponding to the entry elements according to the entry elements which are not associated with the target medical services and the confidence degrees between the medical service elements of the target medical services, and outputs the first entries with the highest confidence degrees as medical response information of the target medical services.

In summary, the medical service entry jump network is constructed according to the interpretation relationship between the medical service and the entry and the jump relationship between the medical service and the medical service, the jump relationship between the medical service is introduced into the medical service entry network, and the medical service correlation is extracted according to the interpretation relationship and the jump relationship. The medical service entry topic network is constructed according to the interpretation relation between the medical service and the entries and the topic hierarchy relation among the entries, the topic hierarchy information of the entries is introduced into the medical service entry network, and the medical service correlation is extracted according to the interpretation relation and the topic hierarchy of the entries. Then, the confidence coefficient is calculated according to the jump confidence coefficient of the jump perception level and the topic confidence coefficient of the topic perception level, which are extracted and determined by the two networks, wherein the confidence coefficient comprises perception characteristics of the jump level, the topic level and the explanation level, and the relevance entries of the medical service are predicted according to the three levels, so that the accuracy of the medical response information can be improved. Moreover, when the interpretation relation of the medical service is less, the entry recommendation can be performed on the medical service by using the other two relations, so that the correlation acquisition capacity of the model on the cold-start medical service is improved, and the medical response information of the medical service is better.

For example, the method provided by the present disclosure is executed by a term recommendation model on a smart medical cloud computing system, and this embodiment provides a training method for the term recommendation model.

Another exemplary embodiment of the present disclosure is given below, for example, on the basis of the foregoing exemplary embodiment, step S113 is followed by step S114 to step S117.

Step S114, obtaining a first annotation object confidence of the first annotation object, wherein the first annotation object comprises a sample medical service with an interpretation relationship and a first annotation object entry, and the first annotation object confidence is the confidence of the sample medical service and the first annotation object entry.

Step S115, obtaining a second annotation object confidence of a second annotation object, wherein the second annotation object comprises a sample medical service without an interpretation relationship and a second annotation object entry, and the second annotation object confidence is the confidence of the sample medical service and the second annotation object entry.

For example, after a network of medical business terms is constructed and business learning is performed on elements on the network, the smart medical response model may be trained based on the confidence level output by the smart medical response model.

For example, based on the method of partial order learning, the medical service and the entry having an interpretation relationship are used as a first labeled object, the medical service and the entry not having the interpretation relationship are used as a second labeled object, a difference parameter value between the first labeled object and the second labeled object is calculated by using a loss function, and a model parameter is adjusted according to the difference parameter value, so that the confidence coefficient theme matching degree of the first labeled object and the second labeled object is increased, and the larger the confidence coefficient difference value of the first labeled object and the second labeled object is, the stronger the distinguishing capability of the intelligent medical response model for the first labeled object and the second labeled object is.

For example, taking an example of calculating a difference parameter value by using a first annotation object and a second annotation object, the smart medical cloud computing system selects a sample medical service from a plurality of medical services, and then obtains a first annotation object and a second annotation object of the sample medical service, that is, obtains a confidence of the sample medical service and an entry having an interpretation relationship, and obtains a confidence of the sample medical service and an entry having no interpretation relationship, and then calls a loss function to calculate a difference parameter value of the two confidences. For example, a set of first annotation object and second annotation object must correspond to the same medical service, i.e. the triple data forming < medical service, first annotation object, second annotation object >, and the difference parameter value is calculated according to the confidence of the first annotation object and the second annotation object in the triple data.

For example, for a medical service, the number of terms having an interpretation relationship is small, and the number of terms having no interpretation relationship is large, so for each first annotation object, a part of the terms having no interpretation relationship is randomly extracted from a plurality of terms having no interpretation relationship according to a certain proportion as a second annotation object to form triple data, and each group of triple data is substituted into a loss function to calculate a difference parameter value.

Step S116, a loss function is called to calculate a difference parameter value between the confidence of the first annotation object and the confidence of the second annotation object.

Step S117, training the intelligent medical response model according to the difference parameter values.

For example, the smart medical cloud computing system adjusts the smart medical response model according to the difference parameter value, so as to minimize the difference parameter value of the confidence coefficient output by the adjusted smart medical response model. For example, the smart medical cloud computing system may minimize the discrepancy parameter value of the confidence of the output by adjusting each jump weight, jump service feature, topic weight, and topic bias for each element.

In summary, in the method provided in this embodiment, the intelligent medical response model is trained by using a method based on partial order learning, so that a difference between the confidence level output by the model to the first labeled object and the confidence level output by the model to the second labeled object is as large as possible, thereby improving the distinguishing capability of the model to the first labeled object and the second labeled object, and improving the capability of accurately recommending the model.

Fig. 3 is a schematic diagram of functional modules of a big data based smart medical interactive device 300 according to an embodiment of the present disclosure, and the functions of the functional modules of the big data based smart medical interactive device 300 are described in detail below.

The obtaining module 310 is configured to obtain medical response information of a target medical service initiated by the smart medical mobile terminal, and obtain an online consultation topic initiated by the smart medical mobile terminal for any medical response entry in the medical response information.

The determining module 320 is configured to obtain medical service knowledge points related to the online consultation subject, and determine a degree of relevance of each medical service knowledge point to each to-be-answered question based on knowledge point code distribution of each medical service knowledge point and question code distribution of each to-be-answered question in the to-be-answered question set.

The calculating module 330 is configured to obtain the heat of the knowledge point corresponding to each medical service knowledge point, and calculate a referenceable degree of each medical service knowledge point for the set of questions to be answered based on the correlation degree and the heat of the knowledge point.

And the selecting module 340 is used for selecting a target knowledge point from the medical service knowledge points to be associated to the set of questions to be answered based on the referenceable degree so as to interact with the online consultation theme initiated by the intelligent medical mobile terminal.

Fig. 4 illustrates a hardware structure of the smart medical cloud computing system 100 for implementing the big data based smart medical interaction method, according to an embodiment of the present disclosure, as shown in fig. 4, the smart medical cloud computing system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a transceiver 140.

In a specific implementation process, at least one processor 110 executes the smart medical cloud computing system stored in the machine-readable storage medium 120 to execute instructions, so that the processor 110 may execute the smart medical interaction method based on big data according to the above method embodiment, the processor 110, the machine-readable storage medium 120, and the transceiver 140 are connected through the bus 130, and the processor 110 may be configured to control the transceiving action of the transceiver 140, so as to perform data transceiving with the smart medical mobile terminal 200.

For a specific implementation process of the processor 110, reference may be made to the above-mentioned various method embodiments executed by the smart medical cloud computing system 100, which implement similar principles and technical effects, and the detailed description of the embodiments is omitted here.

In addition, the readable storage medium is preset with an intelligent medical cloud computing system execution instruction, and when the processor executes the intelligent medical cloud computing system execution instruction, the intelligent medical interaction method based on big data is realized.

Finally, it should be understood that the examples in this specification are only intended to illustrate the principles of the examples in this specification. Other variations are also possible within the scope of this description. Accordingly, by way of example, and not limitation, alternative configurations of the embodiments of the specification can be seen as matching the teachings of the specification. Accordingly, the embodiments of the present description are not limited to only those embodiments explicitly described and depicted herein.

Claims

1. A smart medical interaction method based on big data is applied to a smart medical cloud computing system, the smart medical cloud computing system is in communication connection with a plurality of smart medical mobile terminals, and the method comprises the following steps:

acquiring medical service knowledge points related to the online consultation theme, and determining the correlation degree of each medical service knowledge point for each problem to be responded based on the knowledge point code distribution of each medical service knowledge point and the problem code distribution of each problem to be responded in the problem set to be responded, wherein the knowledge point code distribution comprises a knowledge point range, a knowledge point type, a knowledge point applicable condition, a knowledge point coverage range, a knowledge point query quantity and a knowledge point response quantity, the problem code distribution comprises a problem basic code distribution to be responded and a problem specific code distribution to be responded, the problem basic code distribution to be responded comprises a problem category to be responded, a problem label and a problem grade, and the problem specific code distribution to be responded comprises an extended problem direction of the problem to be responded;

based on the referential degree, selecting a target knowledge point from the medical service knowledge points to be associated with the set of questions to be answered so as to interact with an online consultation subject initiated by the intelligent medical mobile terminal;

the determining the degree of correlation of each medical service knowledge point to each question to be answered based on the knowledge point code distribution of each medical service knowledge point and the question code distribution of each question to be answered in the question set to be answered includes:

inputting the knowledge point code distribution of each medical service knowledge point into a first machine learning unit of a machine learning network, and inputting the question code distribution of each question to be answered into a second machine learning unit of the machine learning network, wherein the first machine learning unit and the second machine learning unit are connected to a global learning unit;

acquiring the correlation degree of each medical service knowledge point output by the global learning unit aiming at each problem to be responded;

wherein the first machine learning unit includes: the device comprises a first nonlinear characteristic fitting subunit and a first fusion subunit connected with the output end of the first nonlinear characteristic fitting subunit;

the first nonlinear feature fitting subunit is configured to perform nonlinear feature fitting processing on a third knowledge point code distribution in the knowledge point code distributions, and the first fusion subunit is configured to fuse a second knowledge point code distribution in the knowledge point code distributions and a knowledge point nonlinear feature fitting feature output by the first nonlinear feature fitting subunit;

the second machine learning unit includes: the input end of the second fusion subunit is connected with the output end of the second nonlinear feature fitting subunit and the output end of the conditional random field subunit;

the second nonlinear feature fitting subunit is configured to perform nonlinear feature fitting processing on a third problem code distribution in the problem code distribution, the conditional random field subunit is configured to perform conditional random field calculation on a first problem code distribution in the problem code distribution, and the second fusion subunit is configured to fuse a second problem code distribution in the problem code distribution, the problem nonlinear feature fitting feature output by the second nonlinear feature fitting subunit, and the conditional random field feature output by the conditional random field subunit.

2. The intelligent medical interaction method based on big data as claimed in claim 1, wherein the calculating the referenceable degree of each medical service knowledge point for the set of questions to be answered based on the correlation degree and the knowledge point heat degree comprises:

calculating the product of the heat degree of the knowledge points corresponding to the medical service knowledge points and the correlation degree of the medical service knowledge points for the problems to be responded to obtain the initial reference degree of the medical service knowledge points for the problems to be responded to;

and weighting the initial reference degree of each medical service knowledge point for each problem to be responded to obtain the referential degree of each medical service knowledge point for the problem set to be responded to.

3. The intelligent medical interaction method based on big data as claimed in claim 1, wherein the selecting a target knowledge point from the medical service knowledge points to be associated with the set of questions to be answered based on the referenceable degree comprises:

according to the sequence of the referential degree from large to small, selecting a set number of medical service knowledge points from the medical service knowledge points as the target knowledge points;

and associating the target knowledge point to the set of questions to be answered.

4. The intelligent medical interaction method based on big data as claimed in any one of claims 1-3, wherein the determining the degree of relevance of each medical service knowledge point to each question to be answered based on the knowledge point code distribution of each medical service knowledge point and the question code distribution of each question to be answered in the set of questions to be answered comprises:

calculating distribution association parameters between the problem code distribution of each to-be-responded problem and the knowledge point code distribution of each medical service knowledge point;

and performing input mapping on the distribution correlation parameters through a Sigmoid function to obtain the correlation degree of each medical service knowledge point for each to-be-responded question.

5. The intelligent big data-based medical interaction method according to claim 4, wherein before calculating distribution association parameters between the question code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point, the method further comprises:

and mapping the problem code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point to the same code distribution grid.

6. The intelligent medical interaction method based on big data as claimed in claim 5, wherein the mapping of the question code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point to the same code distribution grid comprises:

acquiring a problem code distribution range corresponding to the problem code distribution of each problem to be responded and a knowledge point code distribution range corresponding to the knowledge point code distribution of each medical service knowledge point;

and inputting the problem code distribution of each problem to be responded into a kernel function with the kernel function processing range as the problem code distribution range for data conversion, and inputting the knowledge point code distribution of each medical service knowledge point into a kernel function with the kernel function processing range as the knowledge point code distribution range for data conversion, so as to convert the problem code distribution of each problem to be responded and the knowledge point code distribution of each medical service knowledge point into the same code distribution grid.

7. The intelligent big data-based medical interaction method according to claim 5, wherein each question to be answered has a plurality of question coding distributions, the plurality of question coding distributions comprising: the problem code distribution comprises a first problem code distribution, a second problem code distribution and a third problem code distribution, wherein the first problem code distribution is used for describing the problem response strategy to be responded and the service position updating relation, the second problem code distribution is continuously updated along with the service position updating, and the third problem code distribution is not continuously updated along with the service position updating;

each medical service knowledge point has a plurality of knowledge point coding distributions, and the plurality of knowledge point coding distributions comprise: a second knowledge point code distribution which is continuously updated with the service location update and a third knowledge point code distribution which is not continuously updated with the service location update;

before mapping the question code distribution of each question to be answered and the knowledge point code distribution of each medical service knowledge point to the same code distribution grid, the method further comprises:

fusing the multi-question code distribution of each question to be responded, and fusing the multi-knowledge point code distribution of each medical service knowledge point;

wherein, the fusing the distribution of the multiple question codes of each question to be answered comprises:

performing conditional random field calculation on every two first problem code distributions of each problem to be responded to obtain conditional random field characteristics;

carrying out nonlinear feature fitting processing on every two third problem code distributions of each to-be-responded problem to obtain problem nonlinear feature fitting features;

fusing the second problem code distribution, the conditional random field characteristics and the problem nonlinear characteristic fitting characteristics of each problem to be responded;

fusing the coding distribution of the multiple knowledge points of each medical service knowledge point, including:

carrying out nonlinear feature fitting processing on every two third knowledge point code distributions of the medical service knowledge points to obtain nonlinear feature fitting features of the knowledge points;

and fusing the second knowledge point coding distribution of each medical service knowledge point and the nonlinear feature fitting feature of the knowledge point.

8. The intelligent medical interaction method based on big data as claimed in any one of claims 1-3, wherein the step of obtaining the medical response information of the target medical service initiated by the intelligent medical mobile terminal comprises:

constructing a medical service entry network, wherein the medical service entry network comprises elements and association attributes, the elements comprise medical service elements corresponding to medical services and entry elements corresponding to entries, the two elements with interpretation relations or extended association attributes are subjected to association labeling by using the association attributes, the interpretation relations comprise attribute relations between the medical service elements which have initiated an interpretation process and the entry elements, the extended association attributes comprise at least one of jump relations between the medical service elements and topic hierarchy relations between the entry elements, and the medical service entry network is constructed based on big data generated by historical medical services;

performing service learning based on the medical service entry network to obtain medical service learning characteristics of medical service elements and entry learning characteristics of the entry elements, wherein the service learning is used for performing deep learning according to the association attributes among the elements and extracting network contact characteristics in the medical service entry network;

and outputting medical response information of the target medical service according to at least two confidence degrees of medical service elements of the target medical service and at least two entry elements, wherein the confidence degrees are determined according to the medical service learning characteristics and the entry learning characteristics.

9. A smart medical cloud computing system, comprising a processor, a machine-readable storage medium, and a network interface, wherein the machine-readable storage medium, the network interface, and the processor are associated with each other through a bus system, the network interface is configured to be communicatively connected to at least one smart medical mobile terminal, the machine-readable storage medium is configured to store smart medical cloud computing system instructions, and the processor is configured to execute the smart medical cloud computing system instructions in the machine-readable storage medium to perform the big data based smart medical interaction method according to any one of claims 1 to 8.