Method for enhancing epidemic prevention management of infectious diseases through network modeling

Abstract

The invention provides a method for enhancing epidemic prevention management of infectious diseases through network modeling. Comprising the following four processes: firstly, according to detailed epidemic situation contact information recorded by a flow scheduling report, constructing an epidemic situation association diagram; then, by defining a meta path, generating a node sequence in the heterogeneous graph by using random walk, and capturing semantic and structural correlation among various nodes in the heterogeneous graph; then, risk factors are introduced and adjusted, namely, the weight coefficient in the random walk process is adjusted according to the risk factor level provided by the disease prevention and control center, and the risk factors comprise the site ventilation condition, the contact duration, the personnel density and the behavior habits; and finally, carrying out point embedding and risk assessment, assessing the importance of nodes, and predicting high-risk individuals and places. Compared with a traditional network evaluation method, the scheme of the invention improves the accuracy of risk prediction.

Description

Technical Field

[0001] The present disclosure relates to the field of computers, and more specifically, to a method for enhancing epidemic prevention and management of infectious diseases through network modeling. Background Art

[0002] During an epidemic outbreak, accurate and timely tracking and predicting the transmission path of infectious disease prevention and control is the key to controlling the spread of the virus. However, traditional infectious disease prevention and control management methods have significant limitations, which are mainly reflected in the following aspects. First, traditional contact tracing methods usually rely on manual recording and analysis of direct contact relationships. Although these methods have played a certain role in early infectious disease prevention and control, they are powerless when faced with complex infectious disease transmission networks. When it comes to multiple indirect contacts or transmission in different locations, simple contact tracing is difficult to capture potential transmission chains and high-risk individuals. This method often ignores the impact of complex multidimensional interactions and environmental factors on virus transmission, resulting in delayed response and improper resource allocation. In the process of infectious disease transmission, environmental factors such as ventilation conditions, population density, and contact time have an important impact on the risk of virus transmission. For example, in a closed indoor environment, the risk of virus transmission is significantly higher than in a well-ventilated outdoor environment. In addition, individual behavior (such as whether to wear a mask or maintain social distance) also directly affects the probability of virus transmission. Traditional infectious disease prevention and control management methods are difficult to systematically integrate and analyze these environmental factors, resulting in large deviations in risk assessment and prediction. Modern epidemiological research and infectious disease prevention and control management require more complex network analysis methods to comprehensively capture and simulate the dynamic process of infectious disease transmission. Summary of the invention

[0003] The purpose of the embodiments of the present disclosure is to provide a method for enhancing infectious disease prevention and control management through network modeling. In the early stages of an epidemic, it is crucial to promptly identify individuals who are more susceptible to infection. Therefore, in order to identify individuals who are more susceptible to infection in the early stages of an infectious disease, a new meta-path-based embedding method is proposed. By constructing a heterogeneous graph containing detailed epidemic contact element information, nodes that should be controlled earlier are predicted.

[0004] In a general aspect, a method for enhancing infectious disease prevention and control management through network modeling is provided, which includes four processes:

[0005] First, according to the detailed epidemic contact information recorded in the epidemiological investigation report, an epidemic association graph is constructed. The epidemic association graph is a heterogeneous graph that includes infected persons, susceptible persons, and locations as nodes, and includes contacts between infected persons, contacts between infected persons and susceptible persons, contacts between susceptible persons, and relationships between infected persons and locations as edges;

[0006] Then, by defining meta-paths, random walks are used to generate node sequences in heterogeneous graphs to capture the semantic and structural correlations between various nodes in the heterogeneous graphs.

[0007] Then, risk factors are introduced and adjusted, that is, the weight coefficients in the random walk process are adjusted according to the risk factor levels provided by the Centers for Disease Control and Prevention. The risk factors include ventilation conditions of the location, contact time, density of people, and behavioral habits;

[0008] Finally, point embedding and risk assessment are performed. The generated meta-path sequence is trained using the word2vec model to obtain the low-dimensional embedding vectors of the nodes. The risk factors are combined to perform risk assessment on the nodes and locations. The importance of the nodes is evaluated by calculating the cosine similarity between the low-dimensional embedding vectors of the nodes, and high-risk individuals and locations are predicted.

[0009] The node needs to be uniquely coded.

[0010] The specific method of obtaining the low-dimensional embedding vector of the node is: given a network G = (V, E), the goal is to maximize the network probability based on the local structure, that is:

[0011]

[0012] where N(v) is the neighborhood of node v in the network G, and p(c|v;θ) defines the conditional probability of having a context node c given node v;

[0013] In heterogeneous graphs, the node embedding method is used to maximize the heterogeneous context N given a node v. t (v), where t∈T represents the probability of the node type, enabling skip-gram to learn effective node representations in heterogeneous networks:

[0014]

[0015] Among them, N t (v) represents the neighborhood of the t-th type of node v, p(c t |v;θ) is usually defined as the softmax function, that is:

[0016]

[0017] Among them, x v is the embedding vector of node v;

[0018] Random walks are applied to heterogeneous networks to generate paths of multiple types of nodes, and meta-path-based random walks are designed to generate paths that can capture the semantic and structural correlations between different types of nodes, thereby facilitating the conversion of heterogeneous network structures into skip-gram of metapath2vec.

[0019] Random walk of the metapath Defined as:

[0020]

[0021] in For node types V1 and V L The composite relationship between them, according to the defined meta-path, guides the random walk as follows:

[0022]

[0023] in, and Representation Node V t+1 When the neighbor type does not match the currently defined meta-path, the wandering probability is 0, and γ is the correction coefficient affected by the epidemic risk factor.

[0024] The specific process of performing point embedding and risk assessment is as follows: using the word2vec model to train the meta-path sequence to obtain the embedding vector of the node;

[0025] Then calculate the cosine similarity between nodes:

[0026]

[0027] Where x and y are the embedding vectors of the two nodes;

[0028] By identifying the most similar nodes for each node and analyzing their importance in the network, high-risk individuals can be screened;

[0029] Combining node embedding and cosine similarity to evaluate and predict high-risk locations, a risk location dataset is constructed to count the number of connections between each location and infected individuals, and the ranking is performed;

[0030] Specifically:

[0031] First, the importance of infected persons is ranked according to the risk population results predicted by the model, and the top k infected persons are selected. For each infected person, a weight is assigned according to its ranking. The higher the ranking, the higher the weight, which means the greater the risk;

[0032] Second, traverse each infected person, accumulate the weights of the location nodes connected to it, obtain the score of each location, sort the locations according to the score, and obtain the high-risk locations with the highest evaluation value;

[0033] Finally, the list of risk locations given by the model is compared with known risk locations, and the hit rate is calculated to evaluate the model performance.

[0034] The innovation of the embodiment of the present invention lies in:

[0035] The present invention introduces detailed contact element information to construct a heterogeneous graph to more accurately analyze and predict the spread of infectious diseases, and records the basic information of infected persons, the time of symptom onset or positive test time, the specific date and time of the contact event, the location, the length of stay, the number of contacts, etc. In addition, factors such as whether the behavior of the infected person is conducive to the spread of the virus, whether he wears a mask, whether he uses public supplies and facilities, the type of venue, the ventilation conditions of the venue, and whether the contact person wears a mask in a standardized manner are also considered, thereby proposing a risk population identification method based on network modeling, constructing a heterogeneous graph containing rich semantics and diverse nodes, obtaining risk populations while also reflecting the impact of different venues on the spread of the epidemic, thereby improving the representation of rich nodes.

[0036] This approach has the following advantages:

[0037] The use of meta-path-based random walk and node embedding technology enables the evaluation model to capture complex network structures and subtle relationships between nodes, improving the accuracy of risk prediction compared to traditional network evaluation methods.

[0038] The heterogeneous graph model reflects the complexity of the spread of the epidemic more comprehensively by introducing multiple types of nodes and edges. Compared with the traditional single contact tracing method, the method of the present invention can capture multiple relationships (such as direct contact, indirect contact) between different types of nodes (such as infected persons, susceptible persons, and locations), thereby providing a more comprehensive risk assessment result.

[0039] By identifying high-risk individuals and locations, the method of the present invention can return a risk ranking, so that in actual application, resource allocation can be optimized, high-risk groups and key locations can be protected first, thereby improving the efficiency of epidemic prevention and control. Especially in the case of limited medical resources, the application of this method can significantly improve the prevention and control effect and reduce unnecessary waste of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The above and other objects and features of the present disclosure will become more apparent from the following description in conjunction with the accompanying drawings.

[0041] Figure 1 2 is a schematic diagram showing a design framework of an epidemic risk management method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0042] The following specific embodiments are provided to help the reader obtain a comprehensive understanding of the methods, devices and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the methods, devices and / or systems described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0043] The features described herein can be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein, which will be clear after understanding the disclosure of the present application.

[0044] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more.

[0045] Although terms such as "first", "second", and "third" may be used herein to describe various members, components, regions, layers, or portions, these members, components, regions, layers, or portions should not be limited by these terms. Instead, these terms are only used to distinguish one member, component, region, layer, or portion from another member, component, region, layer, or portion. Therefore, without departing from the teachings of the examples described herein, the first member, first component, first region, first layer, or first portion referred to in the examples may also be referred to as the second member, second component, second region, second layer, or second portion.

[0046] In the specification, when an element (such as a layer, a region or a substrate) is described as being “on”, “connected to” or “coupled to” another element, the element may be directly “on”, “connected to” or “coupled to” another element, or one or more other elements may be present therebetween. Conversely, when an element is described as being “directly on”, “directly connected to” or “directly coupled to” another element, there may be no other elements present therebetween.

[0047] The terms used herein are only used to describe various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. The terms "comprise", "include" and "have" indicate the presence of the described features, quantities, operations, components, elements and / or combinations thereof, but do not exclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0048] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by a person of ordinary skill in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0049] Furthermore, in the description of examples, when it is considered that a detailed description of a well-known related structure or function would cause vague interpretation of the present disclosure, such a detailed description will be omitted.

[0050] Figure 1 It is a schematic diagram showing a method for enhancing infectious disease prevention and control management through network modeling according to an embodiment of the present disclosure.

[0051] The invention is based on heterogeneous graphs and random walks. It proposes to use detailed epidemic contact information to construct a heterogeneous graph, and generate a meta-path sequence through random walks. Finally, through node embedding and risk factor analysis, the prediction of high-risk individuals and locations is realized, achieving the effect of epidemic transmission risk assessment and realizing the design purpose and application goals.

[0052] The algorithm design process of the present invention is as follows Figure 1 As shown in the figure: First, detailed epidemic contact information is recorded according to the epidemiological report, and an epidemic association graph is constructed. Then, by defining the meta-path, a random walk is used to generate a node sequence in the heterogeneous graph. Next, the generated meta-path sequence is trained using the word2vec model to obtain the low-dimensional embedding vector of the node. Finally, the risk factors are combined to evaluate the risk of nodes and locations, and high-risk individuals and locations are predicted by calculating the cosine similarity between the node embedding vectors. Through this series of processes, the risk assessment of epidemic transmission can be carried out comprehensively and accurately.

[0053] The method includes the following four parts and processes:

[0054] Epidemic association graph construction: Record detailed epidemic contact information and construct a heterogeneous graph containing multiple nodes (infected persons, susceptible persons, locations) and multiple edges (contact relationships). Node types include infected persons, susceptible persons, and locations, and edge types include contacts between infected persons, contacts between infected persons and susceptible persons, contacts between susceptible persons, and relationships between infected persons and locations.

[0055] Random walk to construct meta-path sequence: Based on the defined meta-path, a random walk method is used to generate node sequences in the heterogeneous graph. Meta-path captures the semantic and structural correlations between various nodes in the heterogeneous graph by defining the relationships between different node types.

[0056] Introduction and adjustment of risk factors: According to the risk factor level provided by the Centers for Disease Control and Prevention (CDC), the weight coefficient in the random walk process is adjusted. Risk factors include ventilation conditions, contact time, personnel density, behavioral habits, etc., to more accurately reflect the actual risk of epidemic transmission.

[0057] Node embedding and risk assessment: The generated meta-path sequences are trained using the word2vec model to obtain low-dimensional embedding vectors of the nodes. The importance of the nodes is assessed by calculating the cosine similarity between the node embedding vectors to predict high-risk individuals and locations.

[0058] In addition, in order to train and deploy the model, you need to build a running environment including pytorch and other necessary dependent libraries on the machine.

[0059] Epidemic association graph construction:

[0060] During the graph construction process, we assign specific attributes and labels to each node and edge to accurately reflect their roles and relationships in the spread of the epidemic. The specific steps are as follows:

[0061] Node type: Node types include Susceptible, Person, Location, etc. Each node type has unique attributes, such as the health status of the infected person, the wearing of masks, the social relationships of ordinary people, the type of location, ventilation conditions, etc.

[0062] Edge type: Edge types include contact between infected persons (Susceptible-Susceptible), contact between infected persons and susceptible persons (Susceptible-Person), contact between susceptible persons and susceptible persons (Person-Person), relationship between infected persons and locations (Susceptible-location), relationship between infected persons and locations (Person-location), etc. Edge attributes include contact time, contact location, contact duration, and various contact elements mentioned above.

[0063] In order to facilitate subsequent feature extraction and prevent sensitive information leakage, the nodes are uniquely encoded. After processing the complex connection relationships in the heterogeneous graph, the attribute information of each node and edge is embedded into the graph structure.

[0064] Random walk constructs the meta-path sequence associated with the epidemic:

[0065] The framework of the word2vec model is used to complete node embedding. In word2vec, given a text corpus, the words and context concepts in the text corpus are mapped to the network. We use random walks to achieve this goal of node corpus, and use the skip-gram model to learn the representation of nodes, which can facilitate the prediction of their structural context, that is, local neighborhood, in homogeneous networks. In general, given a network G = (V, E), the goal is to maximize the network probability based on the local structure, that is:

[0066]

[0067] where N(v) is the neighborhood of node v in the network G, which can be defined in different ways, such as v’s direct neighbors; and p(c|v;θ) defines the conditional probability of having a context node c given node v.

[0068] In heterogeneous graphs, the node embedding method is used to maximize the heterogeneous context N given a node v. t (v), where t∈T represents the probability of a node type, enabling skip-gram to learn effective node representations in heterogeneous networks.

[0069]

[0070] Among them, N t (v) represents the neighborhood of the t-th type of node v, p(c t |v;θ) is usually defined as the softmax function, that is:

[0071]

[0072] Among them, x v is the embedding vector of node v.

[0073] We apply random walks to heterogeneous networks to generate paths of multiple types of nodes. We design a meta-path based random walk to generate paths that can capture the semantic and structural correlations between different types of nodes, thereby facilitating the transformation of heterogeneous network structures into skip-grams of metapath2vec. Formally, the meta-path scheme Defined as a path of the form:

[0074]

[0075] in Defines node types V1 and V L For example, in the epidemic risk association graph, the meta-path "P1L1P2" indicates that a susceptible person (P1) and a susceptible person (P2) have arrived at a location (L1) together, while "S1L1P1L2P2" indicates that an infected person (S1) contacts a susceptible person (P1) through location (L1), and the susceptible person (P1) has a contact relationship with another susceptible person (P2) through location (L2). It should be noted that there are only two types of nodes P and L in the meta-path definition, and the infected person S is obtained through the P node and the infected person set.

[0076] The way to guide the random walk according to the defined meta-path is shown in the formula

[0077]

[0078] in, and Representation Node V t+1 When the neighbor type does not match the currently defined meta-path, the wandering probability is 0. γ is the correction coefficient affected by the epidemic risk factor, that is, various contact factors, and the coefficient is determined according to the risk factor level provided by the CDC (Centers for Disease Control and Prevention).

[0079] Point embedding and risk assessment:

[0080] After generating a series of meta-path sequences, the word2vec model is used to train these sequences to obtain the embedding vectors of the nodes. By using the node sequence generated by random walk as the input of word2vec, a low-dimensional vector representation of each node can be learned. This representation not only retains the structural information of the node, but also captures the semantic relationship between nodes. After obtaining the node embedding vector, the cosine similarity between nodes can be calculated to evaluate the importance of the node and predict high-risk populations and locations. Specifically, we measure the similarity between nodes by calculating the cosine similarity between node vectors. Cosine similarity is defined as follows:

[0081]

[0082] Where (x) and (y) are the embedding vectors of the two nodes. By calculating the cosine similarity matrix, the most similar nodes to each node can be identified, so as to further analyze their importance in the network. Intuitively speaking, the susceptible person who is most similar to the infected person in the feature space is also more likely to be infected, which means that the risk of infection when these two people are in close contact or in contact is high, and they should be given priority control. The evaluation indicator is the hit rate, that is, how many of the predicted risk personnel will be confirmed to be infected in the future. For example, in the epidemic risk association graph constructed at time T, there are three types of nodes in the graph, infected person S, susceptible person P and location L, but in fact the susceptible person can be further subdivided into infected person S at time T+t. T+t and susceptible P T+t We hope that the model can predict the future infected person S as much as possible. T+t , so use the predicted risk groups R and S T+t The hit rate is calculated to evaluate the model performance. Specifically, the model embeds the vector x of the learned infected node S S , we can calculate all remaining susceptible node vectors x P With x s The cosine similarity of the infected nodes is then sorted according to the degree of similarity to obtain the most relevant susceptible nodes {P1, P2, ...}. Based on the sorting results, the susceptible nodes can be scored for the risk of the infected node S. The higher the score, the higher the score. Then the scores of the susceptible nodes for all infected nodes are summed up to obtain the final risk score of the susceptible nodes. The final predicted risk population R is obtained by sorting according to the score size, and then compared with the actual future infected population S. T+t Compare and calculate the hit rate. The formula is as follows: T+t The ratio of the number of people in to the total number of people in R:

[0083]

[0084] At the same time, combining node embedding and cosine similarity, high-risk locations can also be evaluated and predicted. We construct a risk location dataset by counting the number of connections between each location and infected individuals (including individuals who will be infected in the future) and ranking them. Specifically, the number of infected people arriving at each location scores the location risk, and the higher the score, the greater the risk.

[0085] The specific scoring process of the location risk value of the risk prediction algorithm is as follows:

[0086] 1. Sort the importance of infected persons according to the risk population results predicted by the model, select the first k infected persons, and assign weights to each infected person according to their ranking. The higher the ranking, the higher the weight, which means the greater the risk considered by the model.

[0087] 2. Traverse each infected person, accumulate the weights of the location nodes connected to it, and get the score of each location. Sort the locations according to the score to get the high-risk location with the highest evaluation value.

[0088] 3. Compare the list of risk locations given by the model with known risk locations and calculate the hit rate to evaluate model performance.

[0089] Specifically, similar to the calculation method of risk population, the model embeds the vector x according to the learned infected node S S , all remaining location node vectors x can be calculated L With x s The cosine similarity of the infected nodes is then sorted according to the degree of similarity to obtain the location nodes {L1, L2, ...} that are most relevant to the infected node. Based on the sorting results, the risk of the location node with respect to the infected person S can be scored, and the higher the score, the higher the score. Then the scores of the location for all infected people are summed up to obtain the final risk score of the location, and the final predicted risk location C is sorted according to the score size. It is then compared with the list of locations Q with the most infected people in the future and the hit rate is calculated. The formula is as follows, that is, the ratio of the number of people in C who appear correctly in Q to the total number of locations in C:

[0090]

[0091] Although some embodiments of the present disclosure have been shown and described, it will be appreciated by those skilled in the art that modifications may be made to the embodiments without departing from the principles and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for enhancing infectious disease prevention and control management through network modeling, characterized in that: It includes four processes: First, according to the detailed epidemic contact information recorded in the epidemiological investigation report, an epidemic association graph is constructed. The epidemic association graph is a heterogeneous graph that includes infected persons, susceptible persons, and locations as nodes, and includes contacts between infected persons, contacts between infected persons and susceptible persons, contacts between susceptible persons, and relationships between infected persons and locations as edges; Then, by defining meta-paths, random walks are used to generate node sequences in heterogeneous graphs to capture the semantic and structural correlations between various nodes in the heterogeneous graphs. Then, risk factors are introduced and adjusted, that is, the weight coefficients in the random walk process are adjusted according to the risk factor levels provided by the Centers for Disease Control and Prevention. The risk factors include ventilation conditions of the location, contact time, density of people, and behavioral habits; Finally, point embedding and risk assessment are performed. The generated meta-path sequence is trained using the word2vec model to obtain the low-dimensional embedding vectors of the nodes. The risk factors are combined to perform risk assessment on the nodes and locations. The importance of the nodes is evaluated by calculating the cosine similarity between the low-dimensional embedding vectors of the nodes, and high-risk individuals and locations are predicted.

2. A method for enhancing infectious disease prevention and control management through network modeling as claimed in claim 1, characterized in that: The node needs to be uniquely coded.

3. A method for enhancing infectious disease prevention and control management through network modeling as claimed in claim 1, characterized in that: The specific method of obtaining the low-dimensional embedding vector of the node is: given a network G = (V, E), the goal is to maximize the network probability based on the local structure, that is: where N(v) is the neighborhood of node v in the network G, and p(c|v;θ) defines the conditional probability of having a context node c given node v; In heterogeneous graphs, the node embedding method is used to maximize the heterogeneous context N given a node v. t (v), where t∈T represents the probability of the node type, enabling skip-gram to learn effective node representations in heterogeneous networks: Among them, N t (v) represents the neighborhood of the t-th type of node v, p(c t |v;θ) is usually defined as the softmax function, that is: Among them, x v is the embedding vector of node v; Random walks are applied to heterogeneous networks to generate paths of multiple types of nodes, and meta-path-based random walks are designed to generate paths that can capture the semantic and structural correlations between different types of nodes, thereby facilitating the conversion of heterogeneous network structures into skip-gram of metapath2vec.

4. A method for enhancing infectious disease prevention and control management through network modeling as claimed in claim 3, characterized in that: Random walk of the metapath Defined as: in For node types V1 and V L The composite relationship between them, according to the defined meta-path, guides the random walk as follows: in, and Representation Node V t+1 When the neighbor type does not match the currently defined meta-path, the wandering probability is 0, and γ is the correction coefficient affected by the epidemic risk factor.

5. A method for enhancing infectious disease prevention and control management through network modeling as claimed in claim 1, characterized in that: The specific process of performing point embedding and risk assessment is as follows: using the word2vec model to train the meta-path sequence to obtain the embedding vector of the node; Then calculate the cosine similarity between nodes: Where x and y are the embedding vectors of the two nodes; By identifying the most similar nodes for each node and analyzing their importance in the network, high-risk individuals can be screened; Combining node embedding and cosine similarity to evaluate and predict high-risk locations, a risk location dataset is constructed to count the number of connections between each location and infected individuals, and the ranking is performed; Specifically: First, the importance of infected persons is ranked according to the risk population results predicted by the model, and the top k infected persons are selected. For each infected person, a weight is assigned according to its ranking. The higher the ranking, the higher the weight, which means the greater the risk; Second, traverse each infected person, accumulate the weights of the location nodes connected to it, obtain the score of each location, sort the locations according to the score, and obtain the high-risk locations with the highest evaluation value; Finally, the list of risk locations given by the model is compared with known risk locations, and the hit rate is calculated to evaluate the model performance.

Images