An information recommendation method, device, equipment, system and storage medium

Abstract

Embodiments of the present application disclose an information recommendation method, device, equipment, system and storage medium. In the method, first, medical text materials of a target object are acquired. Then, symptom entities in the medical text materials are determined. Next, based on the medical text materials, symptom entity feature information and context feature information corresponding to the symptom entities are extracted. Then, the symptom entity feature information and the context feature information are fused to obtain fusion features corresponding to the symptom entities. Finally, information is recommended based on the fusion features corresponding to the symptom entities in the medical text materials. By extracting the symptom entity feature information and the context feature information corresponding to the symptom entities from the medical text materials, richer symptom-related semantics is used as the basis for the recommended information. Compared with the prior art, the technical solution of the present application enriches the basis category of the recommended information, improves the utilization rate of information in the medical text materials, and thus makes the recommended information more accurate.

Description

Technical Field

[0001] This application relates to the field of medical information processing technology, and in particular to an information recommendation method, apparatus, device, system and storage medium. Background Technology

[0002] Today, people are paying increasing attention to health. Traditionally, the solution to health problems involves seeking help from professional medical personnel and institutions to obtain more specialized and accurate information. Currently, due to factors such as the distribution of medical resources and population distribution, many patients also face difficulties in making appointments. The emergence of intelligent medical assistance technologies can alleviate the pressure caused by these problems.

[0003] In recent years, with the rise of artificial intelligence and machine learning, research on intelligent assistance technologies in medical settings has become a hot topic. Some researchers view the medical information recommendation process as a Markov process, using Deep Q-Network (DQN) in reinforcement learning to process the conversational information based on the doctor-patient dialogue, thereby recommending the information needed by the patient.

[0004] Figure 1A This is a schematic diagram illustrating an example of how the DQN model uses explicit symptoms (i.e., symptoms reported by the patient) and implicit symptoms (i.e., symptoms obtained from questioning the patient) for information recommendation. Figure 1A As can be seen, when recommending information in a medical setting, the system only considers whether symptoms are present (true indicates symptoms are present, false indicates no symptoms are present) for the conversational information, ignoring the important value of other medical information to the recommended information, thus resulting in insufficient accuracy of the recommended information. Summary of the Invention

[0005] This application provides an information recommendation device, equipment, system, and storage medium to improve the accuracy of recommended information in medical scenarios and provide patients with more precise medical information services.

[0006] In view of the above, the first aspect of this application provides an information recommendation method, including:

[0007] Obtain medical text materials from the target object;

[0008] Identify the symptom entities in the medical text material;

[0009] Based on the medical text material, extract symptom entity feature information and context feature information corresponding to the symptom entity;

[0010] The symptom entity feature information and the context feature information are fused to obtain the fused feature corresponding to the symptom entity;

[0011] Information recommendation is performed based on the fusion features corresponding to each symptom entity in the medical text material.

[0012] A second aspect of this application provides an information recommendation device, comprising:

[0013] The material acquisition module is used to acquire medical text materials of the target object;

[0014] The symptom entity determination module is used to determine symptom entities in the medical text material;

[0015] The feature extraction module is used to extract symptom entity feature information and context feature information corresponding to the symptom entity based on the medical text material;

[0016] The feature fusion module is used to fuse the symptom entity feature information and the context feature information to obtain the fused feature corresponding to the symptom entity.

[0017] The information recommendation module is used to recommend information based on the fusion features corresponding to each symptom entity in the medical text material.

[0018] A third aspect of this application provides an information recommendation device, the device including a processor and a memory:

[0019] The memory is used to store program code and transfer the program code to the processor;

[0020] The processor is used to execute the steps in the recommended method provided in the first aspect, based on instructions in the program code.

[0021] The fourth aspect of this application provides a disease diagnosis system, including: the information recommendation device described in the third aspect, and further including: a terminal device;

[0022] The terminal device is used to generate medical text material for the target object in response to an input operation, and to send the medical text material to the information recommendation device; the terminal device is also used to receive information recommended by the information recommendation device, and to output the information recommended by the information recommendation device.

[0023] The fifth aspect of this application provides a computer-readable storage medium for storing program code for performing the steps in the information recommendation method as described in the first aspect.

[0024] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0025] This application provides an information recommendation method, apparatus, device, system, and storage medium. In this method, medical text material of the target object is first acquired. Then, symptom entities in the medical text material are identified. Next, based on the medical text material, symptom entity feature information and contextual feature information corresponding to the symptom entities are extracted. Then, the symptom entity feature information and contextual feature information are fused to obtain the fused features corresponding to the symptom entities. Finally, information recommendation is performed based on the fused features corresponding to each symptom entity in the medical text material. In this application's technical solution, by extracting symptom feature information, preceding context feature information, and following context feature information corresponding to symptom entities from the medical text material, richer symptom-related semantics are used as the basis for recommended information. Compared to existing technologies that only use the presence or absence of symptoms as the basis for recommended information, this application's technical solution enriches the scope of the recommended information, improves the utilization rate of information in the medical text material, and thus makes the recommended information more accurate. Attached Figure Description

[0026] Figure 1A This is a schematic diagram illustrating an example of how the DQN model uses explicit and implicit symptoms for information recommendation.

[0027] Figure 1B A flowchart illustrating an information recommendation method provided in this application embodiment;

[0028] Figure 2 A flowchart illustrating the process of identifying symptom entities in medical text materials;

[0029] Figure 3 A flowchart illustrating the process of extracting multiple feature information corresponding to symptom entities;

[0030] Figure 4 This is a flowchart illustrating the process of recommending information based on the fusion features corresponding to each symptom entity in the medical text material.

[0031] Figure 5 This is a schematic diagram illustrating the initialization of a medical knowledge graph, as provided in an embodiment of this application.

[0032] Figure 6 This is a schematic diagram illustrating the process of information recommendation based on the nodes reached, which involves propagation and reasoning based on the initialized medical knowledge graph.

[0033] Figure 7 A schematic diagram of graph propagation provided in an embodiment of this application;

[0034] Figure 8 This is a sample diagram illustrating how knowledge graph reasoning is used to ultimately determine recommended information based on medical text materials.

[0035] Figure 9A This is another example illustration of determining recommended information based on medical text materials through knowledge graph reasoning;

[0036] Figure 9B A diagram illustrating an intelligent consultation dialogue with a patient;

[0037] Figure 9C This is a schematic diagram illustrating the process of feature aggregation of symptom entities based on the history of consultation dialogue, followed by knowledge dissemination and reasoning to ultimately achieve information recommendation.

[0038] Figure 10A A schematic diagram of a training information recommendation model provided in an embodiment of this application;

[0039] Figure 10B This is a schematic diagram of the structure of an information recommendation device provided in an embodiment of this application;

[0040] Figure 11 A schematic diagram of the structure of a server for information recommendation provided in an embodiment of this application;

[0041] Figure 12 A schematic diagram of the structure of a terminal device for information recommendation provided in an embodiment of this application;

[0042] Figure 13 This is a schematic diagram of the information recommendation system in an embodiment of this application. Detailed Implementation

[0043] Current technologies for intelligent medical information recommendation typically rely solely on the patient's symptom presentation as the basis for recommendations. This approach simplifies the clinical consultation process by having the doctor continuously ask about certain symptoms, and the patient simply answer yes or no to collect conversational material for information recommendations. However, in actual consultations, doctors usually need to recommend information based on specific symptom attributes such as symptom location, frequency, duration, and cause. Because these factors are ignored, the information recommended by current technologies suffers from insufficient accuracy.

[0044] To address the above issues, this application proposes an information recommendation method, apparatus, device, system, and storage medium. In this application's technical solution, the basis for information recommendation is no longer limited to the symptoms themselves; features are also extracted from the symptom entity context of the medical text material, thereby effectively obtaining richer symptom-related semantic auxiliary information for recommendation. This improves the accuracy of the recommended information and provides patients with more credible and reliable information recommendation services in intelligent medical information recommendation scenarios.

[0045] The technical solution provided in this application is applied to a processing device, which is a processing device with information recommendation function, and therefore can be called an information recommendation device. The information recommendation device can be a terminal device or a server with information recommendation function. The technical solution of this application can be executed independently by the terminal device or the server, or it can be applied to a network scenario where the terminal device and the server communicate, operating in cooperation. The terminal device can be a mobile phone, desktop computer, personal digital assistant (PDA), tablet computer, etc. The server can be understood as an application server or a web server. In actual deployment, the server can be a standalone physical server, or a server cluster or distributed system composed of multiple physical servers. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0046] To facilitate understanding, several example implementation scenarios of the solution are introduced below.

[0047] Scenario 1: The patient inputs their medical information on their own device (e.g., a mobile phone). This device communicates with an information recommendation device. The device generates medical text based on the patient's input and sends this text to the information recommendation device. The information recommendation device then makes recommendations based on this medical text.

[0048] The second scenario: Patients input information on their own terminal device, providing medical-related data. This terminal device acts as an information recommendation device, generating medical text materials based on the patient's input. The terminal device then makes information recommendations based on these medical text materials.

[0049] The third scenario: Patients input information at self-service terminals in the hospital's outpatient lobby, providing their medical information. These terminals are connected to an information recommendation system. The terminals generate medical text based on the patient's input and send this text to the information recommendation system, which then makes recommendations based on the medical text.

[0050] The fourth scenario: Patients input their medical information into self-service terminals in the hospital's outpatient lobby. These terminals act as information recommendation devices, generating medical text materials based on the patient's input. The self-service terminals then recommend information based on these medical text materials.

[0051] As an example, the medical information described above can be based on questions asked on personal or self-service terminals and the patient's responses. These questions are not limited to those about symptoms but can also include inquiries about symptom location, frequency, duration, and cause. The medical text material can be generated from these questions and responses.

[0052] As another example, the medical information described above can also be a description of the illness imported into the device. This description can be edited and compiled by the patient themselves, and includes information such as symptoms, symptom location, frequency of attacks, duration, and cause. Additionally, the description can also be a patient's previous medical history. There are no restrictions on the source or format of the description.

[0053] In this embodiment, after the information recommendation device obtains the information to be recommended based on the aforementioned medical text material, it can provide this information to the patient. Specifically, this can be done by displaying the information in text form on an interface providing medical information, by voice output, or by sending a text message. If the patient provides the medical information through their own terminal device or a hospital's self-service terminal device, and another device acts as the information recommendation device, then the information recommendation device needs to send the recommended information to the patient's own terminal device or the hospital's self-service terminal device.

[0054] The above scenarios are merely implementation examples. In practical applications, the person providing patient medical information is not limited to the patient themselves, but can also be the patient's family members, friends, etc. Similarly, the device sending patient medical text materials to the information recommendation device is not limited to the patient's own terminal device, but can also be the terminal device of the patient's family members or friends.

[0055] The information recommendation method provided in the embodiments of this application will be described below. See also Figure 1B This diagram is a flowchart of an information recommendation method. Figure 1B As shown, the method includes:

[0056] S101: Obtain the medical text material of the target object.

[0057] The target audience here refers to the patient. Medical text materials, as previously introduced, can be consultation dialogue texts based on questions and answers. Since the aforementioned text occurred before this medical consultation and serves as the basis for this consultation, for ease of distinction and understanding, the consultation dialogue text is referred to as historical consultation dialogue text. In addition, medical text materials can also be descriptive texts of the patient's condition, such as medical history.

[0058] S102: Identify symptom entities in medical text material.

[0059] A symptom entity refers to a word that conforms to the naming rules for symptoms. Here, a symptom entity can be any symptom entity appearing in the medical text. For example, the first symptom entity appears in line 3 of the medical text, and the second symptom entity appears in line 7. In step S103, described later, the symptom entity feature information and contextual feature information of the first symptom entity, as well as the symptom entity feature information and contextual feature information of the second symptom entity, can be obtained respectively.

[0060] S103: Based on medical text materials, extract symptom entity feature information and context feature information corresponding to symptom entities.

[0061] The contextual feature information corresponding to a symptom entity includes the preceding and following context features. The symptom entity feature information can also be referred to as the symptom entity's own feature information. Extracting the symptom entity feature information, preceding context feature information, and following context feature information corresponding to a symptom entity primarily aims to collect other relevant information besides whether a symptom is present, which helps improve the accuracy of information recommendation. As an example, the preceding context feature information and the following context feature information can be implemented through feature aggregation, which will be discussed in detail later.

[0062] S104: Perform feature fusion on symptom entity feature information and context feature information to obtain the fused feature corresponding to the symptom entity.

[0063] This step involves fusing the symptom entity features and contextual features of the same symptom entity to obtain the fused features corresponding to that symptom entity. In one possible implementation, if the symptom entity features, the preceding context features, and the following context features are each represented by a feature vector, then the fusion operation here can be a concatenation operation of the feature vectors.

[0064] S105: Information recommendation based on the fusion features corresponding to each symptom entity in medical text materials.

[0065] In an optional implementation, a pre-trained information recommendation model can determine the information to be recommended based on fused features, thereby performing information recommendation. The information recommended in this embodiment may include, but is not limited to, one of the following:

[0066] Disease entity, symptom entity, hospital information, department information or doctor information, etc.

[0067] In this context, "disease entity" refers to a word that conforms to the naming rules for diseases. Both symptom entities and disease entities can be identified through a medical entity dictionary. The medical entity dictionary is not limited to recording symptom entities and disease entities; it can also include other categories of medical entities, such as drugs and examinations. Disease entities and symptom entities are related; for example, when a certain disease occurs, a patient typically experiences several symptoms.

[0068] The information recommendation method provided in this application extracts symptom entity feature information, preceding context feature information, and following context feature information from medical text materials, using richer symptom-related semantics as the basis for information recommendation. Compared to existing technologies that only use the presence or absence of symptoms as the basis for information recommendation, the technical solution of this application enriches the scope of information recommendation criteria, improves the utilization rate of information in medical text materials, increases the directional nature of symptoms for recommended information, and makes the recommended information more accurate, thereby more precisely meeting the actual needs of patients.

[0069] In the method described in the foregoing embodiments, the step of determining the symptom entities in the medical text material in S102 can be implemented in the following manner. Figure 2 A flowchart illustrating the process of identifying symptom entities in medical text material. (Example) Figure 2 As shown, this process includes:

[0070] S1021: Segment sentences in medical text materials to obtain multiple words;

[0071] S1022: Match multiple words using a medical entity dictionary to identify the symptom entities that successfully match among the multiple words.

[0072] Segmentation of sentences is a relatively mature technology in this field, therefore the segmentation process and methods will not be limited or elaborated here. If a word obtained from sentence segmentation matches a symptom entity recorded in a medical entity dictionary, the match is successful, and this word belongs to the symptom entity.

[0073] In the method described in the foregoing embodiments, step S103, which extracts symptom entity feature information and context feature information corresponding to symptom entities based on medical text materials, can be implemented in the following manner. Figure 3 This is a flowchart illustrating the process of extracting various feature information corresponding to symptom entities. For example... Figure 3 As shown, this process includes:

[0074] S1031: Encode the statement to obtain the feature vectors of the words in the statement;

[0075] S1032: Use the feature vector of the symptom entity as the symptom entity feature information corresponding to the symptom entity;

[0076] S1033: Based on the feature vector of the symptom entity and the feature vector of the words preceding the symptom entity in the medical text material, aggregate to obtain the preceding context feature information of the symptom entity; based on the feature vector of the symptom entity and the feature vector of the words following the symptom entity in the medical text material, aggregate to obtain the following context feature information of the symptom entity.

[0077] In an optional implementation, a bidirectional Long Short-Term Memory (LSTM) encoder with shared parameters can be used to encode sentences in medical text materials. This yields a feature vector for each word that contains contextual information about the sentence. In this embodiment, the focus is on the feature vectors of words corresponding to symptom entities.

[0078] As an example, if the j-th word in the i-th sentence of the target medical text material is a symptom entity, then its feature vector obtained after passing through a bidirectional LSTM encoder is represented as follows: This is used as the symptom entity feature information corresponding to the symptom entity.

[0079] Medical text materials often contain symptom attribute information, such as location, frequency of attacks, duration, and cause. This information may be recorded in the preceding and / or following context of the symptom entity. Taking a consultation dialogue as an example, the discussion of a symptom entity usually involves more than one round; therefore, symptom attribute information may also exist in the previous and subsequent rounds of dialogue where the symptom entity appears. To extract features that are highly relevant to the recommended information, an attention mechanism can be used to dynamically select the content to be aggregated.

[0080] In practice, the preceding and following contextual features of the symptom entity can be aggregated using the following formulas (1) and (2), respectively.

[0081] Formula (1)

[0082] In formula (1), The feature vector represents the j-th word of the i-th sentence in the target medical text material as the symptom entity. and Each represents the feature vector of any word in the context of the symptom entity. The feature vector obtained after aggregating these words through an attention mechanism is called the contextual feature information of the symptom entity.

[0083] According to formula (1), in order to obtain the contextual feature information of the symptom entity, the contextual feature information of the symptom entity can be obtained by aggregating the feature vector of the symptom entity and the feature vector of the words in the context of the symptom entity in the medical text material.

[0084] Formula (2)

[0085] In formula (2), This represents the feature vector of the j-th word of the i-th sentence in the target medical text material as the symptom entity, and is related to the feature vector in formula (1). Equivalent. and Each represents the feature vector of any word following the symptom entity. The feature vector obtained after aggregating these words through an attention mechanism is called the contextual feature information of the symptom entity.

[0086] According to formula (2), in order to obtain the contextual feature information of the symptom entity, the contextual feature information of the symptom entity can be obtained by aggregating the feature vector of the symptom entity and the feature vector of the words in the context of the symptom entity in the medical text material.

[0087] In practical applications, in step S104, the obtained symptom entity feature information, preceding context feature information, and following context feature information can be concatenated together to form a high-dimensional feature vector three times the length of the symptom entity's feature vector, which serves as the fusion feature. The fusion feature can be represented as a vector as follows: .

[0088] In this embodiment, the contextual and background feature information of a symptom entity is obtained by aggregating its own feature vectors. Therefore, the contextual and background feature information of a symptom entity are also associated with that symptom entity. This provides guidance for accurate subsequent information recommendations.

[0089] Currently, the KR-DS model can be used to incorporate the co-occurrence relationship between symptoms and diseases into a reinforcement learning framework to improve the accuracy of information recommendation. However, this approach only models knowledge using a co-occurrence graph of symptoms and diseases, without directly applying a knowledge graph. Therefore, it has limitations, and its accuracy is restricted by the available co-occurrence relationships between symptoms and diseases. Integrating medical knowledge into medical information recommendation scenarios could help achieve more accurate recommendations.

[0090] Based on this, this application embodiment introduces an implementation method of the above information recommendation in conjunction with a medical knowledge graph.

[0091] join Figure 4 This figure is a schematic diagram illustrating a process for information recommendation based on the fusion features corresponding to various symptom entities in medical text materials, as provided in an embodiment of this application. Figure 4 As shown, this process includes:

[0092] S1041: Obtain a medical knowledge graph, which includes a set of nodes and a set of edges. The set of nodes includes disease nodes and symptom nodes, and the set of edges includes the edges between nodes. Disease nodes represent disease entities, symptom nodes represent symptom entities, and edges represent the association between two connected nodes.

[0093] S1042: Initialize the features of the corresponding symptom nodes in the medical knowledge graph using fusion features to obtain the initialized medical knowledge graph.

[0094] S1043: Based on the initialized medical knowledge graph, propagation and reasoning are performed, and information is recommended according to the nodes reached.

[0095] This application utilizes medical knowledge graphs for knowledge dissemination and reasoning. The aim is to match the known symptom information in the target medical text material with the relationships between symptom entities (i.e., symptom nodes) and disease entities (i.e., disease nodes) in the medical knowledge graph. This will disseminate the information contained in the medical text material from known symptoms to medically related diseases and symptoms that are not mentioned in the material, thus preparing for the determination of the information to be recommended.

[0096] Suppose we obtain a Chinese Medical Knowledge Graph (CMeKG), denoted as G. Graph G includes a set of nodes V and an edge set E, where G = (V, E). If we directly use the obtained graph, we cannot effectively apply the fusion features obtained through splicing, as described earlier; that is, we cannot effectively apply the symptom-related information extracted from the target object's medical text material. For this reason, this application proposes to initialize the directly obtained medical knowledge graph using a fusion feature vector obtained through splicing.

[0097] Suppose a graph G contains M symptom nodes and N disease nodes, each with a corresponding feature vector. For known symptom entities appearing in the medical text, their fused features are used to initialize the corresponding symptom nodes in the graph. For other symptom entities not appearing in the medical text, their feature vectors are initialized with all-zero feature vectors.

[0098] In this embodiment, only the edges between symptom nodes and disease nodes in the initialized knowledge graph can be retained. If the knowledge graph explicitly indicates that certain symptoms are caused by a certain disease, then the corresponding entity nodes are connected. Edges between nodes of the same type are not retained; for example, edges between disease nodes and edges between symptom nodes are deleted. Figure 5 This is a schematic diagram illustrating the initialization of a medical knowledge graph. Figure 5In the diagram, solid nodes represent disease nodes, and hollow nodes represent symptom nodes. Figure 5 The left side shows the graph before initialization. Figure 5 The right side shows the initialized atlas. Nodes marked with "H" represent known symptoms and have been initialized using corresponding fusion features. Nodes marked with "0" represent non-known symptoms and have been initialized using all-zero feature vectors.

[0099] In optional implementations, the co-occurrence frequency of disease and symptom entities can be used to calculate the weight of each edge, reflecting the strength of the relationship between the two entities. For example, if a disease and a symptom frequently appear in the same case, it suggests a stronger likelihood of a close connection between them. Finally, erroneous edges can be manually removed by medical professionals to refine the initialized medical knowledge graph.

[0100] In this embodiment, the initialized knowledge graph undergoes a two-step propagation: the first step propagates from known symptom nodes to disease nodes; the second step propagates from the disease nodes reached in the previous step back to the symptom nodes.

[0101] To facilitate the explanation of the graph propagation process in the initialized knowledge graph, the following will combine... Figure 6 This paper introduces a process for propagation and reasoning based on an initialized medical knowledge graph, and for recommending information based on the nodes reached. Figure 6 This is a schematic diagram of the process. (For example...) Figure 6 As shown, the process includes:

[0102] S10431: In the initialized medical knowledge graph, the first step of propagation from symptom nodes to disease nodes is carried out based on fusion features. Figure 7 This is a schematic diagram of graph propagation provided in an embodiment of this application. Figure 7 Symptom nodes marked as 1 are symptom nodes after feature fusion initialization. These nodes can be understood as symptom entities that are known in the medical text material and belong to known symptom entities. Figure 7 The disease node marked with 2 indicates the disease node reached after the first step of propagation. For example... Figure 7 In this example, "dry mouth" and "acid reflux" are two known symptom entities. Their hidden states are passed to the disease entity "gastric ulcer" through iterative operations of a graph convolutional network.

[0103] S10432: Establish a disease entity feature set based on the post-propagation feature vectors of the disease nodes. After the first step of propagation, the feature vectors of the disease nodes change. For ease of description, the concept of a disease entity feature set is defined in this embodiment. This set specifically includes the feature vectors of the disease nodes reached after the first step of propagation. This disease entity feature set is represented by formula (3).

[0104] Formula (3)

[0105] In formula (3), This represents the set of fused features of symptom nodes representing known symptom entities. This represents the set of entity features of a disease. This represents the transformation parameter matrix of a graph convolutional network. Updates can be made during the model training phase. It is an adjacency matrix of dimension M*N, representing the relationship between symptoms and diseases, where M is the number of symptom nodes in the graph and N is the number of disease nodes in the graph. The degree matrix representing symptom nodes The reverse, It is a diagonal matrix. The elements on the diagonal are The lines and. This represents the nonlinear activation function LeakyReLU.

[0106] S10433: Based on the first step of transmission, the second step of transmission proceeds from the disease node to the symptom node. Figure 7 The symptom node marked with 3 is a related symptom node that has spread from the disease node reached in the first step of transmission to another related symptom node. For example... Figure 7 As shown, in the second step of transmission, "gastric ulcer" is transmitted to other symptom nodes, such as "abdominal bloating" and "nausea".

[0107] S10434: Establish a symptom entity feature set based on the propagated feature vectors of the symptom nodes. After the second propagation, the feature vectors of the symptom nodes changed. For ease of description, the concept of a symptom entity feature set is defined in this embodiment. This set specifically includes the feature vectors of the symptom nodes reached after the second propagation. This symptom entity feature set is represented by formula (4).

[0108] Formula (4)

[0109] In formula (4), Represents the set of entity features of a disease. This represents the set of entity features of the symptoms. This represents the transformation parameter matrix of a graph convolutional network. Updates can be made during the model training phase. It is an adjacency matrix of dimension M*N, representing the relationship between symptoms and diseases, where M is the number of symptom nodes in the graph and N is the number of disease nodes in the graph. Degree matrix representing disease nodes The reverse, It is a diagonal matrix. The elements on the diagonal are The columns and. This represents the nonlinear activation function LeakyReLU.

[0110] After the above two steps of propagation, a set of disease entity features and a set of symptom entity features are obtained, represented as follows: Then we can base it on To recommend information.

[0111] Specifically, S10435: Determine the recommendation probability of each disease entity corresponding to the disease entity feature set based on the disease entity feature set and the symptom entity feature set. Next, S10436: Recommend information based on the recommendation probability of each disease entity. In practice, this can be achieved using a multilayer perceptron neural network combined with a sigmoid function. The system categorizes symptoms to determine the recommendation probability for each disease or symptom entity. Considering that different symptoms or diseases may occur simultaneously, the recommendation probability for each entity is between 0 and 1 and is independent of the others.

[0112] In one alternative implementation, disease entities with the highest determined recommendation probability can be recommended using a first method, while disease entities with other non-zero recommendation probabilities can be recommended using a second method. For example, the first method uses red to indicate that the recommended disease entities require close attention, while the second method uses yellow to indicate that the recommended disease entities require attention. Of course, the first and second methods in the above examples are merely illustrations; in practical applications, different textual descriptions can be used to recommend different information.

[0113] Figure 8 This is a sample diagram illustrating how knowledge graph reasoning can be used to determine recommended information based on medical text materials. Figure 9A This is another example illustration of determining recommended information based on knowledge graph reasoning using medical text materials. Figure 8 and Figure 9A In the displayed historical consultation dialogue texts, P represents the narration or response from the patient's perspective, and D represents the question from the doctor's perspective. Figure 8 and Figure 9AIn the displayed knowledge graph, solid circles represent disease nodes, and hollow circles represent symptom nodes.

[0114] exist Figure 8 and Figure 9A The displayed recommendations include "Considered Diseases" and "Other Possible Diseases." "Considered Diseases" refers to information recommended using the first method, while "Other Possible Diseases" refers to information recommended using the second method. Both methods are provided for patients and doctors' reference.

[0115] In other possible embodiments, the recommendation probability of each symptom entity corresponding to the symptom entity feature set can be determined based on the disease entity feature set and the symptom entity feature set; and when the maximum recommendation probability of the disease entity is less than the first preset probability threshold, the symptom entity whose recommendation probability exceeds the preset second preset probability can be recommended through a third method.

[0116] exist Figure 8 In the example shown, "abdominal pain," "nausea," and "vomiting" are three symptoms that have appeared in the dialogue history, and they are closely related to both "appendicitis" and "pancreatitis." Based on these symptoms, this scheme identifies "appendicitis" as a likely disease entity to recommend. However, due to insufficient confidence in the disease entity (maximum recommendation probability less than the first preset probability threshold), to differentiate between two other possible diseases, "pancreatitis" and "irritable bowel syndrome," this scheme further transfers the entity information to other appendicitis-related symptoms, "fever" and "diarrhea," thus obtaining the recommendation information on the right. Specifically, "fever" and "diarrhea" are treated as symptoms to be clarified, and the textual description of "symptoms to be clarified" serves as a third method for recommending these two symptom entities.

[0117] exist Figure 9A In the example shown, many known symptoms such as "stomach pain," "bloating," "acid reflux," and "belching" all point to the disease node "gastric ulcer," indicating a sufficiently high confidence level (the maximum recommendation probability is greater than or equal to the first preset probability threshold). Therefore, it is unnecessary to ask the patient for additional symptoms, and thus, the recommended symptom entities are no longer displayed through a third-party method. Figure 9AThe suggested recommendations do not include "symptoms to be clarified." Furthermore, when providing "symptoms to be clarified," all predicted non-zero probabilities can be included, or higher-probability symptoms can be selected. For example, symptom entities with probabilities exceeding a preset second probability can be included in the recommended information. As an example, the second preset probability could be set to 0.5. Of course, the top-ranking symptoms can be provided based on actual needs, or the second preset probability can be set to other values. Including symptoms to be clarified in the recommended information helps achieve more accurate information recommendations.

[0118] The graph propagation process described in the above embodiments can be presented to patients or doctors who subsequently examine them in a visual manner. The information recommendation method described in the embodiments of this application may further include:

[0119] The symptom and disease nodes in the initialized medical knowledge graph are displayed in their first state.

[0120] Once the first step of propagation is completed, the disease nodes that have been propagated will be changed from the first state to the second state for display.

[0121] Once the second step of propagation is complete, the symptom nodes that have been propagated will be changed from the first state to the third state for display.

[0122] For example, in a medical knowledge graph, the default display state of each node is the first state. Disease nodes reached in the first step are displayed in the second state, and symptom nodes reached in the second step are displayed in the third state. The second and third states can be the same or different. The displayed node states can be dynamically presented; for example, the graph might first display disease nodes reached in the first step in the second state, and then display symptom nodes reached in the second step in the third state. As an example, the first state of a node in the graph is white, the second state is blue, and the third state is green. Thus, by using different display states, nodes that have been propagated are distinguished from those that haven't, facilitating a more intuitive understanding of this distinction.

[0123] The information recommended in this solution can be directly provided to patients. Of course, to enhance interactivity, the recommended information can also be transformed into corresponding statements based on preset rules and displayed to the user. For example, "Consider disease: gastric ulcer" can be transformed into "You may have a gastric ulcer; please go to the hospital to contact a doctor for further examination." It can also simultaneously match relevant doctors for recommendations.

[0124] In a conversational medical consultation scenario, medical text materials include historical consultation dialogue texts. To ensure the accuracy of the recommended information, this solution continuously updates the text in real time by collecting dialogue texts.

[0125] Therefore, in the information recommendation method provided in this application embodiment, after completing one round of information recommendation, the following steps can be performed:

[0126] Generate question statements based on symptom entities whose recommended probability exceeds a second preset probability;

[0127] Output the question statement;

[0128] Collect the response statements to the questions provided by the target audience;

[0129] Update medical text materials based on question and answer statements.

[0130] In this way, when the interactive consultation dialogue is updated to the next round, materials can be collected based on the new round of dialogue, and more accurate information recommendations can be made in real time.

[0131] To provide patients with more convenient medical information services, after recommending the disease entity, departments and / or doctors can also be recommended. In this embodiment, the information recommendation method may further include:

[0132] Department information is recommended based on the disease entity corresponding to the highest recommendation probability and the affiliation between the disease and the department; and / or, doctor information is recommended based on the disease entity corresponding to the highest recommendation probability and the association between the doctor and their areas of expertise. Diseases and departments typically have an affiliation, with specific diseases handled by specific departments. For example, eye diseases are handled by ophthalmology. Therefore, the department responsible for a disease can be determined based on the recommended disease entity and the aforementioned affiliation. Recommending departments to patients prevents them from registering with the wrong department due to inaccurate understanding of their condition, thus affecting diagnosis and treatment. Doctors typically have areas of expertise within their departments; for example, a doctor's introduction usually lists several diseases they are skilled at treating. Therefore, based on the disease entity in the recommended information and the association between the doctor and their areas of expertise, information about doctors specializing in treating that disease can be recommended to patients. This leverages the doctor's professional expertise and improves treatment outcomes.

[0133] Expected application scenarios include interactive pre-medical intelligent consultation, as follows: Figure 9B The image shows an intelligent consultation dialogue with a patient. Current pre-medical consultations first ask a few relatively fixed questions based on the patient's self-report, then allow the user to select their corresponding symptoms from a limited selection, and finally provide triage suggestions. The technical solution in this application, however, combines... Figure 9BAs shown, this approach enhances the relevance of symptom inquiries, allowing for the extraction of more valuable information from patients and improving the accuracy of subsequent recommendations, thereby increasing patient access efficiency. Furthermore, this solution provides more granular information recommendations, extending beyond disease-specific details to include hospital, department, and doctor information. Considering the varying specialties of different doctors, this approach ensures patients are matched with more suitable and professional physicians, thus improving their overall healthcare experience.

[0134] Figure 9C This diagram illustrates how information recommendation is achieved through knowledge dissemination and reasoning after aggregating features of symptom entities based on the history of online consultations. Figure 9C In the example symptom entity extracted from the dialogue history, "acid reflux" is used. T1, T2, T i-1 T i T i+1 This indicates different rounds of dialogue. Figure 9C The recommended information displayed is determined using a multi-layer perceptron (MLP) classifier.

[0135] In the information recommendation method described in the above embodiments, information recommendation can be specifically implemented through an information recommendation model. This information recommendation model can be used to recommend medical information such as disease entities based on input medical text materials. To facilitate understanding of the implementation of this model, the following will be combined with... Figure 10A The training process of the model is described. Figure 10A A schematic diagram for training an information recommendation model.

[0136] In this embodiment of the application, the steps for training the information recommendation model are as follows:

[0137] 1) Obtain the initial model, which includes a symptom feature extraction neural network and an information recommendation neural network. The symptom feature extraction neural network is used to identify symptom entities in medical text materials. Based on the medical text materials, it extracts symptom entity feature information and contextual feature information corresponding to the symptom entities; it then fuses the symptom entity feature information and contextual feature information to obtain the fused features corresponding to the symptom entities. The neural recommendation neural network is used to recommend information based on the fused features corresponding to each symptom entity in the medical text materials.

[0138] The initial model is the basic model before training. This model needs to be improved through multiple training sessions to obtain an information recommendation model that can achieve the required information recommendation function.

[0139] 2) Obtain the training sample set, which includes multiple medical text sample materials.

[0140] Here, the medical text sample material corresponds to the aforementioned medical text material, specifically it can be a consultation dialogue or a description of the patient's condition. The medical text sample material can also be material from other patients. There are no restrictions on the source of the material here.

[0141] 3) Input the medical text sample material into the symptom feature extraction neural network, and obtain the symptom entity feature information, preceding context feature information and following context feature information corresponding to the symptom entities in the medical text sample material through the symptom feature extraction neural network.

[0142] Combination Figure 10A As shown, the medical text sample material serves as the input to the initial model. The symptom feature extraction neural network is a relatively front-end network in the model, and its output serves as the input to the information recommendation neural network. The symptom feature extraction neural network outputs the self-feature information, contextual feature information, and background feature information of the symptom entity. These can be input in parallel or sequentially into the information recommendation neural network, or they can be fused (i.e., vector concatenation) to obtain fused features, which are then input into the information recommendation neural network. Based on the preceding description, this symptom feature extraction neural network effectively aggregates the symptom entity and contextual features in the medical text sample material.

[0143] 4) Input the fused features into the information recommendation neural network, and obtain the recommendation information corresponding to the medical text sample material through the information recommendation neural network.

[0144] In this embodiment, the information recommendation neural network can be further divided into a first network for graph construction and graph propagation (responsible for knowledge propagation and reasoning), and a second network that determines the recommended information based on the disease entity feature set and symptom entity feature set obtained from the two-step propagation reasoning. Finally, the information recommendation neural network will output recommended information for the previously input medical text sample material. 5) Based on the gap between the information output by the information recommendation neural network and the labels of the medical text sample material, the parameters in the symptom feature extraction neural network and the information recommendation neural network are adjusted until the training cutoff condition is met, thus obtaining the information recommendation model.

[0145] The labels for the medical text sample materials can be provided by experts in the medical field. In an optional implementation, a binary cross-entropy loss function can be used to simultaneously update the parameters of the aforementioned neural network. The binary cross-entropy loss function is a function of the difference between the recommended information and the labels of the medical text sample materials; the larger the difference, the higher the loss value; conversely, the smaller the difference, the smaller the loss value. Training cutoff conditions can be either the loss value reaching a preset value or the number of training iterations reaching a preset number.

[0146] In practical applications, there is a potential need to categorize medical condition descriptions. For example, classifying a medical condition description into a specific disease category. Categorized medical condition descriptions facilitate doctors' work in classifying and archiving medical histories, or for later use in surveys and disease screenings. If the medical text material consists of medical condition descriptions, the information recommendation method can also include categorizing these descriptions based on the recommended information. After categorization, the medical condition descriptions can be labeled with a classification tag to distinguish them from descriptions categorized in other ways.

[0147] This application also compares the information recommendation model trained using this method with other proposed models. Two different datasets were used as input, and the comparisons were made using nine performance evaluation metrics, as shown in Table 1.

[0148] Table 1. Comparison of Evaluation Indicators between This Solution and Other Solutions

[0149]

[0150] For the Chinese Pediatric Medical Dialogue Dataset (CMDD), the evaluation metrics include next-round entity precision P@1, next-round first-position recall R@1 and first-three-position recall R@3, and total entity precision P. f @1. For the Chinese gastroenterology medical diagnosis dataset MDG-D, the evaluation metrics include the accuracy, recall, and F1 score of symptom prediction. s / R s / F1 s F1 score for disease diagnosis (F1) d ), and total entity accuracy P f The comparison schemes include LSTM (Long Short-Term Memory Network), H-LSTM (Hierarchical Long Short-Term Memory Network), TextCNN (Text Convolutional Network), BERT-wmm (Pre-trained Language Model), and GAT (Graph Attention Network).

[0151] As shown in the table above, this approach demonstrates superior performance compared to five common text classification models. It achieves the best results in 8 out of 9 evaluation metrics across the CMDD and MDG-D datasets, including entity accuracy, recall, and F1 score, only losing to the BERT-wmm model in one metric. Furthermore, the neural network in this approach is significantly lightweight, with only one-tenth the number of parameters compared to BERT-wmm.

[0152] This approach uses a graph convolutional network to implement the inference process. However, it can be replaced with various other graph networks, such as graph attention networks or GraphSAGE, but these may introduce more model parameters, thus increasing the training difficulty.

[0153] Based on the information recommendation method described in the foregoing embodiments, this application also provides an information recommendation device. The specific implementation of this device will be described below with reference to embodiments and accompanying drawings. Figure 10B As shown, the information recommendation device 20 includes the following components:

[0154] The module comprises: material acquisition module 21, symptom entity identification module 22, feature extraction module 23, feature fusion module 24, and information recommendation module 25. Among them:

[0155] Material acquisition module 21 is used to acquire medical text materials of the target object;

[0156] Symptom entity determination module 22 is used to determine symptom entities in the medical text material;

[0157] Feature extraction module 23 is used to extract symptom entity feature information and context feature information corresponding to the symptom entity based on the medical text material;

[0158] Feature fusion module 24 is used to fuse the symptom entity feature information and the context feature information to obtain the fused feature corresponding to the symptom entity;

[0159] The information recommendation module 25 is used to recommend information based on the fusion features corresponding to each symptom entity in the medical text material.

[0160] Optionally, the symptom entity determination module 22 includes:

[0161] The word segmentation unit is used to segment sentences in the medical text material into multiple words;

[0162] The determining unit is used to match the multiple words using a medical entity dictionary to identify the symptom entities that are successfully matched among the multiple words.

[0163] Optionally, the contextual feature information includes the preceding and following contextual feature information of the symptom entity; the feature extraction module 23 includes:

[0164] The encoding unit is used to encode the statement to obtain the feature vectors of the words in the statement;

[0165] The first extraction unit is used to extract the feature vector of the symptom entity as the symptom entity feature information corresponding to the symptom entity.

[0166] The second extraction unit is used to aggregate the contextual feature information of the symptom entity based on the feature vector of the symptom entity and the feature vector of the words in the context of the symptom entity in the medical text material.

[0167] The third extraction unit is used to aggregate the contextual feature information of the symptom entity based on the feature vector of the symptom entity and the feature vector of the words in the context of the symptom entity in the medical text material.

[0168] Optionally, the information recommendation module 25 includes:

[0169] The graph acquisition unit is used to acquire a medical knowledge graph, which includes a set of nodes and a set of edges. The set of nodes includes disease nodes and symptom nodes, and the set of edges includes edges between nodes. Disease nodes represent disease entities, symptom nodes represent symptom entities, and edges represent the association between two connected nodes.

[0170] The graph initialization unit is used to initialize the features of the corresponding symptom nodes in the medical knowledge graph using the fusion features, so as to obtain the initialized medical knowledge graph.

[0171] The graph application unit is used to perform propagation and reasoning based on the initialized medical knowledge graph, and to recommend information based on the nodes reached.

[0172] Optionally, the map application unit includes:

[0173] The first propagation unit is used to perform the first step of propagation from symptom nodes to disease nodes based on the fusion features in the initialized medical knowledge graph.

[0174] The first set establishment unit is used to establish a disease entity feature set based on the post-propagation feature vector of the disease nodes to which the disease has been propagated.

[0175] The second propagation unit is used to carry out a second step of propagation from the disease node to the symptom node based on the first step of propagation.

[0176] The second set establishment unit is used to establish a symptom entity feature set based on the propagated feature vectors of the symptom nodes.

[0177] The recommendation probability determination unit is used to determine the recommendation probability of each disease entity corresponding to the disease entity feature set based on the disease entity feature set and the symptom entity feature set.

[0178] The recommendation unit is used to recommend information based on the recommendation probability of each disease entity.

[0179] Optionally, the recommendation unit includes:

[0180] The first recommendation unit is used to recommend disease entities corresponding to the determined maximum recommendation probability through a first method;

[0181] The second recommendation unit is used to recommend disease entities corresponding to other non-zero recommendation probabilities determined by the second method.

[0182] The recommendation probability determination unit is also used to determine the recommendation probability of each symptom entity corresponding to the symptom entity feature set based on the disease entity feature set and the symptom entity feature set;

[0183] The third recommendation unit is used to recommend symptom entities whose recommendation probability exceeds a preset second preset probability in a third manner when the maximum recommendation probability is less than a first preset probability threshold.

[0184] Optionally, the medical text material includes historical consultation dialogue text; the information recommendation device 20 also includes:

[0185] The question generation module is used to generate question statements based on symptom entities whose recommendation probability exceeds a preset second preset probability.

[0186] The question output module outputs the question statement;

[0187] The response statement collection module is used to collect response statements to the questions provided by the target object;

[0188] The text material update module is used to update the medical text material based on the question and the answer.

[0189] Optionally, the information recommendation device 20 further includes: a model training module for training an information recommendation model, which is used to recommend information based on input medical text materials; the model training module includes:

[0190] An initial model acquisition unit is used to acquire an initial model, which includes a symptom feature extraction neural network and an information recommendation neural network.

[0191] A sample set acquisition unit is used to acquire a training sample set, which includes multiple medical text sample materials.

[0192] The first processing unit is used to input the medical text sample material into the symptom feature extraction neural network, determine the symptom entities in the medical text sample material through the symptom feature extraction neural network, extract symptom entity feature information and context feature information corresponding to the symptom entities in the medical text sample material based on the medical text sample material, and obtain the fused features corresponding to the symptom entities in the medical text sample material after feature fusion.

[0193] The second processing unit is used to input the fusion features corresponding to each symptom entity in the medical text sample material into the information recommendation neural network to obtain the information output by the information recommendation neural network.

[0194] The parameter adjustment unit is used to adjust the parameters in the symptom feature extraction neural network and the information recommendation neural network based on the gap between the information output by the information recommendation neural network and the labels of the medical text sample material, until the training cutoff condition is met, thereby obtaining the information recommendation model.

[0195] Optionally, the medical text material is a description of the patient's condition; the information recommendation device 20 also includes:

[0196] The classification module is used to classify the description text of the illness based on the recommended information.

[0197] Optionally, the information recommendation device 20 and the information recommendation module 25 are also used for:

[0198] Recommend department information based on the disease entity corresponding to the maximum recommendation probability and the affiliation between the disease and the department; and / or,

[0199] Doctor information is recommended based on the disease entity corresponding to the maximum recommendation probability and the association between the doctor and the diseases they are good at.

[0200] Optionally, the information recommendation device 20 also includes:

[0201] The node display module is used to display symptom nodes and disease nodes in the initialized medical knowledge graph in a first state; when the first step of propagation is completed, the disease nodes that have been propagated are changed from the first state to the second state for display; when the second step of propagation is completed, the symptom nodes that have been propagated are changed from the first state to the third state for display.

[0202] Based on the implementation scenarios of this solution described above, the technical solution of this application can be applied to a server, which can then be used as an information recommendation device. Figure 11This is a schematic diagram of a server structure provided in an embodiment of this application. The server 900 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 922 (e.g., one or more processors) and memory 932, and one or more storage media 930 (e.g., one or more mass storage devices) for storing application programs 942 or data 944. The memory 932 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the server. Furthermore, the CPU 922 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the server 900.

[0203] Server 900 may also include one or more power supplies 926, one or more wired or wireless network interfaces 950, one or more input / output interfaces 958, and / or one or more operating systems 941, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0204] The steps performed by the server in the above embodiments can be based on this Figure 11 The server structure shown.

[0205] CPU 922 is used to perform the following steps:

[0206] Obtain medical text materials from the target object;

[0207] Identify the symptom entities in the medical text material;

[0208] Based on the medical text material, extract symptom entity feature information and context feature information corresponding to the symptom entity;

[0209] The symptom entity feature information and the context feature information are fused to obtain the fused feature corresponding to the symptom entity;

[0210] Information recommendation is performed based on the fusion features corresponding to each symptom entity in the medical text material.

[0211] This application also provides another information recommendation device, such as... Figure 12As shown, for ease of explanation, only the parts related to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application. The terminal can be any terminal device including mobile phones, tablets, personal digital assistants (PDAs), point-of-sale (POS) terminals, in-vehicle computers, etc. Taking a mobile phone as an example:

[0212] Figure 12 This is a block diagram illustrating a portion of the structure of a mobile phone related to the terminal provided in the embodiments of this application. (Reference) Figure 12 The mobile phone includes: a radio frequency (RF) circuit 1010, a memory 1020, an input unit 1030, a display unit 1040, a sensor 1050, an audio circuit 1060, a wireless fidelity (WiFi) module 1070, a processor 1080, and a power supply 1090, etc. Those skilled in the art will understand that... Figure 12 The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0213] The following is combined Figure 12 A detailed introduction to each component of a mobile phone:

[0214] The RF circuit 1010 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with the processor 1080; additionally, it transmits uplink data to the base station. Typically, the RF circuit 1010 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), a duplexer, etc. Furthermore, the RF circuit 1010 can also communicate wirelessly with networks and other devices. The aforementioned wireless communications may use any communication standard or protocol, including but not limited to Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0215] The memory 1020 can be used to store software programs and modules. The processor 1080 executes various mobile phone functions and data processing by running the software programs and modules stored in the memory 1020. The memory 1020 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 1020 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0216] The input unit 1030 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the mobile phone. Specifically, the input unit 1030 may include a touch panel 1031 and other input devices 1032. The touch panel 1031, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 1031), and drive the corresponding connection devices according to a pre-set program. Optionally, the touch panel 1031 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 1080, and can also receive and execute commands sent by the processor 1080. In addition, the touch panel 1031 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 1031, the input unit 1030 may also include other input devices 1032. Specifically, other input devices 1032 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0217] The display unit 1040 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 1040 may include a display panel 1041, which may optionally be configured as a Liquid Crystal Display (LCD), Organic Light-Emitting Diode (OLED), or similar display panel 1041. Furthermore, a touch panel 1031 may cover the display panel 1041. When the touch panel 1031 detects a touch operation on or near it, it transmits the information to the processor 1080 to determine the type of touch event. Subsequently, the processor 1080 provides corresponding visual output on the display panel 1041 based on the type of touch event. Although in Figure 12 In this embodiment, the touch panel 1031 and the display panel 1041 are two separate components to realize the input and output functions of the mobile phone. However, in some embodiments, the touch panel 1031 and the display panel 1041 can be integrated to realize the input and output functions of the mobile phone.

[0218] The mobile phone may also include at least one sensor 1050, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 1041 according to the ambient light level, and the proximity sensor can turn off the display panel 1041 and / or backlight when the phone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, taps), etc. Other sensors that may be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0219] The audio circuit 1060, speaker 1061, and microphone 1062 provide an audio interface between the user and the mobile phone. The audio circuit 1060 converts the received audio data into electrical signals and transmits them to the speaker 1061, where the speaker 1061 converts them into sound signals for output. On the other hand, the microphone 1062 converts the collected sound signals into electrical signals, which are then received by the audio circuit 1060, converted into audio data, and then processed by the processor 1080 before being transmitted via the RF circuit 1010 to, for example, another mobile phone, or the audio data can be output to the memory 1020 for further processing.

[0220] WiFi is a short-range wireless transmission technology. Through the WiFi module 1070, mobile phones can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access. Although Figure 12 The WiFi module 1070 is shown, but it is understood that it is not an essential component of a mobile phone and can be omitted as needed without changing the essence of the invention.

[0221] The processor 1080 is the control center of the mobile phone, connecting various parts of the phone through various interfaces and lines. It executes various functions and processes data by running or executing software programs and / or modules stored in the memory 1020 and calling data stored in the memory 1020. Optionally, the processor 1080 may include one or more processing units; preferably, the processor 1080 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1080.

[0222] The mobile phone also includes a power supply 1090 (such as a battery) that supplies power to various components. Preferably, the power supply can be logically connected to the processor 1080 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0223] Although not shown, mobile phones may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0224] In this embodiment of the application, the processor 1080 included in the terminal also has the following functions:

[0225] Obtain medical text materials from the target object;

[0226] Identify the symptom entities in the medical text material;

[0227] Based on the medical text material, extract symptom entity feature information and context feature information corresponding to the symptom entity;

[0228] The symptom entity feature information and the context feature information are fused to obtain the fused feature corresponding to the symptom entity;

[0229] Information recommendation is performed based on the fusion features corresponding to each symptom entity in the medical text material.

[0230] This application also provides a computer-readable storage medium for storing program code that implements one or more steps in any of the aforementioned information recommendation methods.

[0231] This application also provides a computer program product including instructions that, when run on a computer, cause the computer to perform one or more steps in any of the aforementioned information recommendation methods.

[0232] Please see Figure 13 , Figure 13 This is a schematic diagram of an embodiment of the information recommendation system in this application, wherein the information recommendation system includes: an information recommendation device and a terminal device;

[0233] The terminal device is used to generate medical text materials for the target object in response to input operations, and to send the medical text materials to the information recommendation device. The terminal device can be the patient's own terminal device, the terminal device (such as a mobile phone) of a relative or friend who helps the patient with online or pre-medical consultations, or a self-service terminal device in the hospital's outpatient hall.

[0234] The information recommendation device is used to acquire medical text materials of a target object; identify symptom entities in the medical text materials; extract symptom entity feature information and context feature information corresponding to the symptom entities based on the medical text materials; fuse the symptom entity feature information and the context feature information to obtain the fused feature corresponding to the symptom entity; and perform information recommendation based on the fused feature corresponding to each symptom entity in the medical text materials.

[0235] The terminal device is also used to output information recommended by the information recommendation device.

[0236] Figure 13 The diagram illustrates several different terminal devices. In practical applications, information recommendation devices can simultaneously provide medical information recommendation services to one or more terminal devices.

[0237] In this application's technical solution, the information recommendation device extracts the self-feature information, preceding context information, and following context information of symptom entities from medical text materials, using richer symptom-related semantics as the basis for information recommendation. Compared to existing technologies that only use the presence or absence of symptoms as the basis for information recommendation, this application's technical solution enriches the scope of information recommendation criteria, improves the utilization rate of information in medical text materials, and makes information recommendation more accurate. Through the interaction between the terminal device and the information recommendation device, patients can obtain the recommended information in a timely manner. Based on the recommended information, they can register for medical appointments and obtain further diagnosis and treatment strategies.

[0238] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0239] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules and units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0240] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0241] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0242] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0243] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An information recommendation method, characterized in that, include: Obtain medical text materials from the target object; Identify the symptom entities in the medical text material; Based on the medical text material, extract symptom entity feature information and context feature information corresponding to the symptom entity; The symptom entity feature information and the context feature information are fused to obtain the fused feature corresponding to the symptom entity; Obtaining a medical knowledge graph; The features of the corresponding symptom nodes in the medical knowledge graph are initialized using the fusion features to obtain the initialized medical knowledge graph; In the initialized medical knowledge graph, the first step of propagation from symptom nodes to disease nodes is carried out based on the fusion features; A set of disease entity features is established based on the post-propagation feature vectors of the disease nodes to which the disease has spread. Based on the first step of transmission, a second step of transmission proceeds from the disease node to the symptom node; A set of symptom entity features is established based on the propagation feature vectors of the symptom nodes to which the propagation occurs; the propagated symptom nodes are other symptom nodes related to the disease nodes to which the propagation occurs. Based on the disease entity feature set and the symptom entity feature set, the recommendation probability of each disease entity corresponding to the disease entity feature set and the recommendation probability of each symptom entity corresponding to the symptom entity feature set are determined. Information recommendation is performed based on the recommendation probability of each disease entity, including: recommending disease entities corresponding to the determined maximum recommendation probability through a first method; and recommending symptom entities whose recommendation probability exceeds a preset second preset probability through a third method when the maximum recommendation probability is less than a first preset probability threshold.

2. The method according to claim 1, characterized in that, The process of determining symptom entities in the medical text material includes: The sentences in the medical text material are segmented into multiple words; By matching the multiple words using a medical entity dictionary, the symptom entities that successfully match the multiple words are identified.

3. The method according to claim 2, characterized in that, The contextual feature information includes the preceding and following contextual feature information of the symptom entity; The step of extracting symptom entity feature information and contextual feature information corresponding to the symptom entity based on the medical text material includes: The statement is encoded to obtain the feature vectors of the words in the statement; The feature vector of the symptom entity is used as the symptom entity feature information corresponding to the symptom entity; Based on the feature vector of the symptom entity and the feature vector of the words preceding the symptom entity in the medical text material, the preceding text feature information of the symptom entity is aggregated. Based on the feature vector of the symptom entity and the feature vector of the words following the symptom entity in the medical text material, the context feature information of the symptom entity is obtained by aggregating them.

4. The method according to claim 1, characterized in that, The medical knowledge graph includes a set of nodes and a set of edges; the set of nodes includes disease nodes and symptom nodes, and the set of edges includes edges between nodes. Disease nodes represent disease entities, symptom nodes represent symptom entities, and edges represent the association between two connected nodes.

5. The method according to claim 1, characterized in that, The information recommendation based on the recommendation probability of each disease entity includes: The second method is used to recommend disease entities corresponding to other non-zero recommendation probabilities.

6. The method according to claim 5, characterized in that, The medical text materials include historical consultation dialogue texts; the method also includes: Generate question statements based on symptom entities whose recommended probability exceeds a second preset probability; Output the question statement; Collect the response statements to the questions provided by the target object; The medical text material is updated based on the question and the response.

7. The method according to any one of claims 1-5, characterized in that, Also includes: A training information recommendation model is used to recommend information based on input medical text materials; the information recommendation model is trained in the following manner: Obtain an initial model, which includes a symptom feature extraction neural network and an information recommendation neural network; Obtain a training sample set, which includes multiple medical text sample materials; The medical text sample material is input into the symptom feature extraction neural network. The symptom entities in the medical text sample material are determined by the symptom feature extraction neural network. Based on the medical text sample material, the symptom entity feature information and context feature information corresponding to the symptom entities in the medical text sample material are extracted. After feature fusion, the fused features corresponding to the symptom entities in the medical text sample material are obtained. The fusion features corresponding to each symptom entity in the medical text sample material are input into the information recommendation neural network to obtain the information output by the information recommendation neural network. Based on the gap between the information output by the information recommendation neural network and the labels of the medical text sample material, the parameters in the symptom feature extraction neural network and the information recommendation neural network are adjusted until the training cutoff condition is met, thereby obtaining the information recommendation model.

8. The method according to any one of claims 1-5, characterized in that, The medical text material is a description of a patient's condition; the method further includes: The description of the illness is categorized based on the recommended information.

9. The method according to claim 5 or 6, characterized in that, Also includes: Recommend department information based on the disease entity corresponding to the maximum recommendation probability and the affiliation between the disease and the department; And / or, Doctor information is recommended based on the disease entity corresponding to the maximum recommendation probability and the association between the doctor and the diseases they are good at.

10. The method according to any one of claims 1, 5, and 6, characterized in that, Also includes: The symptom nodes and disease nodes in the initialized medical knowledge graph are displayed in the first state. When the first step of propagation is completed, the disease nodes that have been propagated to are changed from the first state to the second state for display. When the second step of propagation is completed, the symptom nodes that have been propagated are changed from the first state to the third state for display.

11. An information recommendation device, characterized in that, include: The material acquisition module is used to acquire medical text materials of the target object; The symptom entity determination module is used to determine symptom entities in the medical text material; The feature extraction module is used to extract symptom entity feature information and context feature information corresponding to the symptom entity based on the medical text material; The feature fusion module is used to fuse the symptom entity feature information and the context feature information to obtain the fused feature corresponding to the symptom entity. The information recommendation module includes: a map acquisition unit, a map initialization unit, and a map application unit; The graph acquisition unit is used to acquire medical knowledge graphs; The graph initialization unit is used to initialize the features of the corresponding symptom nodes in the medical knowledge graph using the fusion features, so as to obtain the initialized medical knowledge graph. The map application unit includes: The first propagation unit is used to perform the first step of propagation from symptom nodes to disease nodes based on the fusion features in the initialized medical knowledge graph. The first set establishment unit is used to establish a disease entity feature set based on the post-propagation feature vector of the disease nodes to which the disease has been propagated. The second propagation unit is used to carry out a second step of propagation from the disease node to the symptom node based on the first step of propagation. The second set establishment unit is used to establish a symptom entity feature set based on the propagated feature vectors of the symptom nodes; the propagated symptom nodes are other symptom nodes related to the propagated disease nodes. The recommendation probability determination unit is used to determine the recommendation probability of each disease entity corresponding to the disease entity feature set and the recommendation probability of each symptom entity corresponding to the symptom entity feature set based on the disease entity feature set and the symptom entity feature set. The recommendation unit is used to recommend information based on the recommendation probability of each disease entity; The recommendation unit includes: The first recommendation unit is used to recommend disease entities corresponding to the determined maximum recommendation probability through a first method; The third recommendation unit is used to recommend symptom entities whose recommendation probability exceeds a preset second preset probability in a third manner when the maximum recommendation probability is less than a first preset probability threshold.

12. The apparatus according to claim 11, characterized in that, The symptom entity determination module includes: The word segmentation unit is used to segment sentences in the medical text material into multiple words; The determining unit is used to match the multiple words using a medical entity dictionary to identify the symptom entities that are successfully matched among the multiple words.

13. The apparatus according to claim 12, characterized in that, The contextual feature information includes the preceding and following contextual feature information of the symptom entity; The feature extraction module includes: The encoding unit is used to encode the statement to obtain the feature vectors of the words in the statement; The first extraction unit is used to extract the feature vector of the symptom entity as the symptom entity feature information corresponding to the symptom entity. The second extraction unit is used to aggregate the contextual feature information of the symptom entity based on the feature vector of the symptom entity and the feature vector of the words in the context of the symptom entity in the medical text material. The third extraction unit is used to aggregate the contextual feature information of the symptom entity based on the feature vector of the symptom entity and the feature vector of the words in the context of the symptom entity in the medical text material.

14. The apparatus according to claim 11, characterized in that, The medical knowledge graph includes a set of nodes and a set of edges; the set of nodes includes disease nodes and symptom nodes, and the set of edges includes edges between nodes. Disease nodes represent disease entities, symptom nodes represent symptom entities, and edges represent the association between two connected nodes.

15. The apparatus according to claim 11, characterized in that, The recommendation unit includes: The second recommendation unit is used to recommend disease entities corresponding to other non-zero recommendation probabilities determined by the second method.

16. The apparatus according to claim 15, characterized in that, The medical text material includes historical consultation dialogue texts; the device also includes: The question generation module is used to generate question statements based on symptom entities whose recommendation probability exceeds a preset second preset probability. The question output module is used to output the question statement; The response statement collection module is used to collect response statements to the questions provided by the target object; The text material update module is used to update the medical text material based on the question and the answer.

17. The apparatus according to any one of claims 11-15, characterized in that, Also includes: The model training module is used to train an information recommendation model, which is used to recommend information based on input medical text materials. The model training module includes: An initial model acquisition unit is used to acquire an initial model, which includes a symptom feature extraction neural network and an information recommendation neural network. A sample set acquisition unit is used to acquire a training sample set, which includes multiple medical text sample materials. The first processing unit is used to input the medical text sample material into the symptom feature extraction neural network, determine the symptom entities in the medical text sample material through the symptom feature extraction neural network, extract symptom entity feature information and context feature information corresponding to the symptom entities in the medical text sample material based on the medical text sample material, and obtain the fused features corresponding to the symptom entities in the medical text sample material after feature fusion. The second processing unit is used to input the fusion features corresponding to each symptom entity in the medical text sample material into the information recommendation neural network to obtain the information output by the information recommendation neural network. The parameter adjustment unit is used to adjust the parameters in the symptom feature extraction neural network and the information recommendation neural network based on the gap between the information output by the information recommendation neural network and the labels of the medical text sample material, until the training cutoff condition is met, thereby obtaining the information recommendation model.

18. The apparatus according to any one of claims 11-15, characterized in that, The medical text material is a description of a patient's condition; the device also includes: The classification module is used to classify the description text of the illness based on the recommended information.

19. The apparatus according to claim 15 or 16, characterized in that, The information recommendation module is also used for: Recommend department information based on the disease entity corresponding to the maximum recommendation probability and the affiliation between the disease and the department; and / or, Doctor information is recommended based on the disease entity corresponding to the maximum recommendation probability and the association between the doctor and the diseases they are good at.

20. The apparatus according to any one of claims 11, 15, and 16, characterized in that, Also includes: The node display module is used to display the symptom nodes and disease nodes in the initialized medical knowledge graph in a first state. When the first step of propagation is completed, the disease node that has been propagated is changed from the first state to the second state for display; when the second step of propagation is completed, the symptom node that has been propagated is changed from the first state to the third state for display.

21. An information recommendation device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the steps of the information recommendation method as described in any one of claims 1-10 according to the instructions in the program code.

22. An information recommendation system, characterized in that, The information recommendation device according to claim 21 also includes a terminal device; The terminal device is configured to generate medical text material for the target object in response to an input operation, and to send the medical text material to the information recommendation device; the terminal device is also configured to receive information recommended by the information recommendation device, and to output the information recommended by the information recommendation device.

23. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for performing the steps in the information recommendation method as described in any one of claims 1-10.

24. A computer program product, characterized in that, Includes instructions that, when executed on a computer, cause the computer to perform the steps of the information recommendation method as described in any one of claims 1-10.

Images