Medical Information Recommendation Method, Device, Electronic Device, and Readable Medium

Abstract

The present application provides a medical information recommendation method, apparatus, electronic device, and readable medium. The method includes: obtaining object information of a target object; dividing the object information into status information and attribute information according to the text content of the object information; generating an information weighted vector according to the status information and the attribute information, where the information weighted vector includes the degree of association between each character in the status information and the attribute information and the object status of the target object; determining medical information associated with the target object according to the information weighted vector, and pushing the medical information to the target object. This method can avoid the confusion of recommendation results caused by the mixing of status information and attribute information, make the determined medical information more in line with the object status of the target object, and improve the accuracy of the recommended medical information.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular, to a medical information recommendation method, apparatus, electronic device, and readable medium. Background Art

[0002] With the increasing maturity of Internet technology and big data technology, more and more people learn medical knowledge through the Internet. Correspondingly, many medical information sharing platforms or applications have emerged on the Internet, and these medical information sharing platforms provide users with channels to learn and understand the medical information they are concerned about.

[0003] In the related art, platforms or applications usually analyze according to the input information of users to determine the recommended information associated with the input information for display.

[0004] However, in the above manner, there are often too many or too few recommended information associated with the object information of the target object, resulting in inaccurate determined recommended information and reducing the accuracy and efficiency of the recommendation results. Summary of the Invention

[0005] Based on the above technical problems, the present application provides a medical information recommendation method, apparatus, electronic device, and readable medium, so as to avoid the confusion of recommendation results caused by the mixing of status information and attribute information, make the determined medical information more in line with the object status of the target object, and improve the accuracy of the recommended medical information.

[0006] Other features and advantages of the present application will become apparent through the following detailed description, or be learned in part through the practice of the present application.

[0007] According to one aspect of the embodiments of the present application, a medical information recommendation method is provided, including:

[0008] Obtain the object information of the target object;

[0009] Divide the object information into status information and attribute information according to the text content of the object information;

[0010] Generate an information weighted vector according to the status information and the attribute information, where the information weighted vector includes the degree of association between each character in the status information and the attribute information and the object status of the target object;

[0011] Determine the medical information associated with the target object according to the information weighted vector and push the medical information to the target object.

[0012] In some embodiments of the present application, based on the above technical solutions, dividing the object information into status information and attribute information according to the text content of the object information includes:

[0013] Using an information classification model to perform status information prediction on the object information to obtain the character probability distribution of the object information, where the character probability distribution is used to indicate the probability that each character in the object information is used to describe the object status of the target object;

[0014] According to the character probability distribution of the object information, combine the characters used to describe the object status of the target object into the status information and combine the remaining characters into the attribute information.

[0015] In some embodiments of the present application, based on the above technical solutions, generating an information weighted vector according to the status information and the attribute information includes:

[0016] Using the information classification model, according to the character probability distribution, calculate the attention feature value corresponding to each character in the object information to obtain the information weighted vector.

[0017] In some embodiments of the present application, based on the above technical solutions, determining the medical information associated with the target object according to the information weighted vector and pushing the medical information to the target object includes:

[0018] Using a status prediction model, according to the information weighted vector, determine the status medical information corresponding to the object status of the target object;

[0019] Using an attribute prediction model, according to the information weighted vector, determine the attribute medical information corresponding to the object attribute of the target object;

[0020] Push the status medical information and the attribute medical information to the target object.

[0021] In some embodiments of the present application, based on the above technical solutions, before using the information classification model to perform status information prediction on the object information to obtain the character probability distribution of the object information, the method further includes:

[0022] Obtain historical object information, where the historical object information includes status information and attribute information;

[0023] Label the status information and the attribute information in the historical object information to obtain classification training data;

[0024] According to the classification training data, train the classification model to be trained to obtain the information classification model.

[0025] In some embodiments of the present application, based on the above technical solutions, the historical object information further includes historical status recommendation information and historical attribute recommendation information; training the classification model to be trained based on the classification training data to obtain the information classification model includes:

[0026] Predicting the classification training data through the classification model to be trained to obtain a training weighted vector;

[0027] Predicting through the status model to be trained and the attribute model to be trained according to the training weighted vector to obtain status recommendation information and attribute recommendation information;

[0028] Adjusting the parameters of the classification model to be trained according to the status recommendation information and the historical status recommendation information to obtain the information classification model;

[0029] Adjusting the parameters of the status model to be trained according to the status recommendation information, the training weighted vector, the attribute recommendation information, and the historical status recommendation information to obtain the status prediction model;

[0030] Adjusting the parameters of the attribute model to be trained according to the attribute recommendation information, the training weighted vector, the status recommendation information, and the historical attribute recommendation information to obtain the attribute prediction model.

[0031] In some embodiments of the present application, based on the above technical solutions, adjusting the parameters of the status model to be trained according to the status recommendation information, the training weighted vector, the attribute recommendation information, and the historical status recommendation information to obtain the status prediction model includes:

[0032] Determining a status prediction distribution according to the status recommendation information and the training weighted vector;

[0033] Determining an attribute prediction distribution according to the attribute recommendation information and the training weighted vector;

[0034] Weighting the feature values in the status prediction distribution according to the attribute prediction distribution to obtain a status weighted distribution;

[0035] Adjusting the parameters of the status model to be trained according to the status weighted distribution and the historical status recommendation information to obtain the status prediction model.

[0036] According to one aspect of the embodiments of the present application, a medical information recommendation device is provided, including:

[0037] An information acquisition module, configured to acquire object information of a target object;

[0038] An information division module, configured to divide the object information into status information and attribute information according to the literal content of the object information;

[0039] A vector generation module, configured to generate an information weighted vector according to the status information and the attribute information, where the information weighted vector includes the degree of association between each character in the status information and the attribute information and the object status of the target object;

[0040] An information push module, configured to determine medical information associated with the target object according to the information weighted vector and push the medical information to the target object.

[0041] According to one aspect of the embodiments of the present application, an electronic device is provided, which includes: a processor; and a memory for storing executable instructions of the processor; wherein, the processor is configured to execute the medical information recommendation method in the above technical solution by executing the executable instructions.

[0042] According to one aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the medical information recommendation method in the above technical solution is implemented.

[0043] In the embodiments of the present application, when recommending medical information to a target object, the object information will first be divided into status information and attribute information, and then the relevance between the status information and the attribute information and the object status will be quantified through an information weighted vector. Finally, the medical information to be pushed will be determined according to the information weighted vector. By dividing the status information and the attribute information, and then using the information weighted vector to show the association relationship between these two types of information and the object status, it is possible to determine the medical information to be recommended according to the status information or attribute information that is most relevant to the current object status during the recommendation process, thereby avoiding the confusion of the recommendation results caused by the mixing of the status information and the attribute information, making the determined medical information more in line with the object status of the target object, and improving the accuracy of the recommended medical information.

[0044] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0046] In the accompanying drawings:

[0047] Figure 1 Schematically shows a schematic diagram of the exemplary system architecture of the technical solution of the present application in an application scenario;

[0048] Figure 2 Is a schematic flowchart of the core process of the medical information recommendation method in the present application;

[0049] Figure 3 Is a schematic flowchart of a medical information recommendation method in an embodiment of the present application;

[0050] Figure 4 Is a schematic structural diagram of the training process in an embodiment of the present application;

[0051] Figure 5 Is a schematic structural diagram of the training process in an embodiment of the present application;

[0052] Figure 6 Is a schematic structural diagram of the training process in an embodiment of the present application;

[0053] Figure 7 Schematically shows a block diagram of the composition of the medical information recommendation device in an embodiment of the present application;

[0054] Figure 8 Shows a schematic diagram of the structure of a computer system of an electronic device suitable for implementing the embodiments of the present application. Detailed implementation manners

[0055] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0056] In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present application. However, those skilled in the art will realize that the technical solutions of the present application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be used. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the present application.

[0057] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0058] The flowcharts shown in the drawings are only exemplary illustrations, and do not necessarily include all the content and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be decomposed, while some operations / steps can be combined or partially combined. Therefore, the actual execution order may change according to the actual situation.

[0059] The solution of this application can be applied to the information recommendation scenario, and specifically applied to the scenario of recommending medical information that conforms to the user's current state according to the description information input by the user and according to relevant historical information. For example, applying the solution of this application to an application related to life and health, the user can input content describing their own state into the application, such as examination indicators, symptom descriptions, self-feelings, and relevant medical records and case records. The solution of this application will classify the information according to the content of the input information to obtain, and confirm the information related to the user's current physical state, such as information related to possible diseases, or information related to physical constitution or health status, and then make further medical information recommendations according to the determined information. For example, the user may input a description of their physical symptoms. The solution of this application divides the input description information into information about manifested symptoms, such as headache, low back pain, etc., and information about physical constitution, such as easy fatigue, fear of cold, medical history for 2 years, etc. Then, according to these information, the medical information corresponding to the manifested symptoms and the medical information corresponding to the physical constitution are respectively determined, such as popular science information about possible related diseases and relevant information about how to improve the constitution of being easily fatigued and fearing cold.

[0060] The application scenario of the solution of this application is introduced below. Please refer to Figure 1 , Figure 1 which schematically shows an exemplary system architecture diagram of the technical solution of this application in an application scenario. As Figure 1As shown, this scenario includes a client 110, a business server 120, and a data server 130. Among them, the solution introduced in this application is deployed on the business server 120. The business server 120 provides a medical information recommendation service to the client 110. After the user logs in to the client 110, the user requests recommended medical information from the business server 120 and inputs description information related to their own physical condition as the basis for the recommendation through the client. The business server 120 will then determine the medical information related to the user's physical condition according to the description information sent by the client 110 according to the method introduced in this application and push it to the client 110. When needed, the business server 120 can also request authorization for other associated information from the client 110 and, after obtaining the user's authorization, obtain other medical-related records and information associated with the user from the data server 130, such as medical visit records, case records, and drug purchase records, etc.

[0061] In the above application scenario, the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. This is not restricted here. The terminal device can be a mobile phone, a computer, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, etc., but is not limited thereto. The terminal device and the server can be directly or indirectly connected through wired or wireless communication methods, which are not restricted in this application. The number of terminal devices and servers is also not restricted.

[0062] Next, the core process of the medical information recommendation method in this application will be introduced. Please refer to Figure 2 , Figure 2 is a schematic flowchart of the core process of the medical information recommendation method in this application. As Figure 2As shown, the solution of the present application will first generate time-sensitive content in step 210 according to the input content and in accordance with the time described in the content. Specifically, in the input object information, there will usually be current state information related to the user's current state and attribute information related to the persistent state or attributes of the user himself. The state information usually describes the manifestations that exist in a short period of time, that is, usually only appear in a relatively short time period, such as the symptoms brought about by some acute diseases, such as diarrhea and vomiting, etc., while the attribute information usually describes the manifestations that exist in a long time, that is, usually appear in a relatively long time period, such as weak constitution, long-term headache, liking cold food, etc. The state information is usually more relevant to the user's current state. For example, for a sick user, the state information is usually the information describing the disease, while the attribute information is usually more relevant to the user's long-term state or historical state, such as the user's physical constitution information. There is a mutual correlation between the state information and the attribute information. In step 220, attention feature extraction is performed on the state information and the attribute information, so as to extract the attribute information that is related to the user's state information and more relevant to the current state. The correlation between various information and the user's current state can be reflected in the feature vector output by the attention feature extraction, and it also includes the correlation between the attribute information and the state information, that is, the attribute information with a high degree of correlation with the state information will have a higher eigenvalue in the feature vector. Subsequently, in step 230, according to the feature vector output by the attention feature extraction and the prediction target to be performed, the weights of the state information and the attribute information in the feature vector can be adjusted, so as to strengthen the correlation between the state information and the attribute information and the prediction result. The prediction target is usually related to the content of the medical information to be recommended to the user. For example, recommend medical information related to the user's current disease state, or recommend medical information related to the user's physical constitution state. The prediction target can be selected by the user himself, or determined according to the input object information, or corresponding adjustments are made to all the prediction results that can be output. Finally, according to the adjusted feature vector and the mapping relationship between the feature vector and the relevant medical information, the medical information to be recommended is determined and pushed to the user.

[0063] The medical information recommendation method in the embodiments of the present application will be further introduced below. Please refer to Figure 3 , Figure 3 which is a schematic flowchart of a medical information recommendation method in an embodiment of the present application. This solution can be executed by the client or by the server. Below, taking a computing device such as the server of a medical information recommendation platform as the execution subject as an example for introduction, as Figure 3 shown, this medical information recommendation method at least includes steps S310 to S340, which are introduced in detail as follows:

[0064] Step S310: obtaining object information of the target object.

[0065] The target object may be a client using the recommendation platform. The recommendation platform may receive the description content input by the client as the object information, and may also obtain the corresponding historical information from the associated information platform according to the client's login information. The user usually triggers the recommendation platform to recommend medical information to the client through the client. It may also include the category to be recommended to the target object. For example, when the user triggers a medical recommendation, he or she may select the category of medical information he or she wants to see, such as information about possible related diseases or information about how to adjust the physical condition.

[0066] Step S320: dividing the object information into state information and attribute information according to the text content of the object information.

[0067] Specifically, the medical information recommendation device will analyze the text content of the object information and confirm whether the text content belongs to the state information describing the current state of the target object or the attribute information describing the attributes of the target object itself. For example, the input object information is "10 years of history of palpitations, which has worsened since 1976 and has dizziness. Today, he was admitted to the hospital due to chest tightness, palpitations, and syncope. The current symptoms are palpitations, dizziness, fatigue, chest tightness, chest and back pain, chills and cold limbs, pale red tongue, thin white fur, and weak pulse. The patient said that he usually has loose stools and likes to eat hot drinks." The medical information recommendation device can analyze the object information through semantic analysis or a preset mapping database. Semantic analysis can be performed using models such as neural networks, and the preset mapping database can establish a mapping relationship between various types of text belonging to state information or attribute information through the summary of big data rules and expert experience, thereby dividing the object information. For example, the state information divided from the above object information can be "palpitations, dizziness, fatigue, chest tightness, chest and back pain, chills and cold limbs, pale red tongue, thin white fur, weak pulse, chest tightness, panic, and fainting", and the remaining information is divided into attribute information.

[0068] In one embodiment of the present application, based on the above technical solution, the above step S320, dividing the object information into state information and attribute information according to the text content of the object information, specifically includes the following steps:

[0069] By using an information classification model, state information prediction is performed on the object information to obtain a character probability distribution of the object information, wherein the character probability distribution is used to indicate the probability that each character in the object information is used to describe the object state of the target object;

[0070] According to the character probability distribution of the object information, characters for describing the object state of the target object are combined into the state information and the remaining characters are combined into the attribute information.

[0071] Specifically, the information classification model can be a pointer neural network based on Encoder-Decoder. Through the trained information classification model, the state information prediction can be performed on the object information. Specifically, the information classification model calculates a corresponding character probability for each character in the object information, and this probability is used to represent the probability that the corresponding character is used to describe the object state of the target object. The vectors composed of all characters form the character probability distribution. According to this character probability distribution, the characters with probabilities higher than the probability threshold are determined as the characters that will be used to describe the object state of the target object, and these characters are combined into state information, while the remaining characters, that is, the characters with probabilities lower than the probability threshold, are determined as the characters that will be used to describe the object attributes of the target object, and these characters are combined into attribute information.

[0072] Step S330, generate an information weighted vector according to the state information and the attribute information, where the information weighted vector includes the degree of association between each character in the state information and the attribute information and the object state of the target object.

[0073] In this embodiment, the medical information recommendation device determines the weight values of each character in the state information and the attribute information according to the association relationship between the state information and the attribute information and the object state of the target object. Specifically, the higher the degree of association with the object state, the higher the weight value. The weight value can be determined according to a preset weight mapping relationship, for example, determined by a preset weight algorithm or a weight distribution interval. In one embodiment, the medical information recommendation device first makes a preliminary prediction of the current state related to the target object according to the state information and the attribute information, such as predicting the possible disease state or the possible physical state of the current object, and then further processes the state information and the attribute information according to the predicted disease state or physical state. Or, in one embodiment, the medical information recommendation device infers the association relationship between the state information and the attribute information according to a preset mapping relationship. For example, there is a causal relationship or an interaction relationship between certain physical states and certain disease states. Then, according to these association relationships, the state information characters and attribute information characters with strong associations are given high weights, while the state information characters and attribute information characters with weak associations are given low weights.

[0074] In one embodiment of the present application, based on the above technical solution, in the above step S330, generating an information weighted vector according to the state information and the attribute information specifically includes the following steps:

[0075] Through the information classification model, according to the character probability distribution, calculate the attention feature value corresponding to each character in the object information to obtain the information weighted vector.

[0076] Specifically, according to the character probability distribution, the information classification model calculates the attention features for each character through the trained attention weighting algorithm, so as to obtain the information weighted vector. Each character probability can be converted into a corresponding weight value, and is calculated together with the corresponding vector in the attention weighting algorithm, so as to obtain the information weighted value corresponding to each character, and these values are combined into an information weighted vector.

[0077] Step S340: Determine the medical information associated with the target object according to the information weighted vector, and push the medical information to the target object.

[0078] The medical information recommendation device will pre - establish a mapping relationship between the information weighted vector and the medical information, which can be a trained model or a corresponding algorithm. Through this mapping relationship, the medical information associated with the target object can be determined according to the information weighted vector. Depending on the different purposes of recommendation, the medical information can be the medical information corresponding to the status information or the medical information corresponding to the attribute information. Subsequently, the medical information recommendation device will push the determined medical information to the client for the client to display to the user.

[0079] In an embodiment of the present application, based on the above - mentioned technical solution, the above - mentioned step S340, determining the medical information associated with the target object according to the information weighted vector and pushing the medical information to the target object, specifically includes the following steps:

[0080] Through the state prediction model, determine the status medical information corresponding to the object status of the target object according to the information weighted vector;

[0081] Through the attribute prediction model, determine the attribute medical information corresponding to the object attribute of the target object according to the information weighted vector;

[0082] Push the status medical information and the attribute medical information to the target object.

[0083] The state prediction model is a prediction model used to predict the medical information related to the object status of the target object. Input the information weighted vector into the state prediction model for calculation, so as to be able to determine the status medical information corresponding to the object status of the target object. Similarly, input the information weighted vector into the attribute prediction model for calculation, so as to be able to determine the attribute medical information corresponding to the object attribute of the target object. Then push the status medical information and the attribute medical information to the target object together.

[0084] In an embodiment of the present application, when recommending medical information to a target object, the object information is first divided into status information and attribute information, and then the relevance between the status information and attribute information and the object status is quantified through an information weighting vector. Finally, the medical information to be pushed is determined according to the information weighting vector. By dividing the status information and attribute information and then using the information weighting vector to show the correlation between these two types of information and the object status, it is possible to determine the medical information to be recommended based on the status information or attribute information that is most relevant to the current object status during the recommendation process, thereby avoiding the chaos of the recommendation results caused by the mixing of status information and attribute information, making the determined medical information more in line with the object status of the target object, and improving the accuracy of the recommended medical information.

[0085] In an embodiment of the present application, based on the above technical solution, before predicting the status information of the object information through an information classification model and obtaining the character probability distribution of the object information in the above step, the method of the present application further includes:

[0086] Obtain historical object information, where the historical object information includes status information and attribute information;

[0087] Label the status information and the attribute information in the historical object information to obtain classification training data;

[0088] Train the classification model to be trained according to the classification training data to obtain the information classification model.

[0089] For the convenience of introduction, please refer to Figure 4 , Figure 4 which is a schematic structural diagram of the training process in the embodiment of the present application. As Figure 4As shown, for the training data, the content of the past disease history and the content of the current medical treatment are divided. In order to make the model judge not by relying on the order of precedence, but by semantic features, multiple sentences of history and current predictions are extracted respectively, and the order is randomly shuffled to form a historical object information. Such as: "The history of palpitations is 10 years, which has worsened since 1976, with dizziness. Today, he was admitted to the hospital due to chest tightness, palpitations, and syncope. At present, he has palpitations, dizziness, fatigue, chest tightness, chest and back pain, fear of cold limbs, pale red tongue, thin white fur, and weak pulse. The patient said that he usually has loose stools and likes to eat hot drinks." Take the sentence as the unit, and randomly shuffle: "At present, he has palpitations, dizziness, fatigue, chest tightness, chest and back pain, fear of cold limbs, pale red tongue, thin white fur, and weak pulse. Today, he has chest tightness, palpitations, and syncope. Admission. The patient reported that he usually had loose stools and liked hot drinks. He had a history of palpitations for 10 years, which worsened in 1976, and he also had dizziness. "The state information and attribute information in the historical object information are annotated, and the symptom text close to the current time point is extracted as the result to be predicted by the model, which includes: "palpitations, dizziness, fatigue, chest tightness, chest and back pain, chills and cold limbs, pale red tongue, thin white fur, weak pulse, chest tightness, palpitations, syncope", thus obtaining the training text.

[0090] Next, the training text is input into the classification model to be trained. Usually, neither punctuation is required for input nor for generating results, but only the text characters of the symptoms need to be output. The role of the classification model to be trained is equivalent to copying and selecting the symptoms that are more relevant to the current visit from the original text. Figure 4 As shown in the figure, BiLSTM converts each input character into a hidden layer representation h. z is the attention vector, and z0 is a vector randomly initialized in the initial stage. i ,h) can get the attention probability distribution of each input character with a sum of 1. The subscript of the highest output probability is argmax(Attn(z i ,h)) characters as the input for the next moment. And so on, until the special end identifier is predicted <end>until

[0091] In one embodiment of the present application, based on the above technical solution, the historical object information further includes historical status recommendation information and historical attribute recommendation information. The above steps of training the classification model to be trained according to the classification training data to obtain the information classification model specifically include the following steps:

[0092] Predict the classification training data through the classification model to be trained to obtain a training weighted vector;

[0093] Predict through the status model to be trained and the attribute model to be trained according to the training weighted vector to obtain status recommendation information and attribute recommendation information;

[0094] Adjust the parameters of the classification model to be trained according to the status recommendation information and the historical status recommendation information to obtain the information classification model;

[0095] Adjust the parameters of the status model to be trained according to the status recommendation information, the training weighted vector, the attribute recommendation information, and the historical status recommendation information to obtain the status prediction model;

[0096] Adjust the parameters of the attribute model to be trained according to the attribute recommendation information, the training weighted vector, the status recommendation information, and the historical attribute recommendation information to obtain the attribute prediction model.

[0097] The historical object information further includes historical status recommendation information and historical attribute recommendation information. The historical status recommendation information is the expected push result of the historical object status corresponding to the historical object information, while the historical attribute recommendation information is the expected push result of the historical object attribute corresponding to the historical object information.

[0098] For the convenience of introduction, please refer to Figure 5 , Figure 5 which is a schematic structural diagram of the training process in the embodiment of the present application. As Figure 5 As shown, in the solution of this application, an attention algorithm will be used to further calculate its solution. Among them, the information classification model, the state prediction model, and the attribute prediction model will be jointly trained together. Therefore, first, the classification training data will be predicted through the classification model to be trained to obtain a training weighted vector. Subsequently, the training weighted vector will be used as the input value and input into the state model to be trained and the attribute model to be trained for calculation, so as to obtain the state recommendation information and attribute recommendation information output by these two models. Finally, the parameters of the classification model to be trained, the state model to be trained, and the attribute model to be trained will be adjusted according to the deviation between the input result and its result, so as to obtain the corresponding training result. According to the state recommendation information and the historical state recommendation information, the parameters of the classification model to be trained are adjusted to obtain the information classification model. According to the state recommendation information, the training weighted vector, the attribute recommendation information, and the historical state recommendation information, the parameters of the state model to be trained are adjusted to obtain the state prediction model. According to the attribute recommendation information, the training weighted vector, the state recommendation information, and the historical attribute recommendation information, the parameters of the attribute model to be trained are adjusted to obtain the attribute prediction model. When training the state model to be trained and the attribute model to be trained, the output results of the corresponding other party will also be used to adjust each distribution result.

[0099] Specifically, the calculation process of the training weighted vector is as follows:

[0100]

[0101] Among them, a i is the weight value calculated by the attention mechanism Attn. h i is the hidden layer vector. The output vectors are respectively used as the feature inputs of different downstream prediction tasks (disease, constitution prediction), and the downstream prediction tasks are jointly trained with the generation tasks introduced in the same. In this way, the model can, on the one hand, distinguish between past / present content, and at the same time can dynamically select important features in combination with specific prediction classification tasks, including features related to the past and the present.

[0102] In an embodiment of this application, based on the above technical solution, the above step of adjusting the parameters of the state model to be trained according to the state recommendation information, the attribute recommendation information, and the historical state recommendation information to obtain the state prediction model specifically includes the following steps:

[0103] Determine the state prediction distribution according to the state recommendation information and the training weighted vector;

[0104] Determine the attribute prediction distribution according to the attribute recommendation information and the training weighted vector;

[0105] Predict the distribution according to the attribute, weight the eigenvalues in the state prediction distribution to obtain a state weighted distribution;

[0106] Adjust the parameters of the state model to be trained according to the state weighted distribution and the historical state recommendation information to obtain the state prediction model.

[0107] During the training process, the state prediction distribution is determined according to the state recommendation information and the training weighted vector respectively, and the attribute prediction distribution is determined according to the attribute recommendation information and the training weighted vector, so as to obtain the prediction results of the model for the two types of information. Subsequently, the eigenvalues in the state prediction distribution are weighted according to the attribute prediction distribution to obtain a state weighted distribution. Finally, the parameters of the state model to be trained are adjusted according to the state weighted distribution and the historical state recommendation information to obtain the state prediction model. Specifically, for the convenience of introduction, please refer to Figure 6 , Figure 6 which is a schematic structural diagram of the training process in the embodiments of the present application. As Figure 6 shown, for the two different prediction tasks of state prediction and attribute prediction, Task 1 represents disease prediction and is the target task. Task 2 represents constitution prediction. The input is the prediction description text, and the neural network structure of Transformer + classification layer (Cls) is used respectively, and the predicted distributions, namely the state prediction distribution and the attribute prediction distribution, are output. For example, for the disease prediction task, the predicted disease distribution is output, and for the constitution prediction, the constitution distribution is output. For the distribution of one task, the distribution after affine transformation of the intermediate layer output and the attention vector of the other task is used for adjustment. The specific method is as follows. Let the Transformer output corresponding to Task 1 be H1, the corresponding classification layer be Cls1, the attention weighted vector be V, W V , and b1 be the corresponding parameters. Then, for the target task, the predicted disease distribution can be calculated as:

[0108]

[0109] The Transformer output of Task 2 is H2, then there is a probability vector:

[0110]

[0111] where σ is the Sigmoid function. Among them, P D ′ is not a distribution with a sum of 1, but each dimension is a variable in [0,1]. Use 1 - Γ′ D to adjust the prediction distribution of Task 1:

[0112] P D ′ = P D g(1 - Γ′ D )

[0113] In the above formula, using the features of Task 2 as input, the σ function generates the weights of each dimension of the features (within the range of 0 - 1). The higher the weight, the more important a certain feature dimension is for Task 2. For Task 1, however, the importance of the features in this dimension needs to be reduced. Therefore, the adjusted P D ′ is further adjusted in distribution to exclude interference terms that may be caused by other tasks.

[0114] Conversely, we set Task 2 (physical fitness prediction) as the target task, repeat the above steps, and calculate and adjust the prediction distribution of Task 2.

[0115] It should be noted that although the steps of the method in this application are described in a specific order in the drawings, this does not require or imply that these steps must be executed in that specific order, or that all the steps shown must be executed to achieve the desired result. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution, etc.

[0116] The following introduces the device implementation of this application, which can be used to execute the medical information recommendation method in the above embodiments of this application. Figure 7 Schematically shows the composition block diagram of the medical information recommendation device in the embodiments of this application. As Figure 7 shown, the medical information recommendation device 700 mainly may include:

[0117] An information acquisition module 710, configured to acquire object information of a target object;

[0118] An information division module 720, configured to divide the object information into status information and attribute information according to the text content of the object information;

[0119] A vector generation module 730, configured to generate an information weighted vector according to the status information and the attribute information, where the information weighted vector includes the degree of association between each character in the status information and the attribute information and the object status of the target object;

[0120] An information push module 740, configured to determine medical information associated with the target object according to the information weighted vector and push the medical information to the target object.

[0121] In some embodiments of this application, based on the above technical solutions, the information division module 720 includes:

[0122] A status prediction unit, configured to predict status information of the object information through an information classification model, so as to obtain a character probability distribution of the object information, where the character probability distribution is used to indicate the probability that each character in the object information is used to describe the object status of the target object;

[0123] A character combination unit, configured to combine characters for describing the object status of the target object into the status information and combine the remaining characters into the attribute information according to the character probability distribution of the object information.

[0124] In some embodiments of the present application, based on the above technical solution, the vector generation module 730 includes:

[0125] An attention unit, configured to calculate an attention feature value corresponding to each character in the object information through the information classification model according to the character probability distribution, so as to obtain the information weighted vector.

[0126] In some embodiments of the present application, based on the above technical solution, the information push module 740 includes:

[0127] A status recommendation unit, configured to determine status medical information corresponding to the object status of the target object through a status prediction model according to the information weighted vector;

[0128] An attribute recommendation unit, configured to determine attribute medical information corresponding to the object attribute of the target object through an attribute prediction model according to the information weighted vector;

[0129] A push unit, configured to push the status medical information and the attribute medical information to the target object.

[0130] In some embodiments of the present application, based on the above technical solution, the medical information recommendation device 700 further includes:

[0131] A historical information acquisition module, configured to acquire historical object information, where the historical object information includes status information and attribute information;

[0132] A labeling module, configured to label the status information and the attribute information in the historical object information to obtain classification training data;

[0133] A training module, configured to train a classification model to be trained according to the classification training data to obtain the information classification model.

[0134] In some embodiments of the present application, based on the above technical solution, the historical object information further includes historical status recommendation information and historical attribute recommendation information; the training module includes:

[0135] A classification training unit for predicting the classification training data through the classification model to be trained to obtain a training weighted vector;

[0136] A weighted prediction unit for predicting through the status model to be trained and the attribute model to be trained according to the training weighted vector to obtain status recommendation information and attribute recommendation information;

[0137] A classification adjustment unit for adjusting the parameters of the classification model to be trained according to the status recommendation information and the historical status recommendation information to obtain the information classification model;

[0138] A status adjustment unit for adjusting the parameters of the status model to be trained according to the status recommendation information, the training weighted vector, the attribute recommendation information and the historical status recommendation information to obtain the status prediction model;

[0139] An attribute adjustment unit for adjusting the parameters of the attribute model to be trained according to the attribute recommendation information, the training weighted vector, the status recommendation information and the historical attribute recommendation information to obtain the attribute prediction model.

[0140] In some embodiments of the present application, based on the above technical solutions, the status adjustment unit includes:

[0141] A status distribution prediction subunit for determining a status prediction distribution according to the status recommendation information and the training weighted vector;

[0142] An attribute distribution prediction subunit for determining an attribute prediction distribution according to the attribute recommendation information and the training weighted vector;

[0143] A status weighting subunit for weighting the feature values in the status prediction distribution according to the attribute prediction distribution to obtain a status weighted distribution;

[0144] A parameter adjustment subunit for adjusting the parameters of the status model to be trained according to the status weighted distribution and the historical status recommendation information to obtain the status prediction model.

[0145] It should be noted that the device provided in the above embodiment and the method provided in the above embodiment belong to the same concept. The specific ways in which each module performs operations have been described in detail in the method embodiment and will not be repeated here.

[0146] Figure 8 The structure diagram of a computer system of an electronic device suitable for implementing the embodiments of the present application is shown.

[0147] It should be noted that Figure 8 The computer system 800 of the illustrated electronic device is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0148] As Figure 8 shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 802 or the program loaded from the storage section 808 into the random access memory (RAM) 803. In the RAM 803, various programs and data required for system operation are also stored. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0149] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as required. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as required so that a computer program read from it can be installed into the storage section 808 as required.

[0150] In particular, according to the embodiments of the present application, the processes described in each method flowchart can be implemented as computer software programs. For example, the embodiments of the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication section 809 and / or installed from the removable medium 811. When the computer program is executed by the central processing unit (CPU) 801, various functions defined in the system of the present application are executed.

[0151] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. In the present application, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which computer-readable program code is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0152] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the above module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, as well as the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0153] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more of the above-described modules or units can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0154] Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described herein can be implemented by software or by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0155] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present application. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common general knowledge or conventional technical means in the technical field not disclosed in the present application.

[0156] It should be understood that the present application is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.< / end>

Claims

1. A medical information recommendation method, characterized in that, Including: Obtain the object information of the target object; According to the text content of the object information, divide the object information into status information and attribute information; According to the status information and the attribute information, generate an information weighted vector, where the information weighted vector includes the degree of association between each character in the status information and the attribute information and the object status of the target object; According to the information weighted vector, determine the medical information associated with the target object and push the medical information to the target object; Among them, the step of dividing the object information into status information and attribute information according to the text content of the object information includes: Through an information classification model, perform status information prediction on the object information to obtain the character probability distribution of the object information, where the character probability distribution is used to indicate the probability that each character in the object information is used to describe the object status of the target object; According to the character probability distribution of the object information, combine the characters used to describe the object status of the target object into the status information and combine the remaining characters into the attribute information; Among them, the step of generating an information weighted vector according to the status information and the attribute information includes: Through the information classification model, according to the character probability distribution, calculate the attention feature value corresponding to each character in the object information to obtain the information weighted vector.

2. The method according to claim 1, characterized in that, The step of determining the medical information associated with the target object according to the information weighted vector and pushing the medical information to the target object includes: Through a status prediction model, according to the information weighted vector, determine the status medical information corresponding to the object status of the target object; Through an attribute prediction model, according to the information weighted vector, determine the attribute medical information corresponding to the object attribute of the target object; Push the status medical information and the attribute medical information to the target object.

3. The method according to claim 2, characterized in that, Before performing status information prediction on the object information through the information classification model to obtain the character probability distribution of the object information, the method further includes: Obtain historical object information, where the historical object information includes status information and attribute information; Label the status information and the attribute information in the historical object information to obtain classification training data; According to the classification training data, train the classification model to be trained to obtain the information classification model.

4. The method according to claim 3, characterized in that, The historical object information further includes historical status recommendation information and historical attribute recommendation information; the step of training the classification model to be trained according to the classification training data to obtain the information classification model includes: Through the classification model to be trained, perform prediction on the classification training data to obtain a training weighted vector; Through the status model to be trained and the attribute model to be trained, perform prediction according to the training weighted vector to obtain status recommendation information and attribute recommendation information; According to the status recommendation information and the historical status recommendation information, adjust the parameters of the classification model to be trained to obtain the information classification model; Adjust the parameters of the state model to be trained according to the state recommendation information, the training weighted vector, the attribute recommendation information, and the historical state recommendation information to obtain the state prediction model; Adjust the parameters of the attribute model to be trained according to the attribute recommendation information, the training weighted vector, the state recommendation information, and the historical attribute recommendation information to obtain the attribute prediction model.

5. The method according to claim 4, characterized in that, The adjusting the parameters of the state model to be trained according to the state recommendation information, the training weighted vector, the attribute recommendation information, and the historical state recommendation information to obtain the state prediction model includes: Determine the state prediction distribution according to the state recommendation information and the training weighted vector; Determine the attribute prediction distribution according to the attribute recommendation information and the training weighted vector; Weight the eigenvalues in the state prediction distribution according to the attribute prediction distribution to obtain a state weighted distribution; Adjust the parameters of the state model to be trained according to the state weighted distribution and the historical state recommendation information to obtain the state prediction model.

6. A medical information recommendation device, characterized in that, including: An information acquisition module for acquiring object information of a target object; An information division module for dividing the object information into state information and attribute information according to the text content of the object information; A vector generation module for generating an information weighted vector according to the state information and the attribute information, where the information weighted vector includes the degree of association between each character in the state information and the attribute information and the object state of the target object; An information push module for determining medical information associated with the target object according to the information weighted vector and pushing the medical information to the target object; wherein, the dividing the object information into state information and attribute information according to the text content of the object information includes: Predict the state information of the object information through an information classification model to obtain the character probability distribution of the object information, where the character probability distribution is used to indicate the probability that each character in the object information is used to describe the object state of the target object; Combine the characters used to describe the object state of the target object into the state information according to the character probability distribution of the object information and combine the remaining characters into the attribute information; wherein, the generating an information weighted vector according to the state information and the attribute information includes: Calculate the attention eigenvalue corresponding to each character in the object information through the information classification model according to the character probability distribution to obtain the information weighted vector.

7. An electronic device, characterized in that, including: A processor; A memory for storing executable instructions of the processor; wherein, the processor is configured to execute the medical information recommendation method according to any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, implements the medical information recommendation method according to any one of claims 1 to 5.

Images