CN116860964A

CN116860964A - User portrait analysis method, device and server based on medical management label

Info

Publication number: CN116860964A
Application number: CN202310626489.2A
Authority: CN
Inventors: 杨浩; 沈中一
Original assignee: Shenzhen Guokang Health Management Service Co ltd
Current assignee: Shenzhen Guokang Health Management Service Co ltd
Priority date: 2023-05-30
Filing date: 2023-05-30
Publication date: 2023-10-10

Abstract

According to the user portrait analysis method, the device and the server based on the medical management tags, the medical management texts of the target users are processed through the management tag determination model to obtain a plurality of medical management tags, each medical management tag forms the user portrait of the target users, accuracy and efficiency are high, in the execution process, the model is iterated through the true medical management training text and the false medical management training text together to obtain the management tag determination model, when the medical management update training text is obtained and the update management tag does not belong to the existing real management tag, the existing management tag determination model is not repeatedly calibrated, the assumed management tag is used as a new tag type, and the iteration speed of the management tag determination model is higher. Further, when a new management label type appears, the corresponding assumption management label is correspondingly updated, so that the existing label type distribution condition is reserved, and the existing medical record text recognition is not influenced.

Description

User portrait analysis method, device and server based on medical management label

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a user portrait analysis method, device and server based on medical management labels.

Background

With the development and advancement of smart medical concepts, the management of medical data for patients or users, whether in hospitals or other healthcare facilities, is an important link. For example, by classifying medical management texts such as medical records, physical examination reports and the like of patients or users, user portraits of the users are obtained, and subsequent medical services such as condition tracking, health management, physical examination customization and the like can be assisted. Among them, the bulky nature of medical data presents difficulties for its classification. With the progress of artificial intelligence technology and the continuous promotion of large chips, artificial intelligence enables large data processing in different technical scenes, and a traditional data analysis mode can be replaced by a more efficient machine learning model. Based on this, how to solve the problem of medical big data classification by means of an artificial intelligence model is a breakthrough problem in the art, and how to efficiently maintain an artificial intelligence algorithm is also an indispensable link.

Disclosure of Invention

In view of the above, the embodiment of the application at least provides a user portrait analysis method based on a medical management tag.

The technical scheme of the embodiment of the application is realized as follows:

in one aspect, an embodiment of the present application provides a user portrait analysis method based on a medical management tag, which is applied to a server, and the method includes:

acquiring a plurality of medical management texts of a target user, and respectively inputting the medical management texts of the plurality of users into a management tag determination model after the medical management texts are calibrated in advance;

for each medical management text, extracting a text representation carrier of the medical management text through the management label determining model, and obtaining a medical management label of the medical management text based on the text representation carrier;

constructing a user portrait of the target user through the medical management labels corresponding to the medical management texts;

the management tag determination model is obtained by iterating a true medical management training text and a false medical management training text together, and can estimate an assumed management tag and a real management tag;

When the medical management label is updated, acquiring a medical management update training text, wherein the update management label corresponding to the medical management update training text is different from the existing reality management label;

inputting the medical management update training text into the management tag determination model, and extracting a text representation carrier of the medical management update training text based on the management tag determination model;

determining target center representation carriers respectively corresponding to each assumption management label in a carrier set, and obtaining target assumption management labels matched with the medical management update training texts according to errors between the text representation carriers and the target center representation carriers;

and taking the target assumption management label as the update management label, wherein the management label determination model can estimate the update management label.

In some embodiments, the method further comprises:

obtaining a true medical management training text and a false medical management training text, wherein the false medical management training text is obtained by transformation according to the true medical management training text;

estimating the real medical management training text based on a management label determining model to be calibrated, and determining a first cost of the real medical management training text and a first target label according to a first estimated label obtained by estimation, wherein the first target label comprises a target reality management label corresponding to the real medical management training text and a target assumption management label matched with the real medical management training text;

Estimating the fake medical management training text through the management label determining model to be calibrated, and determining a second cost of the fake medical management training text and a second target label according to a second estimated label obtained through estimation, wherein the second target label comprises a target reality management label and a target assumption management label which are respectively matched with the fake medical management training text;

and generating a target quality evaluation algorithm based on the first cost and the second cost, and repeatedly calibrating the management tag determination model to be calibrated through the target quality evaluation algorithm until the model converges to obtain the calibrated management tag determination model.

In some embodiments, the repeatedly tuning the management tag determination model to be tuned by the target quality assessment algorithm includes:

performing multi-round adjustment on the management tag determination model to be adjusted through the target quality evaluation algorithm, calculating a model change vector corresponding to the current round adjustment when each round adjustment is completed, and iterating a center representation carrier corresponding to each tag type in a carrier set according to the model change vector; the center of each label type obtained by the last round of adjustment represents a carrier, and the center of each label type represents the carrier;

The determining the model based on the management label to be calibrated estimates the real medical management training text, and determining the first cost of the real medical management training text and the first target label according to the estimated first estimated label comprises the following steps:

extracting a text representation carrier of the real medical management training text based on a management label determining model to be calibrated, and estimating according to the text representation carrier of the real medical management training text to obtain a first credible coefficient of the real medical management training text corresponding to all label types;

determining a first initial cost of the real medical management training text corresponding to the target reality management label through the first trusted coefficient and the target reality management label corresponding to the real medical management training text;

determining, by the first trusted coefficient, a first assumed trusted coefficient of the real medical management training text corresponding to the remaining tag types other than the target reality management tag;

determining a first assumption cost of the true medical management training text corresponding to the associated target assumption management label according to the first assumption credibility coefficient;

And determining a first cost of the real medical management training text and a first target label according to the first initial cost and the first assumed cost.

In some embodiments, the determining, by the first confidence coefficient, that the real medical management training text corresponds to a first hypothesized confidence coefficient for the remaining tag types other than the target reality management tag includes:

obtaining a coding result corresponding to the real medical management training text based on a target reality management tag corresponding to the real medical management training text, wherein the number of elements of the coding result is the same as the number of preset tag types;

performing reverse calculation on the coding result corresponding to the real medical management training text to obtain a reverse coding result corresponding to the real medical management training text;

determining a first assumed trusted coefficient of the real medical management training text corresponding to the rest tag types except the target reality management tag according to the first trusted coefficient and the reverse coding result corresponding to the real medical management training text;

said determining a first hypothesized cost of said true medical management training text corresponding to an associated target hypothesized management tag in accordance with said first hypothesized confidence coefficient, comprising:

Acquiring a target assumption management label matched with the real medical management training text;

determining a first assumption cost of the true medical management training text corresponding to the target assumption management label according to the first assumption credibility coefficient and the assumption management label indication corresponding to the target assumption management label;

wherein the obtaining the target hypothesis management tag matched with the true medical management training text comprises:

determining a first center representation carrier corresponding to each assumption management label in a carrier set, wherein the first center representation carrier is a center representation carrier corresponding to each assumption management label in current round adjustment;

determining the proximity of the text representing carrier of the real medical management training text and each first center representing carrier respectively;

and taking the assumption management label corresponding to the highest proximity in the proximity and representing the carrier characterization by the first center as a target assumption management label matched with the real medical management training text.

In some embodiments, the estimating the false medical management training text by the management tag determination model to be calibrated, and determining the second cost of the false medical management training text and the second target tag according to the estimated second estimated tag includes:

Extracting a text representation carrier of the fake medical management training text through the management label determining model to be calibrated, and estimating according to the text representation carrier of the fake medical management training text to obtain second credible coefficients of the fake medical management training text corresponding to all label types;

acquiring at least one true medical management training text used for deploying the false medical management training text, and determining a target assumption management label corresponding to the false medical management training text according to the target assumption management label matched with the at least one true medical management training text;

determining a second initial cost of the false medical management training text corresponding to the target assumption management label according to the second credibility coefficient and the target assumption management label corresponding to the false medical management training text;

determining a second assumed trusted coefficient of the false medical management training text corresponding to the rest label types except the target assumed management label according to the second trusted coefficient;

determining a second hypothesized cost of the pseudo medical management training text corresponding to the associated target reality management tag according to the second hypothesized trusted coefficient;

And determining a second cost of the fake medical management training text and a second target label according to the second initial cost and the second assumed cost.

In some embodiments, the determining, based on the second confidence coefficient, the second hypothesized confidence coefficient for the rest of the tag types other than the target hypothesized management tag for the sham medical management training text includes:

obtaining a coding result corresponding to the false medical management training text based on a target reality management tag matched with the false medical management training text;

performing reverse calculation on the coding result corresponding to the false medical management training text to obtain a reverse coding result corresponding to the false medical management training text;

determining a first assumed trusted coefficient of the true medical management training text corresponding to the rest tag types except the target reality management tag according to the second trusted coefficient and the reverse coding result corresponding to the false medical management training text;

said determining, from said second hypothesized confidence coefficient, a second hypothesized cost for said sham medical management training text corresponding to an associated target reality management tag, comprising:

Acquiring a target reality management tag matched with the false medical management training text;

and determining a second assumption cost of the fake medical management training text corresponding to the target reality management label according to the second assumption credibility coefficient and the reality management label indication corresponding to the target reality management label.

In some embodiments, the obtaining the target reality management tag matching the false medical management training text comprises:

determining second center representation carriers respectively corresponding to the reality management tags in the carrier set, wherein the second center representation carriers are center representation carriers corresponding to the reality management tags in current round adjustment;

determining the proximity of the text representing carrier of the fake medical management training texts and each second center representing carrier respectively;

and taking the second center representing carrier representation reality management label corresponding to the highest proximity in each proximity as a target reality management label matched with the false medical management training text.

In some embodiments, the obtaining the target hypothesis management label matched with the medical management update training text based on the error between the text representation carrier and each target center representation carrier includes:

Determining the proximity between the text representation carrier and each target center representation carrier;

and taking the assumption management label corresponding to the highest proximity in each proximity and representing the carrier characterization by the target center as the target assumption management label matched with the medical management update training text.

In another aspect, an embodiment of the present application provides a user portrait analysis device, including:

the model calling module is used for acquiring a plurality of medical management texts of a target user, and respectively inputting the medical management texts of the plurality of users into a management tag determination model after the medical management texts are calibrated in advance;

the carrier extraction module is used for extracting text representation carriers of the medical management texts through the management label determination model for each medical management text, and obtaining medical management labels of the medical management texts based on the text representation carriers;

the portrait construction module is used for constructing a user portrait of the target user through the medical management labels corresponding to the medical management texts;

The model adjustment module is used for acquiring a medical management update training text when the medical management tag is updated, and the update management tag corresponding to the medical management update training text is different from the existing reality management tag;

In a third aspect, embodiments of the present application provide a server comprising a memory and a processor, the memory storing a computer program executable on the processor, the processor implementing the steps of the method described above when the program is executed.

The technical scheme of the application at least comprises the following beneficial effects:

According to the user portrait analysis method, the device and the server based on the medical management labels, the medical management texts of target users are processed through the management label determining model to obtain a plurality of medical management labels, each medical management label forms a user portrait of the target users, accuracy and efficiency are high, in the execution process, iteration is carried out on the model through the true medical management training texts and the false medical management training texts together to obtain the management label determining model, when the medical management update training texts are obtained and the update management labels do not belong to the existing real management labels, text representing carriers are extracted through the management label determining model which is completed through adjustment, the text representing carriers are compared with target center representing carriers of all the assumption management labels which are obtained through adjustment, then the target assumption management labels corresponding to the medical management update training texts are determined according to errors obtained through comparison, and the target assumption management labels are used as update management labels, so that management label extraction can be carried out on medical record texts of the update management labels. When a new management tag type appears, the existing management tag determination model is not repeatedly calibrated, but one assumption management tag reserved in advance is used as the new tag type, so that the iteration speed of the management tag determination model can be faster. Further, when a new management label type appears, the corresponding assumption management label is correspondingly updated, so that the existing label type distribution condition is reserved, and the existing medical record text recognition is not influenced.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application as claimed.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

Fig. 1 is a schematic implementation flow chart of a user portrait analysis method based on a medical management tag according to an embodiment of the present application.

Fig. 2 is a schematic diagram of a composition structure of a user portrait analysis device according to an embodiment of the present application.

Fig. 3 is a schematic diagram of a hardware entity of a server according to an embodiment of the present application.

Detailed Description

The technical solution of the present application will be further elaborated with reference to the accompanying drawings and examples, which should not be construed as limiting the application, but all other embodiments which can be obtained by one skilled in the art without making inventive efforts are within the scope of protection of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict. The term "first/second/third" is merely to distinguish similar objects and does not represent a particular ordering of objects, it being understood that the "first/second/third" may be interchanged with a particular order or precedence, as allowed, to enable embodiments of the application described herein to be implemented in other than those illustrated or described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing the application only and is not intended to be limiting of the application.

The embodiment of the application provides a user portrait analysis method based on a medical management label, which can be executed by a processor of a server. The server type is not limited herein.

Fig. 1 is a schematic implementation flow chart of a user portrait analysis method based on a medical management tag according to an embodiment of the present application, as shown in fig. 1, the method includes steps 110 to 130 as follows:

step 110, acquiring a plurality of medical management texts of the target user, and respectively inputting the medical management texts of the plurality of users into a management tag determination model after the adjustment in advance.

The medical management text is, for example, a health detection text such as a physical examination report, or may be a medical history text in which health information of the target user such as index check information, diagnosis result information, medical advice information, notes, and the like are recorded. It can be understood that the embodiment of the application does not perform detection analysis on the data to obtain a disease diagnosis result, and only performs label mining on the existing information (such as diagnosis information and examination information) to obtain a label identification result of the medical management text corresponding to the target user. One target user corresponds to a plurality of medical management texts, for example, medical management texts of different detection times corresponding to the same detection item, or medical management texts corresponding to a plurality of different detection items at the same time. The management label determining model is used for analyzing and identifying the medical management text input into the management label determining model to obtain a medical management label corresponding to the medical management text, the medical management label is used for representing global summary information of the corresponding medical management text, for example, the medical management label can be represented by numbers, letters or combinations thereof, for example, A, B, A and B1, and the corresponding label meaning is for example, A-bone abnormality; b-blood type abnormalities; a1-cervical vertebra abnormality of bones; b1-blood hyperlipoidemia, etc., of course, under different demands, the characterization information, the division granularity and the setting rules of the medical management labels may be different, for example, the division modes of the plurality of medical management texts are different, and the corresponding medical management labels are different. The management tag determination model is a machine learning model, which can be constructed by different operators, for example, can comprise a plurality of feature extraction operators, classification operators and the like, can be adaptively trained by adopting a passing neural network architecture, and can be obtained after parameters are adjusted, for example, the model can be Bert.

Step 120, for each medical management text, extracting a text representation carrier of the medical management text through a management label determination model, and obtaining a medical management label of the medical management text based on the text representation carrier.

The process of the management label determining model extracting the text representing carrier of the medical management text is a process of mining the characteristics of the medical management text, and the text representing carrier is characteristic information of the medical management text, such as text vectors and text tensors. The medical management label of the medical management text can be obtained by performing operations such as full connection classification on the text representation carrier.

And 130, constructing a user portrait of the target user through the medical management labels corresponding to the medical management texts.

Each medical management text of the target user corresponds to one medical management tag, and one medical management tag represents the description information of the health state of the target user, so that the plurality of medical management tags can describe the user portrait of the target user, and the user portrait is the description of the health state of the target user.

The process of generating and calibrating the management tag determination model is described below, and in particular, the management tag determination model is a management tag determination model obtained by iterating a real medical management training text and a fake medical management training text together, and the management tag determination model can estimate an assumed management tag and a real management tag. The true medical management training text represents the training text which exists actually, and the false medical management training text represents the fictitious training text. The management tag determination model is not invariable, and as new detection information appears, the model needs to be updated to be capable of mining new medical management tags. The assumed management tag cannot represent tag information and has no exact tag meaning, so that the assumed management tag can be combined with a tag type with objective tag directivity when a model is updated, and the assumed management tag is taken as the new tag type. Each hypothetical management tag exists in a set of carriers, which is the space in which the feature exists, being replaced by a new tag type when it appears. When the management tag determination model is adjusted, the existing real medical management training text and the fictitious fake medical management training text are used at the same time, and generalization of the management tag determination model to the medical text can be improved due to the existence of the fake medical management training text.

The dummy medical management training text may be obtained from existing medical training text or directly fictitious, e.g., a dummy medical management training text may be transformed from a number of real medical management training texts (e.g., to enhance descriptive characters in the real medical management training text, such as prune, change, add interfering characters).

The obtained management tag determination model presets a plurality of reality management tags and a plurality of assumption management tags. The management tag determination model can estimate the real medical management training text, and the estimated result represents the corresponding real management tag. For the false medical management training text, the management tag determination model can also estimate the false medical management training text to obtain the corresponding tag type. When the adjustment is completed, the obtained management label determines that the assumption management label which can be identified by the model is used for being replaced by an update management label when the model label is updated, and the assumption management label is used as an update label type.

The specific updating process comprises the following steps:

step 210, acquiring medical management update training text.

The medical management update training text is updated medical management text, and the corresponding update management label is different from the existing reality management label. The update management label is a medical management label which needs to be updated, and the updated model is expected to have the capability of identifying the update management label, namely the actual management label is the actual medical management label.

According to the embodiment of the application, the medical management update training text is obtained to iterate the adjusted management label determining model, so that the management label determining model can estimate the updated management label.

When the server acquires the medical management update training text, the server acquires the update management label corresponding to the medical management update training text.

Step 220, inputting the medical management update training text into the management label determination model, and extracting a text representation carrier of the medical management update training text based on the management label determination model.

When the adjusted management tag determination model can identify each assumed management tag, the adjusted management tag determination model is iterated based on the medical management update training text without repeatedly adjusting the management tag determination model, and then the management tag determination model can estimate the updated management tag. The medical management update training text is used for iterating the adjusted management label determining model, and the assumption management label reserved in advance is used as an update management label according to the medical management update training text.

For example, the server propagates the medical management update training text into the management tag determination model, and extracts a text representation carrier of the medical management update training text based on the management tag determination model.

Step 230, determining target center representing carriers corresponding to each hypothesis management label in the carrier set, and obtaining the target hypothesis management label matched with the medical management update training text according to errors between the text representing carriers and each target center representing carrier.

The center representing carrier corresponding to the assumed management tag is the cluster center of the corresponding assumed management tag in the carrier set, the target center representing carrier corresponding to each assumed management tag is obtained according to multiple rounds of adjustment, and the real management tag also comprises the corresponding center representing carrier which is the cluster center of the corresponding real management tag in the carrier set. The target center representation carrier corresponding to each label type can be obtained according to the text representation carriers of a plurality of medical training texts corresponding to the label types. In the adjustment process of the management tag determination model, each pre-deployed tag type generates a cluster center, and the management tag determination model predicts the text representation carrier of the medical management training text corresponding to each tag type around the corresponding cluster center, so that the management tag determination model distributes a larger estimated credibility coefficient (namely the probability corresponding to the cluster center tag type) to the tag type corresponding to the medical management training text.

When the text of the medical management training text indicates that the carrier is mapped into the carrier set (can be understood as a feature distribution space), the server determines whether the carrier is in the range of the carrier set corresponding to a label type through the position of the carrier and the distance of each cluster center, so that whether the medical management training text corresponds to the label type or not is determined. After acquiring text representation carriers of a plurality of medical management update training texts, aiming at each medical management update training text, a server represents the position of the carrier in a carrier set according to the text representation carrier of the medical management update training text and the position of a target center representation carrier corresponding to each assumption management label respectively in the carrier set, determines which assumption management label the text representation carrier of the medical management update training text has the smallest cluster center distance with through errors of the two positions, and obtains a target assumption management label matched with the medical management update training text. The server obtains a target assumption management label matched with the medical management update training text according to errors between the text representation carrier and each center representation carrier, and specifically comprises the following steps: based on the proximity of the text representing carrier to each target center representing carrier, the assumption management label with the greatest proximity (i.e., the smallest distance) is determined as the target assumption management label matching the medical management update training text. In a possible case, for a plurality of medical management update training texts belonging to the same update management label, the server may determine the proximity of the text representing carrier of each medical management update training text to the target center representing carrier of each assumption management label, and perform weighted summation or average calculation to obtain the adjustment proximity between the target center representing carrier of each assumption management label, and use the assumption management label which is the most similar as the target assumption management label matched with the medical management update training text. Or the text representing carriers of the medical management update training texts can be extracted respectively, and the text representing carriers of the medical management update training texts form carrier distribution information in a carrier set. For each hypothesized management tag, the server also determines carrier distribution information of all the medical management training texts of each hypothesized management tag, then calculates errors between the carrier distribution information of the updated management tag and the carrier distribution information of each hypothesized management tag (e.g., calculates a similarity between the carrier distribution information, which may be based on the proximity of the text representing the carrier of each medical management update training text of the updated management tag to the text representing the carrier of all the medical management training texts of each hypothesized management tag, and then weight sums) to determine which hypothesized management tag is more similar to the updated management tag, and regards the most similar hypothesized management tag as the target hypothesized management tag that matches the medical management update training text.

Step 240, taking the target assumption management label as an update management label.

In this way, the management tag determination model may make predictions about updated management tags. After the target assumption management label matched with the medical management update training text is obtained, the target assumption management label is corresponding to the update management label corresponding to the medical management update training text. For example, the server takes the target assumption management label as the update management label, so that repeated adjustment of the management label determination model is not performed, and meanwhile, the update management label can be estimated.

According to the user portrait analysis method based on the medical management label, iteration is carried out on the model through the true medical management training text and the false medical management training text, a management label determining model is obtained, when the medical management updating training text is obtained and the updating management label does not belong to the existing real management label, a text representing carrier is extracted through the management label determining model after adjustment, the text representing carrier is compared with target center representing carriers of all the assumed management labels obtained after adjustment, then a target assumption management label corresponding to the medical management updating training text is determined according to errors obtained through comparison, the target assumption management label is used as the updating management label, and therefore management label extraction can be carried out on medical record text of the updating management label. When a new management tag type appears, the existing management tag determination model is not repeatedly calibrated, but one assumption management tag reserved in advance is used as the new tag type, so that the iteration speed of the management tag determination model can be faster. Further, when a new management label type appears, the corresponding assumption management label is correspondingly updated, so that the existing label type distribution condition is reserved, and the existing medical record text recognition is not influenced.

It can be seen that the above model updating process is performed by adopting the concept of continuous learning (Continual Learning), and then the process of determining how the model iterates through the true medical management training text and the false medical management training text by using the management tag is described later.

The method for obtaining the management label determination model obtained by iterating the real medical management training text and the fake medical management training text together specifically comprises the following steps:

in step 310, a true medical management training text and a false medical management training text are obtained.

The false medical management training text is obtained by transformation according to the true medical management training text. And repeatedly adjusting the management label determining model according to the real medical management training text and the false medical management training text, wherein the real medical management training text and the false medical management training text are taken as samples in each round of adjustment. In each round of adjustment, a plurality of real medical management training texts are input, the false medical management training texts are obtained according to random transformation of the plurality of real medical management training texts, the management label determination model to be adjusted is adjusted repeatedly, and finally the adjusted management label determination model is obtained. According to the management label determining model obtained by iterating the real medical management training text and the fake medical management training text, the reality management label and the assumption management label can be estimated at the same time.

Step 320, estimating the real medical management training text based on the management label determining model to be calibrated, and determining a first cost of the real medical management training text and a first target label according to the estimated first estimated label, wherein the first target label comprises a target reality management label corresponding to the real medical management training text and a target assumption management label matched with the real medical management training text.

For the real medical management training text, the server determines a model based on the management label to be calibrated, extracts a text representation carrier of the real medical management training text, and predicts according to the text representation carrier to obtain a predicted result of which real management label the real medical management training text corresponds to, wherein the predicted result is a credibility coefficient of the real medical management training text corresponding to each label type respectively, and the credibility coefficient can be probability or confidence level.

In the embodiment of the application, the estimated result obtained by estimating the true medical management training text is regarded as a first estimated label. The first predictive label is, for example, in the form of a vector, the number of constituent elements of which (i.e., the vector dimension) is the same as the number of types of labels deployed in advance. For easy understanding, the following illustrates that if m real management tags and n assumed management tags are included, for the inputted real medical management training text a, the server extracts the text representation carrier Va of the real medical management training text a based on the management tag determination model to be calibrated, and predicts according to the text representation carrier Va to obtain a first predicted tag as (J) ₁₁ ，J ₁₂ …J _1m ，J ₂₁ …J _2n )。J ₁₁ ，J ₁₂ …J _1m Whether or not the representative real medical management training texts a each correspond to a corresponding real management tag, such as J, as compared with each real management tag ₁₁ Equal to 0, the representative real medical management training text a is not the reality management label 1; j (J) ₁₁ For 1, the representative real medical management training text a is a real management tag 1.J (J) ₂₁ …J _2n Whether or not the representative genuine medical management training texts a each correspond to a corresponding assumption management label as compared with the respective assumption management label.

For the real medical management training text, the server determines the real management label output by the model for prediction and the target real management label actually corresponding to the real medical management training text based on the management label to be calibrated, and determines the cost of the real medical management training text corresponding to the target real management label. For example, for the real medical management training text a, according to the real management label actually corresponding to the first estimated label and the real medical management training text a, the cost of the real medical management training text corresponding to the target real management label can be determined.

The embodiment of the application also deploys the assumption management label in advance, the assumption management label does not exist in any medical management training text in essence, the true medical management training text is required to be carried in to determine the cost of the assumption management label, but the true medical management training text corresponds to one real management label, and when the error between the true medical management training text and the assumption management label is determined, the corresponding real management label is required to be disregarded, so that disturbance is prevented. Then, the server calculates the estimated result of the other label types based on the estimated result obtained by estimating the real medical management training text and the target reality management label corresponding to the real medical management training text. And according to the estimated results corresponding to the types of the rest labels, the true medical management training text is assumed to correspond to an assumed management label to determine the cost, so that the true medical management training text is brought into the adjustment process of the assumed management label. In this way, the server determines that one hypothesis label type is a target hypothesis label type matched with the true medical management training text in the hypothesis label types, and then determines the cost of the true medical management training text corresponding to the target hypothesis management label according to the estimated result of the rest label types corresponding to the fictive true medical management training text and the target hypothesis label type matched with the true medical management training text. Then, for the true medical management training text, the server determines a model based on the management label to be calibrated, determines the cost of the true medical management training text corresponding to the corresponding target reality management label according to the estimated first estimated label, and determines the cost of the true medical management training text corresponding to the matched target assumption management label according to the first estimated label, and establishes the first cost of the true medical management training text and the first target label together.

Step 330, estimating the false medical management training text based on the management label determination model to be calibrated, and determining a second cost of the false medical management training text and a second target label according to the estimated second estimated label, wherein the second target label comprises a target reality management label and a target assumption management label which are respectively matched with the false medical management training text.

If the hypothetical management tags were calibrated only with the genuine medical management training texts, the distribution of the hypothetical management tags in the carrier set would depend only on a small number of samples, and the predictions for the medical record management texts would be inaccurate. Based on the above, the embodiment of the application also adjusts the management label determination model to be adjusted through the fictitious and fake medical management training text. For the fake medical management training texts, the server determines a model based on the management tags to be calibrated, extracts text representing carriers of the fake medical management training texts, and predicts according to the text representing carriers to obtain a prediction result of which tag type the fake medical management training texts correspond to. If the false medical management training text is not true and does not have a corresponding actual tag type, the server firstly acquires the target assumption management tag matched with the false medical management training text and determines the cost of the false medical management training text corresponding to the target assumption management tag corresponding to the false. The target assumption management label matched with the false medical management training text is obtained according to target reality management labels corresponding to a plurality of true medical management training texts used for establishing the false medical management training text. Optionally, the server obtains the target hypothesis management tag matching the true medical management training text at step 320, and the server obtains the target hypothesis management tag matching the false medical management training text by the target hypothesis management tag matching the true medical management training text. In addition, because the false medical management training text does not have an actual tag type substantially, the false medical management training text needs to be brought into a cost acquisition process of the real management tag, and after the false corresponding target assumption management tag of the false medical management training text is determined, the server disregards the false corresponding target assumption management tag, and the false medical management training text does not correspond to the target assumption management tag and corresponds to the estimated result of the other tag types. And through the estimated results corresponding to the types of the rest labels, the false medical management training text is assumed to correspond to a real management label, so that cost determination is carried out, and the false medical management training text is brought into the adjustment process of the real management label. And the server selects one real management label from the real management labels to be regarded as a target real management label matched with the false medical management training text. The cost of the pseudo medical management training text corresponding to the target reality management label can be obtained through the estimated result of the pseudo medical management training text corresponding to the other label types and the target reality management label matched with the pseudo medical management training text. In this way, for the fake medical management training text, the server determines a model based on the management label to be calibrated, determines the cost of the fake medical management training text corresponding to the matched target assumption management label according to the estimated second estimated label, and determines the cost of the fake medical management training text corresponding to the matched target reality management label according to the estimated second label, so as to generate the first cost of the fake medical management training text and the first cost of the second target label together.

And 340, generating a target quality evaluation algorithm based on the first cost and the second cost, and repeatedly calibrating the management tag determination model to be calibrated through the target quality evaluation algorithm until the model converges to obtain the calibrated management tag determination model.

For example, a target quality assessment algorithm is generated by the first cost of the real medical management training text and the second cost of the fake medical management training text, and the model is repeatedly calibrated by the target quality assessment algorithm to be calibrated. The quality assessment algorithm is an algorithmic function that evaluates the performance of the model, e.g., cross entropy function, log likelihood function. The first cost and the second cost can be added to form a target quality assessment algorithm

According to the embodiment of the application, the real medical management training text is acquired, the fake medical management training text is established, the model is adjusted based on the real medical management training text and the fake medical management training text, the first cost corresponding to the real medical management training text and the second cost corresponding to the fake medical management training text are respectively acquired, the target quality evaluation algorithm is determined through the first cost and the second cost, the target quality evaluation algorithm can monitor the adjustment process of the management tag determination model, the adjusted management tag determination model can accurately estimate all real management tags, and meanwhile, the estimation of the assumed management tag is completed on the premise that the estimation of the real management tag is not influenced.

Optionally, the process of repeatedly calibrating the management tag determination model to be calibrated by the target quality evaluation algorithm includes, for example: and determining a model by the management label to be calibrated through a target quality evaluation algorithm, calculating a model change vector corresponding to the current round of calibration when each round of calibration is completed, and iterating the center representation carrier corresponding to each label type in the carrier set according to the model change vector. The center representation carrier corresponding to each label type obtained in the last round of adjustment is the center representation carrier of each label type.

The server performs multi-round adjustment on the management tag determination model to be adjusted through a target quality evaluation algorithm. When each round of adjustment is completed, the server adjusts the corresponding model change vector (representing gradient) of the round of adjustment, and determines the model to be adjusted according to the model change vector, for example, back propagation. The server iterates the central representation carrier corresponding to each label type in the carrier set, and the distribution position of the cluster center of each label type in each round of adjustment is obtained by iterating the cluster center according to the previous round of adjustment. The carrier is represented by the center of each label type, and after multiple rounds of adjustment, the cluster center of each label type becomes stable.

And adding an imaginary sample at the beginning of model adjustment so as to reserve a blank for updating the label type, wherein the imaginary sample does not have a corresponding medical management training text, and a new quality evaluation algorithm is required to be established to support the assumed management label during model adjustment. Therefore, the embodiment of the application provides a new quality evaluation algorithm which can calibrate the true medical management training text and the false medical management training text simultaneously. The quality evaluation algorithm of the management label determination model comprises a first price algorithm obtained according to the real medical management training text and a second cost algorithm obtained according to the fake medical management training text.

Then, for the first price algorithm, estimating the real medical management training text based on the management label determination model to be calibrated, and determining the first cost of the real medical management training text and the first target label according to the estimated first estimated label includes: extracting a text representation carrier of a true medical management training text based on a management label determining model to be calibrated, and estimating according to the text representation carrier of the true medical management training text to obtain a first credible coefficient of the true medical management training text corresponding to all label types; determining a first initial cost of the real medical management training text corresponding to the target reality management label through the first credibility coefficient and the target reality management label corresponding to the real medical management training text; determining a first assumed trusted coefficient of the real medical management training text corresponding to the rest label types except the target reality management label through the first trusted coefficient; determining that the true medical management training text corresponds to the associated first assumed cost according to the first assumed trusted coefficient; and determining the first cost of the real medical management training text and the first target label according to the first initial cost and the first assumed cost. For the true medical management training text, the server determines a model based on the management label to be calibrated, extracts a text representation carrier of the true medical management training text, and predicts according to the text representation carrier to obtain a first credible coefficient of the true medical management training text corresponding to all label types, wherein all label types comprise all real management labels and all assumed management labels. The first credibility coefficient is the estimated result of the real medical management training text. The real management label corresponding to the real medical management training text is regarded as a target real management label, and the target real management label can be determined in advance. The server may obtain, through the first trust coefficient and the target reality management tag, a cost of the real medical management training text corresponding to the target reality management tag, i.e., a first initial cost.

With the first confidence coefficient, the server may determine a first hypothetical confidence coefficient for the real medical management training text corresponding to the remaining tag types other than the target reality management tag. For example, the first confidence coefficient is x, and the true medical management training text corresponds to the first assumed confidence coefficients of the remaining tag types except for the target reality management tag of 1-x.

Optionally, determining, by the first confidence coefficient, a first hypothesized confidence coefficient that the real medical management training text corresponds to a remaining tag type other than the target reality management tag may include: obtaining a coding result corresponding to the real medical management training text based on the target reality management label corresponding to the real medical management training text, wherein the number of elements of the coding result is the same as the number of preset label types; performing reverse calculation on the coding result corresponding to the real medical management training text to obtain a reverse coding result corresponding to the real medical management training text; and determining the first assumed trusted coefficient of the rest label types except the target reality management label according to the first trusted coefficient and the reverse coding result corresponding to the real medical management training text.

For example, the server obtains the encoding result corresponding to the real medical management training text according to the target reality management tag corresponding to the real medical management training text. The number of elements of the encoding result is the same as the number of preset tag types. In one example, the server performs one-time thermal encoding on the target reality management tag corresponding to the real medical management training text to obtain the encoded vector, which may be binarized, for example (1,1,1,0,0,0). The server then performs a reverse calculation on the encoding result (e.g., 1 becomes 0 and 0 becomes 1) to obtain a reverse encoding result. And determining a first presumptive trusted coefficient of the real medical management training text corresponding to the rest label types except the target reality management label according to the first trusted coefficient and the reverse coding result. For example, the first assumed trusted coefficient is a result obtained by performing tensor product calculation on the first trusted coefficient and the reverse coding result.

The adjustment of bringing the real medical management training text into the assumed management label is completed by acquiring the coding result corresponding to the real medical management training text and simulating the situation that the real medical management training text does not correspond to the target real management label and corresponds to the rest label types based on the reverse calculation of the coding result. After obtaining the first hypothesized confidence coefficient, the server determines a first hypothesized cost of the true medical management training text corresponding to the target hypothesized management tag according to the first hypothesized confidence coefficient.

Optionally, determining, according to the first assumed trusted coefficient, a first assumed cost of the real medical management training text corresponding to the associated target assumed management label may specifically include: acquiring a target assumption management label matched with the real medical management training text; and determining that the true medical management training text corresponds to the first assumption cost of the assumption management label according to the first assumption credibility coefficient and the assumption management label indication corresponding to the target assumption management label.

For example, the server first obtains the target assumption management label matched with the real medical management training text, and because the real medical management training text corresponds to the target reality management label and does not correspond to any assumption management label, the server selects one assumption management label from the assumption management labels to be regarded as the target assumption management label matched with the real medical management training text. In this way, it is determined that the true medical management training text corresponds to the first hypothesized cost of the hypothesized management tag based on the first hypothesized confidence coefficient and the hypothesized management tag indication corresponding to the target hypothesized management tag.

According to the target assumption management label matched with the true medical management training text and the assumption management label indication corresponding to the target assumption management label, the first assumption cost of the true medical management training text corresponding to the target assumption management label is obtained, and the problem that the assumption management label does not have medical management training text is solved. The server may then obtain a first cost of the real medical management training text and the first target label based on the first initial cost and the first assumed cost, the first cost being, for example, an addition of the first initial cost and the first assumed cost.

The first initial cost and the first assumption cost are calculated for the real medical management training text, the cost of the real medical management training text is obtained through fusion, and the prediction of the assumption management label is realized on the premise that the model keeps the existing performance. When the target assumption management label matched with the true medical management training text is acquired, the carrier can be represented according to the center of each assumption management label for evaluation. Optionally, obtaining a target hypothesis management tag matching the true medical management training text includes: determining a first center representing carrier corresponding to each assumption management label in the carrier set, wherein the first center representing carrier is the center representing carrier corresponding to each assumption management label in the current round adjustment; determining the proximity of a text representing carrier of a real medical management training text and each first center representing carrier respectively; the assumption management label corresponding to the highest proximity (i.e. the smallest distance) in the proximity and representing the carrier characterization by the first center is taken as a target assumption management label matched with the real medical management training text.

The server determines that the center corresponding to each assumed management label in the current round of adjustment represents the carrier in the carrier set, namely, the first center represents the carrier. And acquiring the proximity of the first center representation carrier corresponding to each hypothesized management label according to the extracted text representation carrier of the real medical management training text, and taking the hypothesized management label represented by the first center representation carrier corresponding to the highest proximity in the proximity as a target hypothesized management label matched with the real medical management training text.

The most similar assumption management labels are used as target assumption management labels according to errors of the text representing carriers in the real medical management training text in the carrier set and the center representing carriers of each assumption management label, and the model is higher in accuracy.

The following describes the acquisition of the second cost, optionally, estimating the false medical management training text based on the management label determination model to be calibrated, and determining the second cost of the false medical management training text and the second target label according to the estimated second estimated label, including: extracting a text representation carrier of a fake medical management training text based on a management label to be calibrated, and estimating according to the text representation carrier of the fake medical management training text to obtain second credibility coefficients of the fake medical management training text corresponding to all label types; acquiring at least one true medical management training text used for deploying the false medical management training text, and determining a target assumption management label corresponding to the false medical management training text according to the target assumption management label matched with the at least one true medical management training text; determining a second initial cost of the false medical management training text corresponding to the target assumption management label according to the second credibility coefficient and the target assumption management label corresponding to the false medical management training text; determining a second assumed trusted coefficient of the false medical management training text corresponding to the rest of label types except the target assumed management label according to the second trusted coefficient; determining a second hypothesized cost of the pseudo-medical management training text corresponding to the associated target reality management tag according to the second hypothesized trusted coefficient; and determining a second cost of the fake medical management training text and the second target label according to the second initial cost and the second assumed cost.

For the fake medical management training texts, the server determines a model based on the management tags to be calibrated, extracts text representing carriers of the fake medical management training texts, predicts according to the text representing carriers, outputs second trusted coefficients of the fake medical management training texts corresponding to all tag types, wherein all tag types comprise all real management tags and all assumed management tags, and the second trusted coefficients are predicted results of the fake medical management training texts.

Because the false medical management training text does not have a corresponding tag type substantially, a presumptive management tag matched with the false medical management training text needs to be determined as a target presumptive management tag corresponding to the false medical management training text. The server obtains at least one true medical management training text for deploying the false medical management training text, and determines a target assumption management label corresponding to the false medical management training text according to the target assumption management label matched with the at least one true medical management training text.

For example, the server uses the obtained target assumption management label matched with the real medical management training text as the target assumption management label corresponding to the false medical management training text according to the real medical management training text for deploying the false medical management training text. In this way, a second initial cost of the sham medical management training text corresponding to the target hypothesis management label may be determined based on the second confidence coefficient and the target hypothesis management label corresponding to the sham medical management training text. The simulated fake medical management training text is adopted to generate a second cost of the fake medical management training text and the second target label, and the simulated fake medical management training text is not falsely corresponding to the target assumption management label, but corresponds to the scenes of the rest label types, so that the fake medical management training text can be brought into the adjustment process of the real management label. Based on the second confidence coefficient, the server re-determines a second hypothesized confidence coefficient for the pseudo medical management training text corresponding to the remaining tag types other than the target hypothesized management tag. For example, if the second confidence coefficient is y, then the second hypothesized confidence coefficient for the rest of the tag types other than the target hypothesized management tag is 1-y for the sham medical management training text.

Alternatively, it is also understood that determining the second hypothesized trusted coefficients of the pseudo medical management training text corresponding to the remaining tag types other than the target hypothesized management tag based on the second trusted coefficients may include: obtaining a coding result corresponding to the false medical management training text according to the target reality management tag matched with the false medical management training text; performing reverse calculation on the coding result corresponding to the false medical management training text to obtain a reverse coding result corresponding to the false medical management training text; and determining the first assumed trusted coefficients of the rest label types except the target reality management labels according to the second trusted coefficients and the reverse coding results corresponding to the fake medical management training texts. For example, the server obtains the encoding result corresponding to the fake medical management training text based on the target hypothesis management tag of the fake correspondence of the fake medical management training text.

In this way, the server determines a second hypothetical trusted coefficient for the pseudo medical management training text corresponding to the remaining tag types other than the target hypothetical management tag based on the second trusted coefficient and the reverse encoding result. Thus, after obtaining the second hypothesized trusted coefficient, determining that the true medical management training text corresponds to the second hypothesized cost of the target hypothesized management tag based on the second hypothesized trusted coefficient.

Similarly, optionally, determining that the sham medical management training text corresponds to the second hypothesized cost of the target reality management tag in accordance with the second hypothesized confidence coefficient may include: acquiring a target reality management tag matched with a false medical management training text; and determining a second assumed cost of the fake medical management training text corresponding to the associated target reality management label according to the second assumed trusted coefficient and the reality management label indication corresponding to the target reality management label. The server firstly acquires a target reality management tag matched with the false medical management training text. Because the false medical management training text does not substantially correspond to the tag type, the server selects a real management tag from the real management tags as a target real management tag matched with the false medical management training text. In this way, the server determines a second hypothesized cost for the pseudo medical management training text corresponding to the target reality management tag based on the second hypothesized confidence coefficient and the reality management tag indication corresponding to the target reality management tag. The second assumption cost of the false medical management training text corresponding to the target reality management label is determined through the target reality management label matched with the false medical management training text and the reality management label indication corresponding to the target reality management label, so that the problem that the false medical management training text does not substantially correspond to the label type is solved. And finally, determining the second cost of the fake medical management training text and the second target label according to the second initial cost and the second assumed cost. Optionally, the second cost is a result of adding the second initial cost to the second hypothesized cost.

The second initial cost of the scene of which the fiction corresponds to the target assumption management label is determined for the fake medical management training text, and the second assumption cost of the scene of which the fiction corresponds to the target reality management label is simultaneously calculated, and the second initial cost is combined to serve as the cost of the fake medical management training text, so that the model can predict the assumption management label while maintaining the identification capability of the existing label type. When the target reality management labels matched with the false medical management training texts are acquired, the evaluation can be performed according to the center representation carrier of each reality management label. Thus, optionally, the process of obtaining a target reality management tag that matches the false medical management training text may include: determining second center representation carriers respectively corresponding to the reality management tags in the carrier set, wherein the second center representation carriers are center representation carriers corresponding to the reality management tags in the current round adjustment; determining the proximity of the text representing carrier of the fake medical management training texts and each second center representing carrier respectively; and taking the second center representing carrier representation reality management label corresponding to the highest proximity in each proximity as a target reality management label matched with the false medical management training text.

For example, the server determines the center-indicating carrier corresponding to each real management tag in the round adjustment among the carrier sets, and regards the center-indicating carrier as the second center-indicating carrier. The server obtains the proximity of the second center representing carrier corresponding to each real management label according to the extracted text representing carrier of the fake medical management training text, and takes the real management label represented by the second center representing carrier corresponding to the highest proximity in the proximity as a target real management label matched with the fake medical management training text.

The error between the text representing carrier in the fake medical management training text and the center representing carrier of each real management label in the carrier set takes the most similar real management label as the target real management label, so that the accuracy of model adjustment can be improved to the greatest extent. Thus, a second cost according to the false medical management training text is obtained, and a target quality evaluation algorithm can be obtained according to the first cost and the second cost.

And repeatedly adjusting the management tag determination model through a target quality evaluation algorithm, wherein when the management tag determination model to be adjusted is repeatedly adjusted according to the target quality evaluation algorithm, the process of iterating the model comprises iterating cluster centers of each tag type. After the adjusted management tag determination model is obtained, the management tag determination model can be iterated directly according to the updated training text for the updated tag type. Optionally, the process of obtaining the target hypothesis management label matched with the medical management update training text according to the error between the text representation carrier and each target center representation carrier may include: determining the proximity between the text representation carrier and each target center representation carrier; and taking the assumption management label of the target center representation carrier representation corresponding to the highest proximity in each proximity as a target assumption management label matched with the medical management update training text.

For example, the server determines in the set of bearers that the center of each hypothesized management tag represents the bearer, i.e., the cluster center of each hypothesized management tag that is tuned. The server acquires the proximity (the smaller the acquirable distance is, the closer the distance is) between the text representing carrier of the medical management update training text and the center representing carrier of each hypothesized management tag through the text representing carrier of the medical management update training text, and takes the hypothesized management tag represented by the center representing carrier corresponding to the highest proximity in each proximity as a target hypothesized management tag matched with the medical management update training text.

In the embodiment of the application, the number of the assumed management labels reserved in advance at the beginning of the model adjustment is fixed, and when the number of the label types is gradually increased, the reserved number may not meet the requirement any more, and at the moment, the management label determination model is adjusted again. In this way, the server takes the relevant data of the updated management label as a real management label, and then acquires new real medical management training texts and false medical management training texts again, and adjusts the management label determination model again.

As a detailed embodiment, the server determines a model for the management tag to be calibrated, acquires a true medical management training text and a false medical management training text, extracts a text representation carrier of the true medical management training text based on the management tag determination model to be calibrated, and predicts according to the text representation carrier of the true medical management training text to obtain a first trusted coefficient of the true medical management training text corresponding to all tag types.

In one aspect, the server determines, via the first confidence coefficient and the target reality management tag corresponding to the real medical management training text, a first initial cost of the real medical management training text corresponding to the target reality management tag. In the second aspect, the server obtains the coding result corresponding to the real medical management training text according to the target reality management label corresponding to the real medical management training text, and performs reverse calculation on the coding result corresponding to the real medical management training text to obtain a reverse coding result corresponding to the real medical management training text. And then determining the first assumed trusted coefficient of the rest label types except the target reality management label by the first trusted coefficient and the reverse coding result corresponding to the real medical management training text. Meanwhile, the server determines current center representation carriers corresponding to the assumption management labels respectively in the carrier set, and the text representation carriers of the true medical management training texts and the proximity of the current center representation carriers are determined respectively. In this way, the server takes the assumption management label of the current center representation carrier representation corresponding to the highest proximity of the proximity as the target assumption management label matched with the real medical management training text. The server then determines a first hypothesized cost for the true medical management training text corresponding to the target hypothesized management tag based on the first hypothesized confidence coefficient and the hypothesized management tag indication corresponding to the target hypothesized management tag. In this way, the server may obtain the first cost of the real medical management training text and the first target label according to the first initial cost and the first assumed cost. In addition, the server predicts the false medical management training text based on a management label determining model to be calibrated, and determines a second cost of the false medical management training text and a second target label according to a second estimated label obtained through prediction, wherein the second target label comprises a target reality management label and a target assumption management label which are matched with the false medical management training text respectively. And finally, the server generates a target quality evaluation algorithm according to the first cost and the second cost, and repeatedly adjusts the management tag determination model to be adjusted through the target quality evaluation algorithm until the model converges to obtain the adjusted management tag determination model. When the management label is required to be updated, the server acquires a medical management update training text, inputs the medical management update training text into a management label determining model, extracts text representing carriers of the medical management update training text based on the management label determining model, determines the proximity between the text representing carriers and each center representing carrier, and takes a presumption management label, which is represented by the center representing carrier and corresponds to the highest proximity in each proximity, as a target presumption management label matched with the medical management update training text. Therefore, the server takes the target assumption management label as the update management label, so that the management label determination model can estimate the update management label.

Based on the foregoing embodiments, the embodiments of the present application provide a user portrait analysis device, where each unit included in the device and each module included in each unit may be implemented by a processor in a computer device; of course, the method can also be realized by a specific logic circuit; in practice, the processor may be a central processing unit (Central Processing Unit, CPU), microprocessor (Microprocessor Unit, MPU), digital signal processor (Digital Signal Processor, DSP) or field programmable gate array (Field Programmable Gate Array, FPGA), etc.

Fig. 2 is a schematic diagram of a composition structure of a user portrait analysis device according to an embodiment of the present application, and as shown in fig. 2, a user portrait analysis device 200 includes:

the model invoking module 210 is configured to obtain a plurality of medical management texts of the target user, and input the medical management texts of the plurality of users into a management tag determination model after adjustment in advance;

a carrier extraction module 220, configured to extract, for each of the medical management texts, a text representation carrier of the medical management text by the management label determination model, and obtain a medical management label of the medical management text based on the text representation carrier;

A portrait construction module 230, configured to construct a user portrait of the target user according to the medical management labels corresponding to the plurality of medical management texts;

the model tuning module 240 is configured to obtain a medical management update training text when the medical management tag is updated, where an update management tag corresponding to the medical management update training text is different from the existing reality management tag; inputting the medical management update training text into the management tag determination model, and extracting a text representation carrier of the medical management update training text based on the management tag determination model; determining target center representation carriers respectively corresponding to each assumption management label in a carrier set, and obtaining target assumption management labels matched with the medical management update training texts according to errors between the text representation carriers and the target center representation carriers; and taking the target assumption management label as the update management label, wherein the management label determination model can estimate the update management label.

The description of the apparatus embodiments above is similar to that of the method embodiments above, with similar advantageous effects as the method embodiments. In some embodiments, the functions or modules included in the apparatus provided by the embodiments of the present application may be used to perform the methods described in the foregoing method embodiments, and for technical details that are not disclosed in the embodiments of the apparatus of the present application, reference should be made to the description of the embodiments of the method of the present application.

In the embodiment of the present application, if the user portrait analysis method based on the medical management tag is implemented in the form of a software function module, and is sold or used as a separate product, the user portrait analysis method may also be stored in a computer readable storage medium. Based on such understanding, the technical solution of the embodiments of the present application may be essentially or some of contributing to the related art may be embodied in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read Only Memory (ROM), a magnetic disk, an optical disk, or other various media capable of storing program codes. Thus, embodiments of the application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.

The embodiment of the application provides a server, which comprises a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor realizes part or all of the steps in the method when executing the program.

Embodiments of the present application provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs some or all of the steps of the above-described method. The computer readable storage medium may be transitory or non-transitory.

Embodiments of the present application provide a computer program comprising computer readable code which, when run in a computer device, causes a processor in the computer device to perform some or all of the steps for carrying out the above method.

Embodiments of the present application provide a computer program product comprising a non-transitory computer-readable storage medium storing a computer program which, when read and executed by a computer, performs some or all of the steps of the above-described method. The computer program product may be realized in particular by means of hardware, software or a combination thereof. In some embodiments, the computer program product is embodied as a computer storage medium, in other embodiments the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK), or the like.

It should be noted here that: the above description of various embodiments is intended to emphasize the differences between the various embodiments, the same or similar features being referred to each other. The above description of apparatus, storage medium, computer program and computer program product embodiments is similar to that of method embodiments described above, with similar advantageous effects as the method embodiments. For technical details not disclosed in the embodiments of the apparatus, the storage medium, the computer program and the computer program product of the present application, reference should be made to the description of the embodiments of the method of the present application.

Fig. 3 is a schematic diagram of a hardware entity of a server according to an embodiment of the present application, as shown in fig. 3, the hardware entity of the server 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program executable on the processor 1001, the processor 1001 implementing the steps in the method of any of the embodiments described above when the program is executed.

The memory 1002 stores a computer program executable on the processor, and the memory 1002 is configured to store instructions and applications executable by the processor 1001, and may also cache data (e.g., image data, audio data, voice communication data, and video communication data) to be processed or already processed by each module in the processor 1001 and the model training apparatus 1000, which may be implemented by a FLASH memory (FLASH) or a random access memory (Random Access Memory, RAM).

The processor 1001 executes a program to implement the user image analysis method based on the medical management tag according to any one of the above. The processor 1001 generally controls the overall operation of the server 1000.

Embodiments of the present application provide a computer storage medium storing one or more programs executable by one or more processors to implement the steps of the medical management label-based user portrait analysis method of any of the embodiments above.

It should be noted here that: the description of the storage medium and apparatus embodiments above is similar to that of the method embodiments described above, with similar benefits as the method embodiments. For technical details not disclosed in the embodiments of the storage medium and the apparatus of the present application, please refer to the description of the method embodiments of the present application. The processor may be at least one of a target application integrated circuit (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD), a programmable logic device (Programmable Logic Device, PLD), a field programmable gate array (Field Programmable Gate Array, FPGA), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, and a microprocessor. It will be appreciated that the electronic device implementing the above-mentioned processor function may be other, and embodiments of the present application are not limited in detail.

The computer storage medium/Memory may be a Read Only Memory (ROM), a programmable Read Only Memory (Programmable Read-Only Memory, PROM), an erasable programmable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable programmable Read Only Memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic random access Memory (Ferromagnetic Random Access Memory, FRAM), a Flash Memory (Flash Memory), a magnetic surface Memory, an optical disk, or a Read Only optical disk (Compact Disc Read-Only Memory, CD-ROM); but may also be various terminals such as mobile phones, computers, tablet devices, personal digital assistants, etc., that include one or any combination of the above-mentioned memories.

It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment" or "in the alternative" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in various embodiments of the present application, the sequence number of each step/process described above does not mean that the execution sequence of each step/process should be determined by its functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The foregoing embodiment numbers of the present application are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above described device embodiments are only illustrative, e.g. the division of the units is only one logical function division, and there may be other divisions in practice, such as: multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. In addition, the various components shown or discussed may be coupled or directly coupled or communicatively coupled to each other via some interface, whether indirectly coupled or communicatively coupled to devices or units, whether electrically, mechanically, or otherwise.

The units described above as separate components may or may not be physically separate, and components shown as units may or may not be physical units; can be located in one place or distributed to a plurality of network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may be separately used as one unit, or two or more units may be integrated in one unit; the integrated units may be implemented in hardware or in hardware plus software functional units.

Those of ordinary skill in the art will appreciate that: all or part of the steps for implementing the above method embodiments may be implemented by hardware related to program instructions, and the foregoing program may be stored in a computer readable storage medium, where the program, when executed, performs steps including the above method embodiments; and the aforementioned storage medium includes: a mobile storage device, a Read Only Memory (ROM), a magnetic disk or an optical disk, or the like, which can store program codes.

Alternatively, the above-described integrated units of the present application may be stored in a computer-readable storage medium if implemented in the form of software functional modules and sold or used as separate products. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the related art in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a removable storage device, a ROM, a magnetic disk, or an optical disk.

The foregoing is merely an embodiment of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think about changes or substitutions within the technical scope of the present application, and the changes and substitutions are intended to be covered by the scope of the present application.

Claims

1. A user image analysis method based on medical management labels, which is applied to a server, the method comprising:

2. The method according to claim 1, wherein the method further comprises:

3. The method according to claim 2, wherein the iteratively tuning the management label determination model to be tuned by the target quality assessment algorithm comprises:

4. The method of claim 3, wherein the determining, by the first confidence coefficient, that the genuine medical management training text corresponds to a first fictitious confidence coefficient for a remaining tag type other than the target reality management tag comprises:

5. The method according to claim 2, wherein the estimating the false medical management training text by the management tag determination model to be calibrated, and determining the second cost of the false medical management training text and the second target tag according to the estimated second estimated tag, includes:

6. The method of claim 5, wherein determining, based on the second confidence coefficient, a second hypothesized confidence coefficient for the pseudo medical management training text corresponding to a remaining tag type other than the target hypothesized management tag comprises:

7. The method of claim 6, wherein the obtaining a target reality management tag that matches the false medical management training text comprises:

8. The method of claim 1, wherein the deriving the target hypothesis management tag matching the medical management update training text based on the error between the text representation carrier and each target center representation carrier comprises:

9. A user portrait analysis device, comprising:

10. A server comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that the processor implements the steps of the method of any of claims 1 to 8 when the program is executed.