Data mining methods and systems applied to remote animal disease diagnosis center platforms

Abstract

This application provides a data mining method and system applied to a remote animal disease diagnosis center platform. It acquires cross-domain attention knowledge vectors from animal disease case data and decomposes them into sample attention knowledge vectors representing risk diffusion trends or non-risk diffusion trends. A first risk diagnosis network is used to perform preliminary risk diagnosis on the sample attention knowledge vectors, and the network is optimized based on the diagnosis results and attention tag attributes, improving diagnostic accuracy. Furthermore, by extracting sample attention knowledge vectors representing risk diffusion trends and loading them into a second risk diagnosis network, deeper risk diagnosis and optimization are performed, ensuring the accuracy of risk diagnosis. Finally, the two optimized risk diagnosis networks are connected to generate a target risk diagnosis model, which can effectively diagnose whether risk diffusion trends exist during animal disease data mining, providing data support for remote diagnosis of animal diseases and significantly improving the efficiency and accuracy of animal disease prevention and control.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a data mining method and system applied to a remote diagnostic center platform for animal diseases. Background Technology

[0002] In the field of animal disease prevention and control, timely and accurate identification of disease transmission trends and potential risks is crucial to ensuring animal health and maintaining public health security. Traditional animal disease diagnosis methods mainly rely on field investigations and laboratory testing. While these methods are effective, they suffer from problems such as long diagnostic cycles, high costs, and limited coverage, making it difficult to meet the needs of large-scale, rapid-response disease prevention and control.

[0003] With the development of technologies such as big data and artificial intelligence, data mining techniques have been gradually introduced into the field of animal disease diagnosis. By constructing animal disease data mining models, valuable information and patterns can be extracted from massive amounts of animal disease case data, providing a scientific basis for disease prediction, early warning, and prevention. However, most existing animal disease data mining methods focus on the extraction and classification of disease characteristics, lacking in-depth analysis and diagnosis of disease risk spread trends. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of this application, embodiments of this application provide a data mining method applied to an animal disease remote diagnostic center platform, the method comprising:

[0005] When the animal disease data mining model performs data mining on animal disease case data, the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data is obtained. Based on the mining results of the animal disease data mining model on the animal disease case data, the cross-domain attention knowledge vector is decomposed into sample attention knowledge vectors of risk diffusion trend or sample attention knowledge vectors of non-risk diffusion trend.

[0006] The sample attention knowledge vector is loaded into the first risk diagnosis network to generate a first risk diagnosis result of the first risk diagnosis network on the sample attention knowledge vector. Based on the first risk diagnosis result and the attention tag attribute of the sample attention knowledge vector, the first risk diagnosis network is optimized.

[0007] From the sample attention knowledge vectors loaded into the first risk diagnosis network, extract the sample attention knowledge vectors with the attention tag attribute of risk diffusion trend, and continue to load them into the second risk diagnosis network to generate the second risk diagnosis result of the second risk diagnosis network on the sample attention knowledge vectors. Based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors, optimize the second risk diagnosis network.

[0008] The optimized first risk diagnosis network is connected with the second risk diagnosis network to generate a target risk diagnosis model. The target risk diagnosis model is used to diagnose whether there is a risk diffusion trend when the animal disease data mining model is performing data mining.

[0009] In another aspect, embodiments of this application also provide a remote diagnostic system for animal diseases, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, the embodiments of this application achieve efficient and accurate mining and analysis of animal disease case data, significantly improving the accuracy and timeliness of animal disease risk diagnosis. Specifically, firstly, by acquiring the cross-domain attention knowledge vectors of the animal disease data mining model during the mining process, and decomposing them into sample attention knowledge vectors of risk diffusion trends and non-risk diffusion trends, it is possible to accurately capture key information directly related to the spread of disease risk, effectively avoiding information redundancy and improving the targeting of risk identification. By introducing a first risk diagnosis network and a second risk diagnosis network, this invention constructs a dual risk diagnosis mechanism. The first risk diagnosis network performs preliminary risk assessment, while the second risk diagnosis network conducts a more in-depth analysis of risk diffusion trends. This mechanism not only improves the accuracy of diagnosis but also ensures the comprehensiveness of the diagnostic results. The risk diagnosis network can automatically adjust network parameters based on the comparison between the diagnostic results and sample data, achieving self-optimization. This capability enables the model to continuously adapt to new epidemic situations and data characteristics, maintaining the stability and advancement of its diagnostic performance. By connecting the optimized first and second risk diagnosis networks, the generated target risk diagnosis model can quickly and accurately determine whether there is a risk spread trend in the animal disease data mining results. This not only greatly improves diagnostic efficiency but also ensures the accuracy and reliability of the diagnostic results, providing strong support for the timely prevention and control of animal diseases. Therefore, the technical solution in this embodiment is applicable to animal disease remote diagnosis center platforms, enabling real-time and remote mining and analysis of animal disease case data, providing decision-makers with timely and accurate epidemic information. This helps strengthen cross-regional cooperation in animal disease prevention and control, improving overall prevention and control effectiveness. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the data mining method applied to the remote diagnostic center platform for animal diseases provided in the embodiments of this application.

[0012] Figure 2This is a schematic diagram of the hardware architecture of the remote diagnosis system for animal diseases provided in the embodiments of this application. Detailed Implementation

[0013] The present application will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a data mining method applied to an animal disease remote diagnosis center platform according to an embodiment of this application. The data mining method applied to the animal disease remote diagnosis center platform will be described in detail below.

[0014] Step S110: Obtain the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data when the animal disease data mining model performs data mining on the animal disease case data. Based on the mining results of the animal disease data mining model on the animal disease case data, decompose the cross-domain attention knowledge vector into sample attention knowledge vectors of risk diffusion trend or sample attention knowledge vectors of non-risk diffusion trend.

[0015] In detail, the animal disease data mining model is a model used for in-depth mining and analysis of animal disease case data. It can extract useful information from a large amount of animal disease data, helping users better understand the transmission patterns and risk factors of diseases. The animal disease case data refers to specific case data about animal diseases, including detailed information such as the type of disease, time of occurrence, location, affected animal species and numbers, symptoms, and treatment. The disease knowledge feature chain is a knowledge representation method used to describe various features related to diseases and their interrelationships. Like a chain, it connects various features of diseases (such as symptoms, transmission routes, susceptible animals, etc.) to form a complete disease knowledge system.

[0016] The cross-domain attention knowledge vector is a multi-dimensional vector used to represent the degree of attention paid to different domains (or "domains") in the disease knowledge feature chain during animal disease data mining. Cross-domain means encompassing multiple different knowledge domains or aspects.

[0017] The sample knowledge vectors related to the risk diffusion trend refer to those sample knowledge vectors that are relevant to the risk diffusion trend of the epidemic. They typically contain features or information indicating the potential spread of the epidemic, such as the speed of transmission and changes in the scope of infection. The sample knowledge vectors for non-risk diffusion trends, on the other hand, represent features or information not directly related to the risk diffusion trend of the epidemic, reflecting the stability or controlled state of the epidemic in certain aspects.

[0018] In this embodiment, when processing animal disease-related data, the first step is to obtain relevant information from the animal disease data mining model when performing data mining on animal disease case data. Specifically, the server stores a large amount of animal disease case data, covering disease situations in different regions and for different types of animals. For example, in a large animal breeding base, there are various animals, such as pigs, cattle, and chickens. For pig disease case data, the disease case description data may include detailed information such as the pig's age, breed, recent diet, and whether it has been in contact with other animals. The disease condition description data will record whether there are conditions such as accelerated virus transmission (which is the trend of disease risk factors) in the case description data of diseases like swine fever.

[0019] The server utilizes an animal disease data mining model to mine this animal disease case data. This data mining model includes a disease case feature extraction unit, a cross-domain association unit, and a knowledge reasoning unit (the knowledge reasoning unit further includes an epidemic condition feature extraction subunit and an epidemic condition feature reconstruction subunit). Taking swine disease data as an example, the disease case feature extraction unit extracts features such as the pig's age and breed from the disease case description data, generating a disease case feature vector. For instance, for a 3-month-old white pig, its breed characteristics, dietary characteristics, etc., are quantified as elements in the feature vector.

[0020] Then, the cross-domain association unit, based on a pre-defined classification standard for disease-related knowledge domains (this standard may be formulated based on years of animal disease research results and industry norms, such as classifying diseases by type (e.g., infectious diseases, parasitic diseases); symptom categories (e.g., fever, cough); and sources of infection (e.g., animal-to-animal transmission, environmental transmission), divides each element in the disease case feature vector into different knowledge subsets. For example, a pig's age may be related to the source of infection and be classified into the corresponding knowledge subset. Next, a cross-domain knowledge association rule set is set (e.g., younger pigs may be more susceptible to certain diseases; this is a rule set based on the structure of the disease knowledge system and the prior needs of animal disease diagnosis). For each knowledge subset, the elements within it are traversed, and the association relationships with elements in other knowledge subsets are determined and marked. If a white pig breed is found to be associated with certain specific sources of infection, it is marked. Next, a disease knowledge feature chain framework is constructed, starting from a core knowledge subset (such as a subset related to the source of disease infection). Related knowledge subsets are then connected step-by-step along element association markers. Elements within these knowledge subsets are then arranged according to progressively refined associations, and their completeness is checked and optimized. This includes checking whether the feature chain covers all key knowledge areas of swine fever and whether it can be used for effective swine fever diagnosis. If problems exist, missing knowledge elements are added (such as newly discovered elements related to swine fever transmission routes), and unreasonable associations are adjusted (such as correcting incorrect associations between breeds and susceptibility). Finally, the disease knowledge feature chain is generated.

[0021] Simultaneously, the epidemic disease feature extraction subunit represents the epidemic disease description data using graph features, generating a disease trend knowledge graph. For example, for swine fever, the virus transmission speed and changes in the infection range are constructed into a graph. The epidemic disease feature reconstruction subunit performs feature reconstruction based on the cross-domain feature path data generated by fusing the disease knowledge feature chain and the disease trend knowledge graph, generating the mining results of the animal disease data mining model on the animal disease case data. In this process, the server obtains the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data from the animal disease data mining model.

[0022] Based on the data mining results of animal disease case data using an animal disease data mining model, the server decomposes cross-domain attention knowledge vectors into either example attention knowledge vectors representing risk diffusion trends or non-risk diffusion trends. For example, if the mining results indicate that the spread of swine fever in a certain breeding area is accelerating and the number of infected pigs is increasing, then the cross-domain attention knowledge vector related to this situation will be decomposed into example attention knowledge vectors representing risk diffusion trends. If the mining results show that the swine fever epidemic is effectively controlled in a certain area, the spread is slowing down, and the number of infected pigs is stable or decreasing, then the corresponding cross-domain attention knowledge vector will be decomposed into example attention knowledge vectors representing non-risk diffusion trends.

[0023] Step S120: Load the sample attention knowledge vector into the first risk diagnosis network, generate the first risk diagnosis result of the first risk diagnosis network on the sample attention knowledge vector, and optimize the first risk diagnosis network based on the first risk diagnosis result and the attention tag attribute of the sample attention knowledge vector.

[0024] In detail, the first risk diagnosis network is a network model used for preliminary diagnosis of disease risk. It can receive sample attention knowledge vectors as input and output preliminary diagnostic results regarding the spread trend of disease risk. The first risk diagnosis result is the output of the first risk diagnosis network after processing the input sample attention knowledge vectors, and is usually a probability value or score, representing the likelihood that the disease has a risk of spreading.

[0025] The attention tag attribute refers to a tag information carried in the sample attention knowledge vector, which is used to identify which disease risk state (such as risk spread trend, non-risk spread trend) the feature or information represented by the vector is related to.

[0026] For example, in this embodiment, the server loads the sample attention knowledge vector obtained in the previous step into the first risk diagnosis network. Assume the first risk diagnosis network is a neural network built based on a deep learning algorithm, and this neural network has some pre-set initial parameters. Once the sample attention knowledge vector is loaded, the first risk diagnosis network begins its computation. For example, for the sample attention knowledge vector of risk spread trends obtained from previous swine disease cases (containing knowledge vectors transformed from information such as the speed of swine fever transmission, the age distribution of infected pigs, and the farming environment), the first risk diagnosis network will process this information according to its own neuron structure and weight settings.

[0027] The first-risk diagnostic network analyzes the sample knowledge vectors and generates a first-risk diagnostic result. For example, the diagnostic result might be a probability value, indicating whether swine fever has a high-risk spread trend (e.g., probability value of 0.8) or a low-risk spread trend (e.g., probability value of 0.2) under the current circumstances. Simultaneously, the sample knowledge vectors have a focus tag attribute, which is previously labeled based on the actual situation of the disease case data. For example, cases with a clear swine fever risk spread trend are labeled with a "high-risk spread" tag.

[0028] Based on the initial risk diagnosis result and the attention tag attributes of the sample attention knowledge vector, the server begins to optimize the initial risk diagnosis network. If the initial risk diagnosis result is that swine fever has a high-risk spread trend (probability value of 0.8), and the attention tag attribute of the sample attention knowledge vector is also "high-risk spread," this indicates that the initial risk diagnosis network's diagnosis is correct, but further optimization may be needed to improve accuracy. The server will adjust parameters such as neuron weights in the initial risk diagnosis network according to a certain optimization algorithm (such as backpropagation). For example, if the neuron weight corresponding to the knowledge vector related to the spread rate of swine fever in the initial risk diagnosis network was 0.3 before optimization, after this correct diagnosis, it may be adjusted to 0.32 according to the algorithm to make the diagnosis more accurate when encountering similar sample attention knowledge vectors in the future. If the initial risk diagnosis result does not match the attention tag attribute, for example, the diagnosis result is low-risk spread (probability value of 0.2), while the tag attribute is "high-risk spread," then the server will increase the magnitude of the weight adjustment to correct the network's diagnostic bias.

[0029] Step S130: Extract the sample attention knowledge vectors with the attention tag attribute of risk diffusion trend from the sample attention knowledge vectors loaded into the first risk diagnosis network, continue to load them into the second risk diagnosis network, generate the second risk diagnosis result of the second risk diagnosis network on the sample attention knowledge vectors, and optimize the second risk diagnosis network based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors.

[0030] Specifically, the sample training labeled data refers to the labeled information provided for sample data during model training. For the second risk diagnosis network, the sample training labeled data contains detailed information and evaluation results regarding the epidemic risk spread trend, used to guide model learning and optimization. The second risk diagnosis network is a more refined and specialized network model than the first risk diagnosis network, used to further analyze and diagnose the sample knowledge vectors that have undergone preliminary diagnosis, in order to more accurately assess the epidemic risk spread trend. The second risk diagnosis result is the output of the second risk diagnosis network after processing the input sample knowledge vectors, and is usually more detailed and accurate than the first risk diagnosis result, providing more in-depth risk assessment information.

[0031] In this embodiment, the server extracts sample knowledge vectors with the attribute of risk spread trend from the sample knowledge vectors already loaded into the first risk diagnosis network. Continuing with the above-mentioned swine disease example, from the previously processed sample knowledge vectors related to swine fever, it identifies those knowledge vectors marked with risk spread trend (such as the rapid spread of swine fever in the breeding area).

[0032] The extracted sample knowledge vectors are then loaded into the second risk diagnosis network. This second risk diagnosis network is specifically designed to further refine the diagnosis of risk spread trends; it may have a different structure and parameters than the first risk diagnosis network. For example, it may focus more on analyzing specific patterns of risk spread, such as details like whether swine fever spreads through contact or airborne transmission.

[0033] After receiving these sample knowledge vectors, the second risk diagnosis network processes them and generates a second risk diagnosis result. For example, regarding the risk spread trend of swine fever, the second risk diagnosis network may diagnose that swine fever is mainly caused by the risk spread through direct contact between pigs, and give a specific risk spread assessment result based on factors such as the spread speed and contact frequency, such as the severity of the risk spread being "high" (which may be expressed numerically as 0.7).

[0034] Meanwhile, these examples focus on knowledge vectors with example training annotation data, which is based on in-depth analysis and professional judgment annotation of a large number of swine disease cases. For example, regarding the risk spread of swine fever through contact transmission, detailed information such as the specific risk spread level and the credibility of the transmission route is annotated based on actual observation and analysis.

[0035] Based on the second risk diagnosis results and the example training annotation data of the example attention knowledge vectors, the server optimizes the second risk diagnosis network. If the second risk diagnosis results match the example training annotation data, for example, diagnosing that swine fever is transmitted through contact and causes a high risk of spread (value of 0.7), and the example training annotation data also indicates that the risk of spread is high in this case, then the server will fine-tune the parameters of the second risk diagnosis network according to the optimization algorithm. For example, if the neuron weight corresponding to the knowledge vector related to swine fever contact transmission in the second risk diagnosis network was 0.4 before optimization, it may be adjusted to 0.41. If the second risk diagnosis results do not match the example training annotation data, for example, the diagnosis result is a risk of spread of "medium" (value of 0.5), while the annotation data is "high," then the server will significantly adjust the weights and other parameters to improve the accuracy of the network.

[0036] Step S140: Connect the optimized first risk diagnosis network with the second risk diagnosis network to generate a target risk diagnosis model. The target risk diagnosis model is used to diagnose whether there is a risk diffusion trend when the animal disease data mining model is performing data mining.

[0037] In this embodiment, the server connects the optimized first risk diagnosis network and the second risk diagnosis network to generate a target risk diagnosis model. For example, the first risk diagnosis network is mainly responsible for making a preliminary judgment on the general trend of risk spread in the data mining results of the animal disease data mining model, while the second risk diagnosis network performs a more in-depth and detailed analysis of the risk spread trend. Connecting them together forms a more comprehensive and accurate target risk diagnosis model.

[0038] This targeted risk diagnostic model is specifically designed to diagnose whether a risk spread trend exists during data mining by animal disease data mining models. For example, when an animal disease data mining model mines new swine disease case data, the targeted risk diagnostic model can accurately diagnose whether a risk spread trend of swine fever exists during the mining process. If the animal disease data mining model detects anomalies in the health data of pigs in a certain breeding area, the targeted risk diagnostic model can combine the preliminary judgment of the overall risk trend by the first risk diagnostic network and the judgment of the specific pattern and degree of risk spread by the second risk diagnostic network to accurately provide a diagnostic result on whether a risk spread trend of swine fever exists, thus providing a strong basis for decision-making in the prevention and control of animal diseases.

[0039] Based on the above steps, this application embodiment achieves efficient and accurate mining and analysis of animal disease case data, significantly improving the accuracy and timeliness of animal disease risk diagnosis. Specifically, firstly, by acquiring the cross-domain attention knowledge vectors generated during the mining process of the animal disease data mining model, and decomposing them into sample attention knowledge vectors representing risk diffusion trends and non-risk diffusion trends, it is possible to accurately capture key information directly related to disease risk diffusion, effectively avoiding information redundancy and improving the targeting of risk identification. By introducing a first risk diagnosis network and a second risk diagnosis network, this invention constructs a dual risk diagnosis mechanism. The first risk diagnosis network performs preliminary risk assessment, while the second risk diagnosis network conducts a more in-depth analysis of risk diffusion trends. This mechanism not only improves the accuracy of diagnosis but also ensures the comprehensiveness of the diagnostic results. The risk diagnosis network can automatically adjust network parameters based on the comparison between the diagnostic results and sample data, achieving self-optimization. This capability enables the model to continuously adapt to new epidemic situations and data characteristics, maintaining the stability and advancement of its diagnostic performance. By connecting the optimized first and second risk diagnosis networks, the generated target risk diagnosis model can quickly and accurately determine whether there is a risk spread trend in the animal disease data mining results. This not only greatly improves diagnostic efficiency but also ensures the accuracy and reliability of the diagnostic results, providing strong support for the timely prevention and control of animal diseases. Therefore, the technical solution in this embodiment is applicable to animal disease remote diagnosis center platforms, enabling real-time and remote mining and analysis of animal disease case data, providing decision-makers with timely and accurate epidemic information. This helps strengthen cross-regional cooperation in animal disease prevention and control, improving overall prevention and control effectiveness.

[0040] In one possible implementation, step S110 includes:

[0041] Step S111: Obtain animal disease case data, which includes a disease case description data and an epidemic condition description data. The epidemic condition description data is used to record whether there is a disease risk factor in the disease case description data and its disease change trend.

[0042] Step S112: Using the animal disease data mining model, obtain the disease knowledge feature chain based on the disease case description data, obtain the disease trend knowledge graph based on the disease condition description data, and generate the cross-domain attention knowledge vector of the disease knowledge feature chain based on the disease knowledge feature chain and the disease trend knowledge graph.

[0043] In this embodiment, taking a swine disease case as an example, the disease case description data includes a variety of information, such as the breed of the pig being a Landrace, its age being 6 months, its weight being 100 kg, the stocking density of the farm being 1.5 pigs per square meter, and recent feed changes. The disease severity description data records the changing trends of disease risk factors, such as whether the infection rate of classical swine fever virus is increasing in this herd, and whether the duration of abnormal body temperature in infected pigs is becoming longer.

[0044] Next, the server utilizes an animal disease data mining model to obtain disease knowledge feature chains based on disease case description data. The disease case feature extraction unit within the animal disease data mining model extracts features from the disease case description data. For example, features such as pig breed, age, and weight are quantified into disease case feature vectors using specific algorithms. For the Landrace pig breed, there might be corresponding codes in the vector based on breed characteristics; 6 months of age and 100 kg of weight are also converted into specific numerical representations. Then, the cross-domain association unit, based on pre-defined disease-related knowledge domain classification standards, divides each element in the disease case feature vector into different knowledge subsets. This classification standard might be based on factors such as disease type, symptom category, and disease source of infection. For example, a pig's age might be related to the disease source of infection and thus be classified into the corresponding knowledge subset; a pig breed might be related to disease susceptibility and also be classified into the corresponding subset. Based on the structure of the disease knowledge system and the prior requirements for animal disease diagnosis, a cross-domain knowledge association rule set for the cross-domain association unit is defined. For example, the Landrace pig breed may be more susceptible to swine fever in certain environments; this is an example of an association rule. For each knowledge subset, its elements are traversed, and the cross-domain knowledge association rule set is used to determine whether there is an association between that element and elements in other knowledge subsets. If so, the name or identifier of the other related knowledge subset is marked for that element. For example, if a correlation is found between pig weight and stocking density, the element of weight is marked as associated with the stocking density subset. Based on the knowledge subsets carrying association tags, a preliminary disease knowledge feature chain framework is constructed. Starting with core knowledge subsets, such as disease infection source-related subsets, related knowledge subsets are gradually connected along the association tags of elements, arranged in logical order to form a chain-like structure framework. The position of each knowledge subset is determined based on its association with other knowledge subsets and its importance weight in disease diagnosis. For example, the disease infection source-related subset is at the front of the chain because it plays a crucial role in the occurrence of the disease. For each knowledge subset in the initial disease knowledge feature chain framework, the elements in the knowledge subset are arranged according to the further refined association relationships based on the association tags, generating a refined disease knowledge feature chain framework. Finally, the refined disease knowledge feature chain framework is checked for completeness and optimized. From the perspective of the disease knowledge system, it is checked whether it covers all key knowledge areas of the disease. From the perspective of disease diagnosis, it is checked whether effective disease diagnosis can be carried out based on this framework. If there are any missing key information or unreasonable association relationships, such as the discovery that information related to a newly emerging swine fever transmission route is not covered, it is supplemented or adjusted to generate the disease knowledge feature chain.

[0045] Simultaneously, the server utilizes an animal disease data mining model to obtain a disease trend knowledge graph based on the disease description data. The disease feature extraction subunit represents the disease description data using graph features. For swine fever, the upward trend of the swine fever virus infection rate and changes in the duration of abnormal body temperature in infected pigs are constructed into a graph. Nodes represent different disease factors, and edges represent the relationships between factors; for example, there may be a positive correlation between an increase in the infection rate and a longer duration of abnormal body temperature.

[0046] Finally, based on the disease knowledge feature chain and the disease trend knowledge graph, the server generates a cross-domain attention knowledge vector for the disease knowledge feature chain. This vector integrates various correlation information in the disease knowledge feature chain and disease change trend information in the disease trend knowledge graph. For example, the correlation between factors such as pig breed, age, weight, and stocking density in the disease knowledge feature chain, combined with information such as the increasing trend of swine fever virus infection rate and the longer duration of abnormal body temperature in infected pigs in the disease trend knowledge graph, generates a cross-domain attention knowledge vector through a specific algorithm. This vector can comprehensively reflect the overall situation of animal disease case data in terms of disease knowledge features and disease trends, providing an important basis for subsequent analysis and diagnosis.

[0047] In one possible implementation, the animal disease data mining model includes a disease case feature extraction unit, a cross-domain association unit, and a knowledge reasoning unit, wherein the knowledge reasoning unit includes an epidemic condition feature extraction subunit and an epidemic condition feature restoration subunit.

[0048] Step S112 includes:

[0049] Step S1121: Using the disease case feature extraction unit in the animal disease data mining model, feature extraction is performed on the disease case description data to generate disease case feature vectors. Using the cross-domain association unit in the animal disease data mining model, the disease case feature vectors are cross-domain knowledge-associated to generate disease knowledge feature chains of the animal disease case data.

[0050] Step S1122: Using the disease condition feature extraction subunit of the animal disease data mining model, the disease condition description data is represented by graph features to generate a disease condition trend knowledge graph of the animal disease case data.

[0051] Step S1123: Using the epidemic condition feature restoration subunit of the animal disease data mining model, feature restoration is performed based on the cross-domain feature path data generated by fusing the disease knowledge feature chain with the condition trend knowledge graph, generating the mining result of the animal disease data mining model on the animal disease case data, and obtaining the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data by the animal disease data mining model when generating the mining result.

[0052] In this embodiment, the server utilizes the disease case feature extraction unit in the animal disease data mining model to extract features from disease case description data to generate disease case feature vectors. Taking pig disease cases as an example, the disease case description data includes a wealth of information such as the pig's age, breed, breeding environment, and recent contact history. The disease case feature extraction unit quantifies this information. If the pig's age is 3 months, it will be converted into a specific numerical code; if the breed is Duroc, it will also be encoded according to a pre-set coding system; parameters such as temperature and humidity in the breeding environment will also be converted into corresponding numerical representations. For example, a temperature of 20 degrees Celsius and humidity of 60% will be mapped to a specific numerical range. Combining these quantified data on the pig's age, breed, and breeding environment generates the disease case feature vector.

[0053] Next, the server utilizes the cross-domain association unit of the animal disease data mining model to perform cross-domain knowledge association on the feature vectors of disease cases to generate disease knowledge feature chains for animal disease case data. The cross-domain association unit first classifies disease-related knowledge domains according to pre-defined classification standards. These standards are based on factors such as disease type, symptom categories, and infection sources. For example, for swine diseases, disease types may be classified as viral, bacterial, etc.; symptom categories include fever, cough, etc.; and infection sources include pig-to-pig contact, feed contamination, etc. Elements in the disease case feature vectors are then assigned to different knowledge subsets according to this classification standard. For instance, a pig's age may be related to the infection source, so it is assigned to the infection source knowledge subset; breed is related to susceptibility to the disease, so it is assigned to the corresponding knowledge subset. Then, based on the structure of the disease knowledge system and the prior requirements for animal disease diagnosis, a set of cross-domain knowledge association rules is set. For example, younger pigs may be more susceptible to certain viral diseases, which is an association rule. For each knowledge subset, its elements are traversed, and the relationships between elements and elements in other knowledge subsets are determined based on the cross-domain knowledge association rule set. For example, if a Duroc pig breed is found to be associated with a specific disease source, the element is marked with the name or identifier of other related knowledge subsets. A preliminary disease knowledge feature chain framework is constructed based on the knowledge subsets carrying association tags. Starting with core knowledge subsets, such as disease source-related subsets, related knowledge subsets are gradually arranged and connected in logical order along the association tags of elements, forming a chain-like structure framework. The position of each knowledge subset is determined based on its association with other knowledge subsets and its importance weight in disease diagnosis. For example, the disease source subset occupies a key position in the disease knowledge feature chain framework because it plays an important role in the occurrence and spread of the disease. For each knowledge subset in the preliminary disease knowledge feature chain framework, the elements in the knowledge subset are again arranged according to further refined association relationships based on the association tags, generating a refined disease knowledge feature chain framework. Finally, the detailed disease knowledge feature chain framework is checked for completeness and optimized. From the perspective of disease knowledge system, it is checked whether it covers all key knowledge areas of the disease, and from the perspective of disease diagnosis, it is checked whether it can be used for effective diagnosis. If there is any missing key information, such as the lack of information on the transmission route of a new swine disease, or unreasonable correlation, it is supplemented or adjusted to generate the disease knowledge feature chain.

[0054] Subsequently, the server utilizes the disease condition feature extraction subunit of the animal disease data mining model to perform graph feature representation on the disease condition description data to generate a disease trend knowledge graph of animal disease case data. Taking swine fever as an example, the disease condition description data includes information such as changes in the infection rate of swine fever virus in the pig herd and the symptom development trend of infected pigs. The disease condition feature extraction subunit constructs this information into a graph structure. The infection rate of swine fever virus is used as one node, and the symptom development trend of infected pigs is used as another node. If there is a correlation between the increase in the infection rate and the aggravation of symptoms, the two nodes are connected by an edge. Different disease factors are used as nodes, and the relationships between factors are used as edges, thus constructing a disease trend knowledge graph. This graph can intuitively reflect the relationships between various factors in the disease condition description data and the development trend of the disease.

[0055] Finally, the server utilizes the epidemic condition feature restoration subunit of the animal disease data mining model to perform feature restoration based on the cross-domain feature path data generated by fusing the disease knowledge feature chain and the disease trend knowledge graph. This generates the mining results of the animal disease data mining model on the animal disease case data and obtains the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data during the generation of the mining results. The disease knowledge feature chain contains various disease-related features such as pig breed and age, as well as their correlations. The disease trend knowledge graph contains information such as the swine fever virus infection rate and the development trend of symptoms in infected pigs. The cross-domain feature path data after fusing the two contains rich and comprehensive information. The epidemic condition feature restoration subunit performs feature restoration on these cross-domain feature path data according to a specific algorithm. For example, based on the relationship between pig breed and disease susceptibility in the disease knowledge feature chain, combined with the changes in the swine fever infection rate in the disease trend knowledge graph, the risk assessment results of swine fever infection in pig herds and other mining results are restored. In the process of generating mining results, the server obtains the cross-domain attention knowledge vector of the disease knowledge feature chain of animal disease case data from the animal disease data mining model. This cross-domain attention knowledge vector integrates various relationships in the disease knowledge feature chain and information such as the disease development trend in the disease trend knowledge graph, providing an important basis for subsequent operations such as disease risk analysis.

[0056] In one possible implementation, step S1121 includes:

[0057] Step S1121-1: Using cross-domain association units, based on pre-set disease-related knowledge domain classification standards, each element in the disease case feature vector is divided into different knowledge subsets, which are divided based on the type of disease, symptom category, and source of disease infection.

[0058] Step S1121-2: Based on the structure of the disease knowledge system and the prior requirements for animal disease diagnosis, a set of cross-domain knowledge association rules for the cross-domain association unit is set. The setting of the cross-domain knowledge association rules revolves around the potential connections between different knowledge subsets.

[0059] Step S1121-3: For each knowledge subset, traverse the elements in the knowledge subset and determine whether there is a relationship between the element and the elements in other knowledge subsets based on the cross-domain knowledge association rule set. If there is a relationship, mark the element with the name or identifier of the other knowledge subsets associated with it, so that the elements in each knowledge subset clearly know the relationship with other knowledge subsets, and generate a knowledge subset with association mark.

[0060] Step S1121-4: Based on the knowledge subsets carrying association tags, construct a preliminary disease knowledge feature chain framework. The construction process of this preliminary disease knowledge feature chain framework starts from the core knowledge subset and connects related knowledge subsets step by step along the association tags of the elements. Specifically, the connection method is to arrange the knowledge subsets with relationships in a logical order to form a chain structure framework. In the chain structure framework, the position of each knowledge subset is determined according to the relationship between the knowledge subset and other knowledge subsets and the importance weight in disease diagnosis.

[0061] Step S1121-5: For each knowledge subset in the preliminary disease knowledge feature chain framework, the elements in the knowledge subset are arranged according to the advanced refined association relationship based on the association tag to generate a refined disease knowledge feature chain framework.

[0062] Step S1121-6 involves performing a completeness check and optimization on the refined disease knowledge feature chain framework. From the perspective of the disease knowledge system, this involves checking whether the refined disease knowledge feature chain framework covers all key knowledge areas of the disease. From the perspective of disease diagnosis, it involves checking whether the refined disease knowledge feature chain framework can be used for effective disease diagnosis. It also involves checking whether the refined disease knowledge feature chain framework has missing key information or unreasonable relationships. If the refined disease knowledge feature chain framework is found to have completeness issues, it is optimized. Optimization methods include supplementing missing knowledge elements, adjusting unreasonable relationships, and rearranging the order of knowledge subsets or elements. This generates the disease knowledge feature chain, which reflects the disease knowledge features in the animal disease case data and the relationships between these disease knowledge features.

[0063] In this embodiment, firstly, the server utilizes cross-domain association units to classify each element in the disease case feature vector into different knowledge subsets based on pre-defined disease-related knowledge domain classification standards. These knowledge subsets are divided according to the type of disease, symptom category, and source of infection. Taking a pig disease case as an example, assume the disease case feature vector includes elements such as the pig's age (3 months), breed (Landrace), weight (50 kg), recent contact history (contact with pigs from another farm), and the sanitary conditions of the farm (general). According to the disease-related knowledge domain classification standards, elements such as age and weight may be related to the source of infection. For example, young and lighter pigs may be more sensitive to certain disease sources, so these elements will be classified into the disease source knowledge subset. The pig breed may be related to the type of disease; the Landrace breed may have a higher susceptibility to certain specific types of diseases, so the breed element is classified into the disease type knowledge subset. Recent contact history and farm hygiene conditions are clearly related to the source of disease infection and are also classified into the knowledge subset of the source of disease infection.

[0064] Next, based on the structure of the disease knowledge system and the prior requirements for animal disease diagnosis, the server sets a set of cross-domain knowledge association rules for cross-domain association units. The setting of these rules revolves around the potential connections between different knowledge subsets. For swine diseases, from the perspective of the disease knowledge system structure, different types of diseases may be associated with specific symptom categories and sources of infection. For example, viral diseases may be more easily transmitted in environments with high pig density and poor sanitation, and may be accompanied by symptoms such as fever and cough. From the perspective of the prior requirements for animal disease diagnosis, to quickly diagnose a viral disease like swine fever, it is necessary to focus on the correlation between factors such as pig breed, age, recent contact history, and farm sanitation. Therefore, the established association rules might include: if the pig breed is Landrace and the pig is young, the risk of infection with viral diseases (such as swine fever) increases after contact with outside pigs and when the farm sanitation is poor.

[0065] Then, for each knowledge subset, the server iterates through the elements within the subset and, based on the cross-domain knowledge association rule set, determines whether the element has any association relationship with elements in other knowledge subsets. If such a relationship exists, the server marks the element with the name or identifier of the associated other knowledge subset, ensuring that each element within a knowledge subset clearly understands its association relationship with other knowledge subsets, thus generating a knowledge subset with association markers. Continuing with the example of swine diseases, in the disease infection source knowledge subset, for the element "recent contact history," according to the association rules, it is found to be associated with "viral diseases" in the disease type knowledge subset. Because recent contact with imported pigs increases the risk of viral disease infection, the element "recent contact history" is marked with an association marker with "viral diseases." Similarly, in the disease type knowledge subset, for the element "swine fever," because swine fever is more easily transmitted in young pigs (age element), the element "swine fever" is marked with an association marker with the "age" element in the disease infection source knowledge subset. This generates knowledge subsets with association markers.

[0066] Subsequently, based on the knowledge subsets carrying association tags, the server constructs a preliminary disease knowledge feature chain framework. The construction process of this preliminary disease knowledge feature chain framework begins with the core knowledge subset and gradually connects related knowledge subsets along the association tags of the elements. Specifically, the connection method involves arranging related knowledge subsets in logical order to form a chain-like structure framework. In this chain-like structure framework, the position of each knowledge subset is determined based on its association with other knowledge subsets and its importance weight in disease diagnosis. In the case of swine diseases, the disease infection source knowledge subset may be one of the core knowledge subsets because it is the source factor of disease transmission. Starting from this core knowledge subset, based on the association tags, if a disease infection source is found to be associated with a disease type, the disease type knowledge subset is connected to the disease infection source knowledge subset. For example, because factors such as pig contact history and hygiene conditions are associated with the swine fever disease type, the disease infection source knowledge subset and the disease type knowledge subset are connected in this logical order. When determining the position of each knowledge subset, its importance weight in disease diagnosis is considered. For example, the source of infection is crucial in the early diagnosis of swine diseases, so it is positioned relatively early in the chain structure framework; while the type of disease, although also important, is more helpful for further diagnosis after the source of infection is identified, so it is positioned relatively late.

[0067] For each knowledge subset in the initial disease knowledge feature chain framework, the server further arranges the elements within the knowledge subset according to more refined association relationships based on association tags, generating a more detailed disease knowledge feature chain framework. In the disease infection source knowledge subset, elements such as the pig's recent contact history and farm hygiene conditions are originally included. Based on more detailed association relationships, it may be discovered that different origins of pigs from other regions in the recent contact history have different impacts on the risk of disease infection; for example, pigs from high-incidence areas pose a higher risk of infection. Therefore, according to this more refined association relationship, the elements in the recent contact history are arranged according to the risk level of the pigs' origin. Similarly, in the disease type knowledge subset, for diseases like classical swine fever (CSF), the relevant elements may be more refined based on the differences in pathogenicity of different strains of the CSF virus to pigs of different ages and breeds.

[0068] Finally, the server performs a completeness check and optimization on the refined disease knowledge feature chain framework. From the perspective of the disease knowledge system, it checks whether the refined disease knowledge feature chain framework covers all key knowledge areas of the disease. For example, for swine diseases, in addition to the already considered factors such as breed, age, contact history, sanitary conditions, and disease type, it is also necessary to check whether factors that may affect the disease, such as vaccination status and feed source, are included. If it is found that the key knowledge area of ​​vaccination status is not covered, relevant knowledge elements need to be added. From the perspective of disease diagnosis, it checks whether the refined disease knowledge feature chain framework can be used for effective disease diagnosis, and whether there are any missing key information or unreasonable associations. For example, if it is found that the association between pig breed and a certain disease is incorrectly set in the current refined framework, such as Landrace pigs being incorrectly set to be highly susceptible to a certain disease when they are actually less susceptible, this is an example of an unreasonable association. When the refined disease knowledge feature chain framework is found to have incompleteness issues, it is optimized. Optimization methods include supplementing missing knowledge elements, adjusting unreasonable relationships, and rearranging the order of knowledge subsets or elements. If feed source-related knowledge elements are missing, they are added; if unreasonable relationships exist, such as the incorrect association between breed and disease susceptibility, they are adjusted; if the order of knowledge subsets or elements affects diagnostic efficiency or accuracy, their order is rearranged. Through this completeness check and optimization process, a disease knowledge feature chain is ultimately generated. This chain reflects the disease knowledge features in animal disease case data and the relationships between these features, providing a comprehensive and accurate basis for subsequent disease analysis, diagnosis, and prevention and control operations.

[0069] In one possible implementation, the method further includes:

[0070] Step A110: Obtain sample disease case description data sequence. For each sample disease case description data in the sample disease case description data sequence, obtain the prior disease change trend of the sample disease case description data. The prior disease change trend refers to the disease change trend of disease risk factors existing in the sample disease case description data.

[0071] Step A120: Obtain the epidemic condition description template. The epidemic condition description template is used to record the trend of disease condition changes in the epidemic case description data, including whether there are epidemic risk factors.

[0072] Step A130: Based on the prior disease change trend of the sample epidemic case description data and the epidemic disease description template, generate sample epidemic disease description data. One prior disease change trend generates one sample epidemic disease description data. The sample epidemic disease description data is used to record whether the prior disease change trend exists in the sample epidemic case description data.

[0073] Step A140: Based on the sample disease case description data and the sample epidemic condition description data, configure animal disease case data as positive samples.

[0074] In this embodiment, the first step is to acquire a sequence of sample disease case description data. Taking pig disease as an example, this sequence may contain disease case description data from multiple different pig herds or individual pigs. For example, for pig herd A, the disease case description data includes information such as the breed of pig being Large White pigs, the herd size being 500 pigs, the temperature range of the rearing environment being 18-22 degrees Celsius, and the feed composition being mainly corn and soybean meal; for another pig herd B, the disease case description data might be that the breed of pig being Duroc pigs, the herd size being 300 pigs, the temperature range of the rearing environment being 20-23 degrees Celsius, and that there has been a recent change in some feed, etc. For each sample disease case description data in this sequence, the server needs to acquire the prior disease change trend of the sample disease case description data. Here, the prior disease change trend refers to the disease change trend of the disease risk factors existing in the sample disease case description data. In the case of pig herd A, if there is a classical swine fever virus infection, the preliminary trend of disease change may be that the number of infected pigs increased from 5 to 10 in the past week, indicating that the classical swine fever virus infection is on the rise in pig herd A. For pig herd B, if there is a certain bacterial disease, the preliminary trend of disease change may be that the duration of abnormal body temperature in sick pigs increased from an average of 2 days to 3 days, reflecting that the disease is worsening in pig herd B.

[0075] Next, the server retrieves the epidemic disease description template. This template is used to record the trend of disease changes in the disease case description data, including whether there are risk factors associated with the disease. For example, the epidemic disease description template may include recording items for multiple risk factors for swine diseases, such as viral infection rate, bacterial infection range, and changes in the severity of symptoms in infected animals. It also specifies how to record the trend of these factors, whether to record the increase or decrease in the number of infections according to time series (e.g., daily, weekly) or to record the aggravation or relief of the disease according to quantitative indicators of symptoms (e.g., changes in body temperature, cough frequency, etc.).

[0076] Then, based on the prior disease trend and epidemic disease description template of the sample disease case description data, the server generates sample epidemic disease description data. One sample epidemic disease description data is generated for each prior disease trend. The sample epidemic disease description data is used to record whether there is a prior disease trend in the sample disease case description data. For the swine fever outbreak in pig herd A, based on the prior disease trend (the number of infected pigs increased from 5 to 10 in the past week) and the epidemic disease description template (recording the change in the number of virus infections weekly), the generated sample epidemic disease description data will clearly record that the number of swine fever virus-infected pigs in pig herd A showed an upward trend this week. For the bacterial disease in pig herd B, based on the prior disease trend (the duration of abnormal body temperature in sick pigs increased from an average of 2 days to 3 days) and the epidemic disease description template (recording the change in disease according to the duration of symptoms in sick animals), the generated sample epidemic disease description data will record that the duration of abnormal body temperature in sick pigs with this bacterial disease in pig herd B showed a prolonged trend.

[0077] Finally, based on the sample disease case description data and sample epidemic condition description data, the server was configured with animal disease case data as positive samples. For pig herd A, the sample epidemic condition description data, including the Large White pig breed, the number of pigs (500), the breeding environment temperature (18-22 degrees Celsius), the feed composition mainly composed of corn and soybean meal, and the increase in the number of pigs infected with classical swine fever virus, was combined to form a positive sample of animal disease case data. For pig herd B, the sample epidemic condition description data, including the Duroc pig breed, the number of pigs (300), the breeding environment temperature (20-23 degrees Celsius), recent feed changes, and the prolonged duration of abnormal body temperature in pigs infected with bacterial diseases, was also integrated and configured as a positive sample of animal disease case data. These positive sample animal disease case data can accurately reflect the occurrence and development of diseases, containing various characteristics of disease cases and important information such as disease trend changes, providing a data foundation with practical significance and value for subsequent disease data mining, analysis, diagnosis, and training and optimization of related models. For example, when training animal disease data mining models, these positive samples can serve as correct examples, helping the model learn the relationship between disease-related features and disease progression trends, thereby improving the model's ability to identify and analyze diseases. This enables the model to more accurately predict the development of diseases and take effective control measures in actual disease prevention and control work.

[0078] In one possible implementation, the method further includes:

[0079] Step B110: Determine the negative sample disease condition of the sample disease case description data based on the prior disease condition change trend of the sample disease case description data. The negative sample disease condition is the disease condition change trend of disease risk factors that do not exist in the sample disease case description data.

[0080] Step B120: Based on the negative sample condition of the sample epidemic case description data and the epidemic condition description template, generate sample epidemic condition description data. One negative sample condition generates one sample epidemic condition description data. The sample epidemic condition description data is used to record whether the negative sample condition exists in the sample epidemic case description data.

[0081] Step B130: Based on the sample disease case description data and the sample epidemic condition description data, configure animal disease case data as negative samples.

[0082] In this embodiment, the negative sample disease status of the sample disease case description data is first determined based on the prior disease status change trend of the sample disease case description data. Here, negative sample disease status refers to the disease status change trend of disease risk factors that are not present in the sample disease case description data. Taking a pig disease case as an example, for sample disease case description data of a certain pig herd, it is assumed that the prior disease status change trend indicates that the swine fever virus infection rate is gradually increasing, the number of infected pigs is constantly increasing, and the body temperature of sick pigs is abnormally high. Based on this, the server determines the negative sample disease status. One way to determine the negative sample disease status is to determine multiple negative sample disease statuses of the sample disease case description data based on the prior disease status change trend, generating a negative sample disease status sequence of the sample disease case description data. For example, for a swine fever epidemic, negative sample disease status opposite to an increase in the virus infection rate could be a stable or decreasing virus infection rate; opposite to an increase in the number of infected pigs, it could be a decrease in the number of infected pigs; opposite to a large abnormal increase in the body temperature of sick pigs, it could be a decrease in the abnormal increase in body temperature or a return to normal body temperature, etc. These negative sample disease statuses constitute a negative sample disease status sequence. Then, a negative sample disease sequence is randomly selected from this negative sample disease sequence to generate a negative sample disease. For example, the negative sample disease of a decrease in the number of infected pigs is randomly selected.

[0083] Alternatively, the server can obtain the prior disease trend of other sample disease case description data besides the sample disease case description data from the sample disease case description data sequence. From the prior disease trend of other sample disease case description data, negative sample disease conditions of the sample disease case description data can be extracted, generating a negative sample disease condition sequence. Then, a negative sample disease condition is generated by randomly selecting from this negative sample disease condition sequence. For example, in the sample disease case description data of another pig herd, if the prior disease trend shows that the infection rate of swine influenza is low and stable, then this stable swine influenza infection rate can be included as one of the negative sample disease conditions in the current swine fever epidemic sample disease case description data and added to the negative sample disease condition sequence. After random selection, if this negative sample disease condition is selected, it is determined as the negative sample disease condition of that sample disease case description data.

[0084] Alternatively, the server determines the shared disease trend of the prior disease trend in the sample disease case description data. This shared disease trend represents the disease risk factor trend that is not present in the sample disease case description data but has a correlation with the prior disease trend. Based on this shared disease trend, a negative sample disease sequence is generated from the sample disease case description data. A negative sample disease is then randomly selected from this sequence and used as the negative sample disease in the sample disease case description data. For example, in the sample disease case description data for swine fever, the prior disease trend is an increase in swine fever virus infection rate. Swine fever virus infection may be related to the state of the pig's immune system. In other diseases related to the immune system, there is a possibility that a certain bacterial infection can enhance the pig's immune system, thus providing resistance to swine fever. Therefore, this increase in bacterial infection rate (which has a correlation with the prior disease trend) can be included as a shared disease trend in the negative sample disease sequence. If this disease is selected after random sampling, it becomes the negative sample disease.

[0085] Next, based on the negative sample disease conditions and the epidemic disease description template from the sample disease case description data, the server generates sample epidemic disease description data. One negative sample disease condition generates one sample epidemic disease description data set, which is used to record whether a negative sample disease condition exists in the sample disease case description data. For example, for the previously identified negative sample disease condition of a decrease in the number of infected pigs in a swine fever epidemic, according to the epidemic disease description template (which specifies the weekly changes in the number of infected pigs), the generated sample epidemic disease description data will record a decreasing trend in the number of swine fever virus-infected pigs in that sample disease case description data.

[0086] Finally, based on the sample disease case description data and sample epidemic condition description data, the server is configured with animal disease case data as negative samples. For the example of swine fever, sample disease case description data including pig breed, herd size, breeding environment, and feed conditions, along with sample epidemic condition description data recording the decreasing trend in the number of infected pigs, are combined to form a negative sample of animal disease case data. These negative sample animal disease case data correspond to the positive sample animal disease case data and are crucial in subsequent animal disease data mining model training, optimization, and analysis. For example, during model training, positive samples allow the model to learn the normal development patterns and characteristics of diseases, while negative samples allow the model to better identify situations different from the normal pattern, thereby improving the model's accuracy and generalization ability. In disease risk assessment, negative sample data can provide a basis for comparison with positive sample data, helping to more comprehensively analyze the development trend of diseases and providing a reference for formulating reasonable disease prevention and control strategies.

[0087] In one possible implementation, step B110 includes:

[0088] Based on the prior disease change trend of the sample disease case description data, multiple negative sample disease conditions are determined from the sample disease case description data. A negative sample disease condition sequence is generated from the sample disease case description data. A single negative sample disease condition is then randomly selected from this sequence and used as the negative sample disease condition of the sample disease case description data. Alternatively,

[0089] Obtain the prior disease change trends of other sample disease case description data besides the sample disease case description data in the sample disease case description data sequence. Extract the negative sample disease condition from the prior disease change trends of the other sample disease case description data, generate the negative sample disease condition sequence of the sample disease case description data, and randomly select a negative sample disease condition from the negative sample disease condition sequence to generate a negative sample disease condition as the negative sample disease condition of the sample disease case description data. Alternatively,

[0090] A shared disease change trend is determined from the prior disease change trend of the sample disease case description data. The shared disease change trend is the disease change trend of disease risk factors that do not exist in the sample disease case description data but have a correlation feature with the prior disease change trend. Based on the shared disease change trend, a negative sample disease sequence is generated from the sample disease case description data. A negative sample disease is generated by randomly selecting from the negative sample disease sequence, which serves as the negative sample disease of the sample disease case description data.

[0091] In this embodiment, when the condition of multiple negative samples in the sample disease description data is determined based on the prior disease change trend of the sample disease case description data, a negative sample condition sequence is generated, and then randomly selected, the operation process is as follows.

[0092] Suppose there is sample disease case description data of a pig herd, in which the prior disease trend shows that the infection rate of swine fever virus has been rising continuously over the past month, from the initial 5% to 15%. At the same time, the severity of clinical symptoms of sick pigs has also increased. For example, the average body temperature of pigs with fever has increased from 40°C to 41°C, and the symptoms of rapid breathing have become more obvious, with the respiratory rate increasing from 40 breaths per minute to 50 breaths per minute.

[0093] Based on this prior trend in disease progression, the server begins to identify multiple negative sample conditions. For a scenario where the swine fever virus infection rate is rising, negative sample conditions could be either a stable or declining infection rate. For example, the infection rate might remain at 5% or decrease to 3%. For cases where the fever and rapid breathing symptoms in infected pigs worsen, negative sample conditions could be either a stable or declining body temperature (e.g., a temperature remaining at 40°C or decreasing to 39.5°C) or a reduction in rapid breathing (e.g., a respiratory rate decreasing to 35 breaths per minute or returning to a normal 30 breaths per minute).

[0094] The server aggregates these negative sample disease conditions, generating a negative sample disease condition sequence for the case description data. This sequence includes various possible scenarios that contradict the prior disease trend, covering multiple aspects such as the swine fever virus infection rate and symptoms in sick pigs. The server then randomly selects from this negative sample disease condition sequence. For example, if the random selection selects a negative sample disease condition where the swine fever virus infection rate drops to 3%, the sick pig's body temperature remains at 40℃, and the respiratory distress symptoms lessen to 35 breaths per minute, this will be used as the negative sample disease condition for the case description data.

[0095] Alternatively, one can obtain the prior disease change trend of other sample disease case description data besides the sample disease case description data in the sample disease case description data sequence, extract the negative sample disease condition from the sample disease case description data, and then generate a negative sample disease condition sequence and randomly select it.

[0096] For example, consider a sequence of sample disease case descriptions from a group of pigs. The sample disease case descriptions from pig herd A are of current interest. Their prior disease trend shows a gradually increasing infection rate of swine influenza virus, with infected pigs exhibiting progressively worsening lethargy. Meanwhile, in the sample disease case descriptions from pig herd B, the prior disease trend shows that the infection rate of swine influenza virus begins to decline after a period of time, and the mental state of infected pigs shows signs of gradual improvement.

[0097] The server extracts negative sample disease conditions related to pig herd A from the prior disease trend of pig herd B. For example, if pig herd A experiences an increase in swine influenza virus infection rate and worsening lethargy in infected pigs, the negative sample disease conditions extracted from pig herd B would be a decrease in swine influenza virus infection rate and an improvement in the mental state of infected pigs. The server combines these negative sample disease conditions extracted from other sample disease case description data to generate a negative sample disease condition sequence for sample disease case description data (pig herd A). Then, this sequence is randomly selected; assuming the randomly selected sequence represents the negative sample disease conditions of decreasing swine influenza virus infection rate and improved mental state of infected pigs, it is identified as the negative sample disease condition for the sample disease case description data of pig herd A.

[0098] In addition, the server can generate negative sample disease sequences and randomly select them by determining the shared disease change trend of the prior disease change trend of sample disease case description data.

[0099] Suppose that in the sample disease case description data of a certain pig herd, the prior disease trend is an increase in porcine circovirus infection rate, accompanied by a decline in the pigs' immune function, manifested as a decrease in the proportion of lymphocytes and a reduction in the number of white blood cells. The server needs to determine the shared disease trend, that is, the disease trend of disease risk factors that are not present in the sample disease case description data but have a correlation with the prior disease trend.

[0100] Analysis of disease knowledge and related data revealed a close correlation between the quantity of a certain probiotic in the pig's gut and the pig's immune function. Studies of other pig populations showed that an increase in the quantity of this probiotic in the pig's gut enhanced its immune function, effectively resisting porcine circovirus infection. Therefore, this increase in the quantity of this probiotic represents a shared disease trend correlated with prior disease trends (increased porcine circovirus infection rate and decreased pig immune function).

[0101] Based on this shared trend of disease changes, the server generates a negative sample disease sequence from the sample disease case description data. For example, different degrees of increase in probiotic count can constitute this sequence, such as increases of 10%, 20%, etc. Then, this negative sample disease sequence is randomly selected. Suppose a negative sample disease with a 20% increase in probiotic count is selected, it will be used as the negative sample disease in the sample disease case description data.

[0102] By employing the three methods described above, the disease status of negative samples can be accurately determined based on the prior disease trend of the sample disease case description data. This lays the foundation for subsequent generation of sample disease disease description data and configuration of animal disease case data with negative samples, and is of great significance in the training and optimization of animal disease data mining models and disease risk analysis. For example, in model training, the introduction of negative sample disease status enables the model to better identify the differences between different disease development patterns, thereby improving the accuracy and comprehensiveness of the model's judgment on the disease situation.

[0103] In one possible implementation, the method further includes:

[0104] Step C110: Obtain the prior disease change trend of each sample disease case description data in the sample disease case description data sequence.

[0105] Step C120: Based on the sample disease case description data sequence, calculate the correlation between any two prior disease change trends. Two different prior disease change trends appearing simultaneously in the same sample disease case description data are considered as one correlation.

[0106] Step C130: When the correlation between two different prior disease change trends meets the target requirements, it is determined that the two different prior disease change trends have a correlation feature relationship.

[0107] In this embodiment, the prior disease trend of each sample disease case description in the sample disease case description data sequence is first obtained. Taking swine disease cases as an example, in the sample disease case description data sequence, for the sample disease case description data of pig herd A, it may contain information related to classical swine fever. Its prior disease trend is that the classical swine fever virus infection rate has increased from 5% to 10% in the past week, and the abnormal increase in body temperature of sick pigs has increased from an average of 39.5℃-40℃ to 40℃-40.5℃. For the sample disease case description data of pig herd B, if it is about swine influenza, the prior disease trend may be that the number of infected pigs has increased from 20 to 30 in the recent period, and the severity of coughing in sick pigs has intensified, with the coughing frequency increasing from 10 times per hour to 15 times per hour. The server will obtain the prior disease trend in the sample disease case description data of these different pig herds one by one.

[0108] Next, based on the sample disease case description data sequence, the correlation between any two prior disease trends is calculated. Suppose we want to calculate the correlation between the prior disease trend of classical swine fever (CSF) in pig herd A and the prior disease trend of swine influenza (SIA) in pig herd B. Since the definition of correlation is that two different prior disease trends appearing simultaneously in the same sample disease case description data is considered a single correlation, it is necessary to traverse the entire sample disease case description data sequence. For example, in sample disease case description data where some pig herds are simultaneously infected with both CSF and SIA, if there are 10 such pig herds, and the simultaneous increase in CSF virus infection rate and the increase in the number of infected pigs with SIA occurs 3 times in these herds, then according to the correlation calculation method (number of correlations divided by the total number of samples), the correlation between these two prior disease trends is 3 / 10 = 0.3. As another example, let's calculate the correlation between the prior disease trend of increased CSF virus infection rate and increased abnormal body temperature in infected pigs in pig herd A and the prior disease trend of weight loss in pigs infected with a certain bacterial disease in pig herd C. By traversing the sample disease case description data sequence, it was found that there are 2 pig herds that simultaneously exhibit these three disease change trends. Assuming there are a total of 8 pig herds containing related diseases, the correlation between them is 2 / 8 = 0.25.

[0109] Finally, when the correlation between two different prior disease trends meets the target requirement, the server determines that these two different prior disease trends have a correlation characteristic relationship. For example, the target requirement is that a correlation characteristic relationship exists when the correlation is greater than 0.2. The correlation between the previously calculated prior disease trends of classical swine fever in pig herd A and classical swine influenza in pig herd B is 0.3. Since 0.3 is greater than 0.2, the server determines that the two prior disease trends of increased classical swine fever virus infection rate and increased number of pigs infected with classical swine influenza have a correlation characteristic relationship. Similarly, the correlation between the prior disease trends of increased classical swine fever virus infection rate and increased abnormal body temperature in sick pigs in pig herd A and the prior disease trend of weight loss in pigs infected with a certain bacterial disease in pig herd C is 0.25. Since 0.25 is greater than 0.2, a correlation characteristic relationship is also determined between them. The determination of this correlation characteristic relationship helps to gain a deeper understanding of the intrinsic relationship between different diseases or between different disease trends of the same disease, providing an important basis for comprehensive disease analysis, diagnosis, and the formulation of prevention and control strategies. For example, in disease prevention and control, if a correlation is found between the increase in the infection rate of swine fever virus and the increase in the number of pigs infected with swine influenza, then while controlling swine fever, more attention should be paid to the prevention and control of swine influenza to prevent the two diseases from affecting each other and causing the epidemic to worsen.

[0110] In one possible implementation, step S110 may further include:

[0111] Obtain the mining results of the animal disease case data by the animal disease data mining model, wherein the mining result is one of the first mining result and the second mining result.

[0112] Based on the categories of the mining results and the label data of the animal disease case data, the cross-domain attention knowledge vector is decomposed into sample attention knowledge vectors of the first type of risk diffusion trend, sample attention knowledge vectors of the second type of risk diffusion trend, sample attention knowledge vectors of the first type of non-risk diffusion trend, or sample attention knowledge vectors of the second type of non-risk diffusion trend.

[0113] Step S130 includes:

[0114] Step S131: Extract the sample attention knowledge vectors with the attention tag attribute of the first type of risk diffusion trend from the sample attention knowledge vectors loaded into the first risk diagnosis network, and continue to load them into the second risk diagnosis network of the first network label, generate the second risk diagnosis result of the second risk diagnosis network of the first network label on the sample attention knowledge vectors, and optimize the second risk diagnosis network of the first network label based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors.

[0115] Step S132: Extract sample attention knowledge vectors with the attention tag attribute of the second type of risk diffusion trend from the sample attention knowledge vectors loaded into the first risk diagnosis network, continue to load them into the second risk diagnosis network of the second network label, generate the second risk diagnosis result of the second risk diagnosis network of the second network label on the sample attention knowledge vectors, and optimize the second risk diagnosis network of the second network label based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors.

[0116] Step S140 includes:

[0117] Step S141: Connect the optimized first risk diagnosis network with the second risk diagnosis network labeled with the first network tag to generate a target risk diagnosis model. This target risk diagnosis model is used to diagnose whether there is a risk diffusion trend when the animal disease data mining model obtains the first mining result.

[0118] Step S142: Connect the optimized first risk diagnosis network with the second risk diagnosis network with the second network label to generate a target risk diagnosis model. The target risk diagnosis model is used to diagnose whether there is a risk diffusion trend when the animal disease data mining model obtains the second mining result through risk diagnosis.

[0119] In this embodiment, firstly, taking swine diseases as an example, it is assumed that the first data mining result indicates that the overall infection situation of swine fever in a certain area is in the initial development stage, the number of infected pigs is relatively small but has a slow upward trend, and the spread of the virus is limited to a few farms; the second data mining result may indicate the outbreak of swine influenza in the area, with the number of infected pigs increasing significantly in a short period of time, the spread involving multiple farms and a trend of spreading to surrounding areas. The label data of animal disease case data includes pre-labeled information such as the accurate type of disease, the severity of the disease, and whether there is a risk of spread. For example, for swine fever case data, the label data may be labeled as swine fever type, moderate severity, and currently has a low risk of spread; for swine influenza case data, the label data may be labeled as swine influenza type, high severity, and has a high risk of spread.

[0120] Furthermore, regarding the case of swine fever, if the mining result is the first mining result (initial development stage), based on the low-risk spread trend in the tag data, the server will extract sample attention knowledge vectors for the first type of risk spread trend related to this low-risk spread trend from the cross-domain attention knowledge vector. This vector may contain knowledge elements related to factors such as the slow spread of the swine fever virus within farms and the limited activity range of infected pigs. Simultaneously, it will also extract sample attention knowledge vectors for the first type of non-risk spread trend, which may contain knowledge elements related to factors that inhibit virus spread, such as good disease prevention measures at farms and relatively high immunity in pig herds. Regarding the case of swine influenza, if the mining result is the second mining result (outbreak and spread), based on the high-risk spread trend in the tag data, the server will extract sample attention knowledge vectors for the second type of risk spread trend. This may contain knowledge elements related to high-risk spread, such as the high infectivity of the swine influenza virus and high-density contact between pig herds, as well as sample attention knowledge vectors for the second type of non-risk spread trend, such as the possible protective effect of localized control measures on some pig herds.

[0121] Next, for the sample knowledge vectors related to the first type of risk spread trend, the server extracts the sample knowledge vectors with the attribute of "first type of risk spread trend" from the sample knowledge vectors loaded into the first risk diagnosis network, and continues to load them into the second risk diagnosis network labeled by the first network. For example, after the sample knowledge vectors related to the first type of risk spread trend of swine fever are loaded into the second risk diagnosis network labeled by the first network, this network will analyze and process them. Assume that the second risk diagnosis network labeled by the first network is a diagnostic network based on a neural network, which has been pre-trained with a large amount of swine disease data. This network analyzes the sample knowledge vectors related to swine fever, and the knowledge elements contained therein, such as the spread rate of the swine fever virus and the activity range of infected pigs, are processed by the neurons in the network. Based on its own algorithm and parameters, the network generates the second risk diagnosis result of the second risk diagnosis network labeled by the first network for the sample knowledge vectors. For example, the diagnosis result may indicate that the risk spread trend of swine fever may remain at a low level in the short term, but if certain factors (such as changes in the breeding environment or the introduction of new pig herds) change, the risk may increase. Then, based on this second risk diagnosis result and the example training labeled data of the example attention knowledge vector, the second risk diagnosis network labeled by the first network is optimized. The example training labeled data contains an accurate assessment of the spread trend of swine fever risk, such as the risk spread speed and impact range labeled according to the actual epidemic development data. If there is a deviation between the second risk diagnosis result and the example training labeled data, for example, if the diagnosis result underestimates the risk spread speed, the server will adjust the parameters in the second risk diagnosis network labeled by the first network according to a specific optimization algorithm (such as the backpropagation algorithm). For example, the weights of neurons related to the spread speed of swine fever virus may be appropriately increased to improve the accuracy of the network in diagnosing the spread trend of swine fever risk.

[0122] Similarly, for the sample attention knowledge vector of the second type of risk spread trend, the server extracts the sample attention knowledge vector with the attention tag attribute of the second type of risk spread trend from the sample attention knowledge vector loaded into the first risk diagnosis network, and continues to load it into the second risk diagnosis network of the second network label. Taking swine flu as an example, when the sample attention knowledge vector of the second type of risk spread trend related to swine flu is loaded into the second risk diagnosis network of the second network label, this network begins to perform diagnosis. The neurons in the network process knowledge elements such as the high infectivity of the swine flu virus and the high density of contact in pig herds, generating the second risk diagnosis result of the second risk diagnosis network of the second network label on the sample attention knowledge vector. For example, the diagnosis result may indicate that the risk spread trend of swine flu is very high under the current conditions and may further expand its spread in a short period of time. Then, the second risk diagnosis network of the second network label is optimized based on this second risk diagnosis result and the sample training annotation data of the sample attention knowledge vector. The sample training annotation data contains detailed annotations on the risk spread trend of swine flu, such as the accurate value of the spread speed and the infection range in different time periods. If the diagnostic results do not perfectly match the sample training and labeled data—for example, if the diagnostic results overestimate the spread rate—the server will adjust the parameters in the second risk diagnosis network of the second network label through optimization algorithms. For instance, the weights of neurons related to the infectivity of the swine flu virus may be appropriately reduced so that the network can more accurately diagnose the risk spread trend of swine flu.

[0123] Finally, the server connects the optimized first risk diagnosis network with the second risk diagnosis network to generate the target risk diagnosis model.

[0124] In cases where the optimized first risk diagnosis network is connected to the second risk diagnosis network labeled with the first network label to generate a target risk diagnosis model, this target risk diagnosis model is used to diagnose whether a risk spread trend exists when the animal disease data mining model obtains the first mining result. For example, in the case of swine fever, when the animal disease data mining model obtains the first mining result (initial development stage), the target risk diagnosis model can integrate the preliminary judgment of the first risk diagnosis network on the overall situation of swine fever and the further analysis of the risk spread trend of swine fever by the second risk diagnosis network labeled with the first network label. The first risk diagnosis network may have already made a preliminary assessment of the basic symptoms and infection range of swine fever, while the second risk diagnosis network labeled with the first network label focuses on diagnosing the first type of risk spread trend of swine fever. After connecting them, the target risk diagnosis model can more accurately determine whether a risk spread trend of swine fever exists when the animal disease data mining model obtains the first mining result.

[0125] In the case where the optimized first risk diagnosis network is connected to the second risk diagnosis network with the second network label to generate a target risk diagnosis model, this target risk diagnosis model is used to diagnose whether there is a risk spread trend when the animal disease data mining model obtains the second mining result. Taking swine influenza as an example, when the animal disease data mining model obtains the second mining result (outbreak and spread), the target risk diagnosis model combines the preliminary judgment of the first risk diagnosis network on the overall situation of swine influenza and the in-depth analysis of the risk spread trend of the second type of swine influenza by the second network label. The first risk diagnosis network may have already determined the basic situation of swine influenza, such as the severity and the number of infected pigs, while the second risk diagnosis network with the second network label focuses on analyzing the knowledge elements related to the high-risk spread trend of swine influenza. The connected target risk diagnosis model can more accurately determine whether there is a risk spread trend of swine influenza when the animal disease data mining model obtains the second mining result, thus providing a more accurate and targeted decision-making basis for the prevention and control of swine diseases.

[0126] In one possible implementation, the method further includes:

[0127] Step D110: Obtain the target mining results generated by the animal disease data mining model through risk diagnosis of the target animal disease case data. The target animal disease case data includes a target disease case description data and a target epidemic condition description data. The target epidemic condition description data is used to record the trend of disease change in the target disease case description data, indicating whether there are disease risk factors.

[0128] Step D120: Obtain the cross-domain attention knowledge vector of the disease knowledge feature chain of the target animal disease case data by the animal disease data mining model, load the cross-domain attention knowledge vector into the target risk diagnosis model, the target risk diagnosis model is generated by connecting the optimized first risk diagnosis network and the second risk diagnosis network, and use the first risk diagnosis network to generate the first risk diagnosis result.

[0129] Step D130: When the first risk diagnosis result indicates that the animal disease data mining model has a risk diffusion trend during data mining, the cross-domain attention knowledge vector is loaded into the second risk diagnosis network to generate the second risk diagnosis result.

[0130] Step D140: When the second risk diagnosis result indicates that the animal disease data mining model has a risk diffusion trend when performing data mining, it is determined that the animal disease data mining model has a risk diffusion trend when performing data mining on the target animal disease case data.

[0131] And, in step D150, when the first risk diagnosis result indicates that the animal disease data mining model does not have a risk diffusion trend when performing data mining, or when the second risk diagnosis result indicates that the animal disease data mining model does not have a risk diffusion trend when performing data mining, it is determined that the animal disease data mining model does not have a risk diffusion trend when performing data mining on the target animal disease case data.

[0132] In this embodiment, taking swine diseases as an example, the target disease case description data in the target animal disease case data may include information such as the breed of pig being Landrace, the herd size being 300 heads, a recent change in feed sources, and the ambient temperature being 18-22 degrees Celsius. The target disease condition description data records the trend of disease change in the target disease case description data, indicating whether there are disease risk factors. For example, the infection rate of classical swine fever virus may have increased from 3% to 5% in the past week, and the proportion of sick pigs with abnormal body temperature may have increased from 10% to 15%. The animal disease data mining model performs risk diagnosis based on this data and generates target mining results. It is assumed that the target mining results indicate that the pig herd is at risk of classical swine fever infection, the number of infected pigs may continue to increase, and the spread may be expanding.

[0133] Next, the server obtains the cross-domain attention knowledge vector of the disease knowledge feature chain of the target animal disease case data from the animal disease data mining model. This cross-domain attention knowledge vector integrates various correlation information in the disease knowledge feature chain and the disease change trend information in the disease trend knowledge graph. For example, the correlation between factors such as pig breed, age, weight, and stocking density in the disease knowledge feature chain, combined with information such as the rising trend of swine fever virus infection rate and the increase in the proportion of infected pigs with abnormal body temperature in the disease trend knowledge graph, generates a cross-domain attention knowledge vector through a specific algorithm. Then, the server loads this cross-domain attention knowledge vector into the target risk diagnosis model, which is generated by connecting the optimized first risk diagnosis network and the second risk diagnosis network. The first risk diagnosis result is generated using the first risk diagnosis network. After receiving the cross-domain attention knowledge vector, the first risk diagnosis network processes it according to its own structure and algorithm. For example, the neurons in the first risk diagnosis network analyze each element in the cross-domain attention knowledge vector, such as the relationship between pig breed and swine fever susceptibility, the impact of pig herd size on disease transmission, and the rate of increase in swine fever virus infection rate. Based on these analyses, the first risk diagnosis network generates a first risk diagnosis result. It is assumed that the first risk diagnosis result characterizes a risk diffusion trend in the animal disease data mining model during data mining. For example, the result may show that classical swine fever has a high probability of spreading from the current pig herd to other pig herds due to factors such as the rising infection rate of the classical swine fever virus and the possibility of contact between pig herds.

[0134] When the first risk diagnosis result indicates a risk diffusion trend in the animal disease data mining model during data mining, the server loads cross-domain knowledge vectors of interest into the second risk diagnosis network to generate a second risk diagnosis result. The second risk diagnosis network is a network that performs a deeper and more detailed analysis of the risk diffusion trend. Upon receiving the cross-domain knowledge vectors of interest, it further mines the information within them. For example, the second risk diagnosis network will analyze the transmission mode of the swine fever virus in more detail, whether the risk diffusion is caused by direct contact or indirect contact (such as through contaminated feed, water sources, etc.), and the role of pigs of different ages and weights in this transmission process. Suppose the second risk diagnosis result indicates a risk diffusion trend in the animal disease data mining model during data mining, such as determining that swine fever is mainly transmitted through direct contact between pig herds, and that due to the large size and high density of pig herds, this transmission method will cause swine fever to spread rapidly to surrounding pig herds in a short period of time.

[0135] In this scenario, the server determined that the animal disease data mining model exhibited a risk diffusion trend when mining data on target animal disease cases. This result indicates a risk of swine fever spreading within the pig herd, necessitating appropriate control measures such as isolating infected pigs, thoroughly disinfecting pigpens, and restricting the movement of pigs.

[0136] On the other hand, if the first risk diagnosis result indicates that the animal disease data mining model does not show a risk diffusion trend during data mining—for example, the result shows that although there is a classical swine fever virus infection, the spread of the virus is effectively controlled due to factors such as good hygiene conditions in the breeding environment and high immunity in the pig herd, preventing it from spreading to other pig herds—or, if the second risk diagnosis result indicates that the animal disease data mining model does not show a risk diffusion trend during data mining—for example, the second risk diagnosis network analysis finds that although there is a classical swine fever virus infection, the transmission route is effectively blocked, and contact between pig herds is strictly controlled, preventing spread—then in both cases, the server determines that the animal disease data mining model does not show a risk diffusion trend when mining the target animal disease case data. This result means that the current disease situation in the pig herd is under control and large-scale prevention and control measures are not required, but continuous monitoring of the disease's development is still necessary to prevent changes in the situation.

[0137] Therefore, the target risk diagnosis model can be used to further judge the risk diffusion trend of animal disease case data based on the data mining results of animal disease data mining model, providing an accurate basis for animal disease prevention and control decisions, and helping to control the spread and development of animal diseases in a timely and effective manner.

[0138] Figure 2The following is a schematic diagram of the hardware structure of an animal disease remote diagnosis system 100 provided in this application embodiment for implementing the above-described data mining method applied to an animal disease remote diagnosis center platform. Figure 2 As shown, the animal disease remote diagnosis system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0139] In one possible design, the animal disease remote diagnosis system 100 can be a single server or a group of servers. The server group can be centralized or distributed (e.g., the animal disease remote diagnosis system 100 can be a distributed system). In some embodiments, the animal disease remote diagnosis system 100 can be local or remote. For example, the animal disease remote diagnosis system 100 can access information and / or data stored in machine-readable storage medium 120 via a network. As another example, the animal disease remote diagnosis system 100 can directly connect to machine-readable storage medium 120 to access stored information and / or data. In some embodiments, the animal disease remote diagnosis system 100 can be implemented on an animal disease remote diagnosis system platform. By way of example only, the animal disease remote diagnosis system can include a private cloud, semantically relevant cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-layered cloud, etc., or any aggregation thereof.

[0140] Machine-readable storage medium 120 may store data and / or instructions. In some embodiments, machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, machine-readable storage medium 120 may store data and / or instructions used by the animal disease remote diagnostic system 100 to perform or use in order to accomplish the exemplary methods described in this application.

[0141] In the specific implementation process, one or more processors 110 execute computer-executable instructions stored in machine-readable storage medium 120, so that processor 110 can execute the data mining method applied to the animal disease remote diagnosis center platform as described in the above method embodiment. The processor 110, machine-readable storage medium 120 and communication unit 140 are connected through bus 130. The processor 110 can be used to control the sending and receiving actions of communication unit 140.

[0142] The specific implementation process of processor 110 can be found in the various method embodiments executed by the above-mentioned animal disease remote diagnosis system 100. The implementation principle and technical effect are similar, and will not be repeated here.

[0143] Furthermore, this application embodiment also provides a readable storage medium containing computer-executable instructions. When the processor executes the computer-executable instructions, the data mining method applied to the animal disease remote diagnosis center platform described above is implemented.

[0144] It should be noted that, in order to simplify the description disclosed in this application and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of the embodiments of this application sometimes combines multiple features into a single embodiment, drawing, or description thereof. Similarly, it should be noted that, in order to simplify the description disclosed in this application and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of the embodiments of this application sometimes combines multiple features into a single embodiment, drawing, or description thereof.

Claims

1. A data mining method applied to a remote animal disease diagnosis center platform, characterized in that, The method includes: When the animal disease data mining model performs data mining on animal disease case data, the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data is obtained. Based on the mining results of the animal disease data mining model on the animal disease case data, the cross-domain attention knowledge vector is decomposed into sample attention knowledge vectors of risk diffusion trend or sample attention knowledge vectors of non-risk diffusion trend. The sample attention knowledge vector is loaded into the first risk diagnosis network to generate a first risk diagnosis result of the first risk diagnosis network on the sample attention knowledge vector. Based on the first risk diagnosis result and the attention tag attribute of the sample attention knowledge vector, the first risk diagnosis network is optimized. From the sample attention knowledge vectors loaded into the first risk diagnosis network, extract the sample attention knowledge vectors with the attention tag attribute of risk diffusion trend, and continue to load them into the second risk diagnosis network to generate the second risk diagnosis result of the second risk diagnosis network on the sample attention knowledge vectors. Based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors, optimize the second risk diagnosis network. The optimized first risk diagnosis network is connected with the second risk diagnosis network to generate a target risk diagnosis model. The target risk diagnosis model is used to diagnose whether there is a risk diffusion trend in the animal disease data mining model during data mining. The mining results of the animal disease case data based on the animal disease data mining model decompose the cross-domain attention knowledge vector into sample attention knowledge vectors with risk diffusion trends or sample attention knowledge vectors without risk diffusion trends, including: Obtain the mining results of the animal disease case data by the animal disease data mining model, wherein the mining result is one of the first mining result and the second mining result; Based on the categories of the mining results and the tag data of the animal disease case data, the cross-domain attention knowledge vector is decomposed into sample attention knowledge vectors of the first type of risk diffusion trend, sample attention knowledge vectors of the second type of risk diffusion trend, sample attention knowledge vectors of the first type of non-risk diffusion trend, or sample attention knowledge vectors of the second type of non-risk diffusion trend. The step involves extracting sample attention knowledge vectors with the attention tag attribute of risk diffusion trend from the sample attention knowledge vectors loaded into the first risk diagnosis network, and then loading them into the second risk diagnosis network to generate a second risk diagnosis result for the sample attention knowledge vectors. Based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors, the second risk diagnosis network is optimized, including: From the sample attention knowledge vectors loaded into the first risk diagnosis network, extract the sample attention knowledge vectors with the attention tag attribute of the first type of risk diffusion trend, and continue to load them into the second risk diagnosis network of the first network label. Generate the second risk diagnosis result of the second risk diagnosis network of the first network label on the sample attention knowledge vectors. Based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vectors, optimize the second risk diagnosis network of the first network label. From the sample attention knowledge vector loaded into the first risk diagnosis network, extract the sample attention knowledge vector with the attention tag attribute of the second type of risk diffusion trend, and continue to load it into the second risk diagnosis network of the second network label. Generate the second risk diagnosis result of the second risk diagnosis network of the second network label on the sample attention knowledge vector. Based on the second risk diagnosis result and the sample training annotation data of the sample attention knowledge vector, optimize the second risk diagnosis network of the second network label. The step of connecting the optimized first risk diagnosis network with the second risk diagnosis network to generate the target risk diagnosis model includes: The optimized first risk diagnosis network is connected to the second risk diagnosis network with the first network label to generate a target risk diagnosis model. This target risk diagnosis model is used to diagnose whether there is a risk diffusion trend when the animal disease data mining model obtains the first mining result; and The optimized first risk diagnosis network is connected to the second risk diagnosis network with the second network label to generate a target risk diagnosis model. The target risk diagnosis model is used to diagnose whether there is a risk diffusion trend when the animal disease data mining model obtains the second mining result.

2. The data mining method applied to the remote diagnostic center platform for animal diseases according to claim 1, characterized in that, The cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data when the animal disease data mining model performs data mining on the animal disease case data includes: Obtain animal disease case data, which includes a disease case description data and an epidemic condition description data. The epidemic condition description data is used to record whether there is a disease risk factor in the disease case description data and the trend of disease condition changes. The animal disease data mining model is used to obtain a disease knowledge feature chain based on the disease case description data and a disease trend knowledge graph based on the disease condition description data. Based on the disease knowledge feature chain and the disease trend knowledge graph, a cross-domain attention knowledge vector of the disease knowledge feature chain is generated.

3. The data mining method applied to the remote diagnostic center platform for animal diseases according to claim 2, characterized in that, The animal disease data mining model includes a disease case feature extraction unit, a cross-domain association unit, and a knowledge reasoning unit. The knowledge reasoning unit includes an epidemic disease feature extraction subunit and an epidemic disease feature restoration subunit. The process involves using the animal disease data mining model to obtain a disease knowledge feature chain based on the disease case description data, obtaining a disease trend knowledge graph based on the epidemic condition description data, and generating a cross-domain attention knowledge vector for the disease knowledge feature chain based on the disease knowledge feature chain and the disease trend knowledge graph, including: Using the disease case feature extraction unit in the animal disease data mining model, feature extraction is performed on the disease case description data to generate disease case feature vectors. Using the cross-domain association unit in the animal disease data mining model, the disease case feature vectors are cross-domain knowledge-associated to generate disease knowledge feature chains of the animal disease case data. Using the disease condition feature extraction subunit of the animal disease data mining model, the disease condition description data is represented by graph features to generate a disease condition trend knowledge graph of the animal disease case data. Using the disease condition feature restoration subunit of the animal disease data mining model, feature restoration is performed based on the cross-domain feature path data generated by fusing the disease knowledge feature chain with the disease trend knowledge graph, generating the mining results of the animal disease data mining model on the animal disease case data, and obtaining the cross-domain attention knowledge vector of the disease knowledge feature chain of the animal disease case data by the animal disease data mining model when generating the mining results; The step of using the cross-domain association unit of the animal disease data mining model to perform cross-domain knowledge association on the disease case feature vectors and generate a disease knowledge feature chain for the animal disease case data includes: Using cross-domain association units, based on pre-defined disease-related knowledge domain classification standards, each element in the disease case feature vector is divided into different knowledge subsets, which are divided based on the type of disease, symptom category, and source of infection. Based on the structure of the disease knowledge system and the prior requirements for animal disease diagnosis, a set of cross-domain knowledge association rules is set for the cross-domain association unit. The setting of the cross-domain knowledge association rules revolves around the potential connections between different knowledge subsets. For each knowledge subset, the elements in the knowledge subset are traversed, and the cross-domain knowledge association rule set is used to determine whether there is an association relationship between the element and the elements in other knowledge subsets. If there is, the element is marked with the name or identifier of the other knowledge subsets associated with it, so that the elements in each knowledge subset clearly know the association relationship with other knowledge subsets, and a knowledge subset with association mark is generated. Based on knowledge subsets carrying association tags, a preliminary disease knowledge feature chain framework is constructed. The construction process of this preliminary disease knowledge feature chain framework starts from the core knowledge subset and connects related knowledge subsets step by step along the association tags of the elements. Specifically, the connection method is to arrange the knowledge subsets with relationships in a logical order to form a chain structure framework. In the chain structure framework, the position of each knowledge subset is determined according to the relationship between the knowledge subset and other knowledge subsets and the importance weight in disease diagnosis. For each knowledge subset in the initial disease knowledge feature chain framework, the elements in the knowledge subset are arranged according to the advanced and refined association relationship based on the association tags to generate a refined disease knowledge feature chain framework. The detailed disease knowledge feature chain framework undergoes integrity checks and optimization. From the perspective of the disease knowledge system, it is checked whether the detailed disease knowledge feature chain framework covers all key knowledge areas of the disease. From the perspective of disease diagnosis, it is checked whether the detailed disease knowledge feature chain framework can be used for effective disease diagnosis. It is also checked whether the detailed disease knowledge feature chain framework has missing key information or unreasonable relationships. When the detailed disease knowledge feature chain framework is determined to have integrity issues, it is optimized. Optimization methods include supplementing missing knowledge elements, adjusting unreasonable relationships, and rearranging knowledge subsets or the order of elements. This generates the disease knowledge feature chain, which reflects the disease knowledge features in the animal disease case data and the relationships between these disease knowledge features.

4. The data mining method applied to the remote diagnostic center platform for animal diseases according to claim 1, characterized in that, The method further includes: Obtain a sequence of sample disease case description data. For each sample disease case description data in the sample disease case description data sequence, obtain the prior disease change trend of the sample disease case description data. The prior disease change trend refers to the disease change trend of disease risk factors existing in the sample disease case description data. Obtain an epidemic condition description template, which is used to record the trend of disease condition changes in epidemic case description data, including whether there are epidemic risk factors. Based on the prior disease change trend of the sample epidemic case description data and the epidemic disease description template, sample epidemic disease description data is generated. One prior disease change trend generates one sample epidemic disease description data. The sample epidemic disease description data is used to record whether the prior disease change trend exists in the sample epidemic case description data. Based on the sample disease case description data and the sample epidemic condition description data, animal disease case data are configured as positive samples.

5. The data mining method applied to the remote diagnostic center platform for animal diseases according to claim 4, characterized in that, The method further includes: Based on the prior disease change trend of the sample disease case description data, the negative sample disease condition of the sample disease case description data is determined. The negative sample disease condition is the disease change trend of disease risk factors that do not exist in the sample disease case description data. Based on the negative sample disease condition of the sample epidemic case description data and the epidemic disease description template, sample epidemic disease description data is generated. One negative sample disease condition generates one sample epidemic disease description data. The sample epidemic disease description data is used to record whether the negative sample disease condition exists in the sample epidemic case description data. Based on the sample disease case description data and the sample epidemic condition description data, animal disease case data are configured as negative samples.

6. The data mining method applied to the remote diagnostic center platform for animal diseases according to claim 5, characterized in that, The determination of the negative sample disease condition based on the prior disease change trend of the sample disease case description data includes: Based on the prior disease change trend of the sample disease case description data, multiple negative sample disease conditions are determined from the sample disease case description data. A negative sample disease condition sequence is generated from the sample disease case description data. A single negative sample disease condition is then randomly selected from the negative sample disease condition sequence and used as the negative sample disease condition of the sample disease case description data; or... Obtain the prior disease change trend of other sample disease case description data besides the sample disease case description data in the sample disease case description data sequence; extract the negative sample disease from the prior disease change trend of the other sample disease case description data; generate a negative sample disease sequence of the sample disease case description data; randomly select from the negative sample disease sequence to generate a negative sample disease as the negative sample disease of the sample disease case description data; or... A shared disease change trend is determined from the prior disease change trend of the sample disease case description data. The shared disease change trend is the disease change trend of disease risk factors that do not exist in the sample disease case description data but have a correlation feature with the prior disease change trend. Based on the shared disease change trend, a negative sample disease sequence is generated from the sample disease case description data. A negative sample disease is generated by randomly selecting from the negative sample disease sequence, which serves as the negative sample disease of the sample disease case description data.

7. The data mining method applied to the remote diagnostic center platform for animal diseases according to claim 6, characterized in that, The method further includes: Obtain the prior disease change trend of each sample disease case description data in the sample disease case description data sequence; Based on the sample disease case description data sequence, the correlation between any two prior disease change trends is calculated. Two different prior disease change trends appearing simultaneously in the same sample disease case description data are considered as one correlation. When the correlation between two different prior disease change trends meets the target requirements, it is determined that the two different prior disease change trends have a correlation feature relationship.

8. The data mining method applied to an animal disease remote diagnostic center platform according to any one of claims 1-7, characterized in that, The method further includes: The target mining results generated by the animal disease data mining model through risk diagnosis of target animal disease case data are obtained. The target animal disease case data includes a target disease case description data and a target epidemic condition description data. The target epidemic condition description data is used to record whether there are disease risk factors in the target disease case description data and the trend of disease condition changes. Obtain the cross-domain attention knowledge vector of the disease knowledge feature chain of the target animal disease case data by the animal disease data mining model, load the cross-domain attention knowledge vector into the target risk diagnosis model, the target risk diagnosis model is generated by connecting the optimized first risk diagnosis network and the second risk diagnosis network, and use the first risk diagnosis network to generate the first risk diagnosis result; When the first risk diagnosis result indicates that the animal disease data mining model has a risk diffusion trend during data mining, the cross-domain attention knowledge vector is loaded into the second risk diagnosis network to generate the second risk diagnosis result. When the second risk diagnosis result indicates that the animal disease data mining model has a risk diffusion trend when performing data mining, it is determined that the animal disease data mining model has a risk diffusion trend when performing data mining on the target animal disease case data. And, when the first risk diagnosis result indicates that the animal disease data mining model does not have a risk diffusion trend when performing data mining, or when the second risk diagnosis result indicates that the animal disease data mining model does not have a risk diffusion trend when performing data mining, it is determined that the animal disease data mining model does not have a risk diffusion trend when performing data mining on the target animal disease case data.

9. A remote diagnostic system for animal diseases, characterized in that, The animal disease remote diagnosis system includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the data mining method applied to the animal disease remote diagnosis center platform as described in any one of claims 1-7.

Images