A method for assessing epidemic transmission risk in cross-border e-commerce networks

Abstract

The present invention discloses a method for assessing the risk of epidemic spread in a cross-border e-commerce network, which relates to the field of anomaly detection technology. By acquiring e-commerce data and virus transmission data, an e-commerce logistics topology network and an epidemic spread path topology network are established. According to the relationship between nodes in the two networks, a multi-layer complex network is established, and cross-layer anomaly detection is performed on the multi-layer complex network to determine and lock abnormal logistics information and conduct epidemic spread risk assessment. The present invention provides a method for assessing the risk of epidemic spread in a cross-border e-commerce network, which is applied to a cross-border e-commerce network, and detects abnormal subgraphs in a weak abnormal feature network, further detects imported goods carrying viruses, and conducts a quantitative assessment of the risk.

Description

Technical Field

[0001] The present invention relates to the field of anomaly detection technology, and in particular to a method for assessing the risk of epidemic spread in a cross-border e-commerce network. Background Art

[0002] A weak anomaly feature network refers to a network in which data features related to the abnormal event (or structure) under study cannot be observed in the data set that constitutes the network. For example, in a cross-border e-commerce network, there is no abnormal feature data related to the spread of COVID-19. In addition, traditional anomaly detection methods are highly dependent on abnormal feature data. When applied to a cross-border e-commerce network that lacks epidemic feature data, it is impossible to complete the spread detection of the epidemic virus.

[0003] Regarding anomaly detection of COVID-19, one part uses medical technology to detect the virus, and the other part uses medical data to build deep learning algorithms for classification. For example, Ozturk, Wang, Taresh, and Toraman all use patient X-ray chest images to train various deep neural network classification algorithms for detecting COVID-19. They all study anomaly detection problems on data sets with strong abnormal characteristics of COVID-19.

[0004] On the other hand, anomaly detection algorithms related to e-commerce are mainly aimed at detecting fraudulent behavior in the e-commerce process, such as e-commerce fraud anomaly detection algorithms for user behavior on e-commerce platforms. There is currently no anomaly detection algorithm for epidemic detection in e-commerce data. However, it has been found that there are many batches of imported goods that have passed through epidemic outbreak areas during transportation and carried the virus, so it is necessary to study methods for epidemic detection using cross-border e-commerce data.

[0005] To address this problem, the present invention proposes a method for assessing the risk of epidemic spread in cross-border e-commerce networks. The method is applied to cross-border e-commerce networks. By detecting abnormal subgraphs in weak abnormal feature networks, imported goods carrying the COVID-19 virus are further detected, and the risk is quantitatively assessed. Summary of the invention

[0006] The purpose of the present invention is to provide a method for assessing the risk of epidemic spread in cross-border e-commerce networks, which is applied to cross-border e-commerce networks. By detecting abnormal subgraphs in weak abnormal feature networks, imported goods carrying the COVID-19 virus are further detected, and the risk is quantitatively assessed.

[0007] The present invention provides a method for assessing the risk of epidemic spread in a cross-border e-commerce network, comprising the following steps:

[0008] Obtain the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network;

[0009] Based on the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network, a multi-layer complex network is constructed;

[0010] Adopt anomaly detection algorithm to perform cross-layer anomaly detection on multi-layer complex networks;

[0011] Determine abnormal logistics information based on abnormal detection results;

[0012] Conduct epidemic transmission risk assessment based on abnormal logistics information.

[0013] Furthermore, the step of obtaining the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network includes:

[0014] Collect e-commerce data and viral spread data;

[0015] Build an e-commerce logistics topology network based on e-commerce data;

[0016] Construct a topological network of epidemic transmission paths based on virus transmission data;

[0017] The network features of the e-commerce logistics topology network and the epidemic transmission path topology network were screened respectively. According to the similarity of the network features, the relationship between the nodes of the e-commerce logistics topology network and the epidemic transmission path topology network was obtained, and the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network was obtained.

[0018] Furthermore, the network characteristics include: node attribute characteristics and mutual relationship characteristics.

[0019] Furthermore, the multi-layer complex network is C=(G, M), where G={G i}, i∈{1,…,n} represents a set of layer graphs, G i =(V i ,E i ) is the i-th layer graph, E ij represents the aligned edge between the two layers of graph i and j, V i represents the set of graph nodes at the i-th layer, E i represents the edge set in the i-th layer graph, and G1 is the global epidemic network.

[0020] Furthermore, the step of using an anomaly detection algorithm to perform cross-layer anomaly detection on a multi-layer complex network includes:

[0021] Anomaly detection algorithm is used to detect abnormal subgraphs of G1 layer in multi-layer complex networks;

[0022] Based on the abnormal subgraph of G1 layer, modeling is carried out according to the similarity law;

[0023] Apply the network alignment method and calculate the alignment matrix;

[0024] According to the feature transfer learning method, the target layer G is calculated i The abnormal subgraph of .

[0025] Furthermore, the objective function of the anomaly detection algorithm is:

[0026]

[0027]

[0028] Among them, α is the abnormal significance level, S1 is the abnormal subgraph of G1, and F α (S1) represents the outlier evaluation function of S1 for the outlier significance level α, Φ represents a parameter-free scanning statistic function, N α (S1) represents the number of abnormal nodes in S1, N(S1) represents the total number of nodes in S1, S i G i The abnormal subgraph, M 1i G1 and G i The alignment relationship matrix of , T represents the feature transfer function.

[0029] Furthermore, the risk assessment function of the epidemic spread risk assessment is A: The abnormal subgraph S i =(V i ,E i ), when A(V i )≥ξ, it is judged to have reached the risk warning level, where ξ is the risk warning threshold parameter.

[0030] Furthermore, it also includes:

[0031] Trace historical commodity transaction data based on abnormal logistics information;

[0032] Track the logistics tracks of different batches based on historical commodity transaction data;

[0033] The epidemic characteristic data with time series are integrated to detect the remaining logistics information overlapping with the abnormal logistics space.

[0034] Compared with the prior art, the present invention has the following significant advantages:

[0035] The present invention proposes a method for assessing the risk of epidemic spread in a cross-border e-commerce network. By acquiring e-commerce data and virus transmission data, an e-commerce logistics topology network and an epidemic spread path topology network are established. According to the relationship between the two networks, a multi-layer complex network is established, and cross-layer anomaly detection is performed on the multi-layer complex network to determine and lock abnormal logistics information and conduct epidemic spread risk assessment. The present invention provides a method for assessing the risk of epidemic spread in a cross-border e-commerce network. It is applied to a cross-border e-commerce network, and by detecting abnormal subgraphs in a weak abnormal feature network, imported goods carrying viruses are further detected, and the risk is quantitatively assessed. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 An evaluation framework diagram of an epidemic spread risk assessment method for a cross-border e-commerce network provided by an embodiment of the present invention;

[0037] Figure 2 A multi-layer complex network structure diagram of a method for assessing the risk of epidemic spread in a cross-border e-commerce network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following is a clear and complete description of the technical solutions of the embodiments of the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0039] This paper mainly studies the quantitative assessment of the risk of imported goods carrying the COVID-19 virus, thereby causing the spread of the epidemic (abnormal events), in a complex network model built based on cross-border e-commerce transaction data. It belongs to the anomaly detection problem based on the weak abnormal feature graph structure.

[0040] As mentioned above, whether it is graph-based or deep learning-based anomaly detection technology, the detection performance of existing algorithms is largely dependent on the amount of abnormal feature data, and in multi-layer complex networks with "weak (few) abnormal features", there is a general problem of low detection accuracy. In order to solve this problem, the present invention needs to solve the following two key problems: (1) Explore the similarity of node features or network structures between cross-border e-commerce networks and global epidemic networks, and implement network alignment algorithms in order to realize multi-layer complex network models; (2) In this network, study the optimization algorithm that integrates network alignment and anomaly detection, in order to realize an effective anomaly detection algorithm based on network alignment.

[0041] The present invention will not only be limited to the risk assessment of the currently popular new coronavirus pneumonia, but will also be applicable to the risk assessment of all infectious disease outbreaks, which will be conducive to promoting the rapid development of the public health safety prevention and control system.

[0042] Example 1

[0043] Reference Figure 1 and Figure 2 The present invention provides a method for assessing the risk of epidemic spread in a cross-border e-commerce network, comprising the following steps:

[0044] Step S1: Collect e-commerce data and virus propagation data, and crawl relevant transaction data on multiple cross-border e-commerce platforms, and crawl virus propagation data from websites;

[0045] Step S2: construct an e-commerce logistics topology network based on e-commerce data;

[0046] Step S3: Construct an epidemic transmission path topology network based on virus transmission data;

[0047] Step S4: The network features of the e-commerce logistics topology network and the epidemic transmission path topology network are screened respectively, and the mutual relationship between the nodes of the e-commerce logistics topology network and the epidemic transmission path topology network is obtained according to the similarity of the network features, and the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network is obtained, wherein the network features include: node attribute features and mutual relationship features, and the nodes include national nodes and regional nodes;

[0048] Step S5: Based on the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network, a multi-layer complex network is constructed, where Figure 1 As shown, the multi-layer complex network is C = (G, M), where G = {G i}, i∈{1,…,n} represents a set of layer graphs, G i =(V i ,E i ) is the i-th layer graph, E ij represents the aligned edge between the two layers of graph i and j, V i represents the set of graph nodes at the i-th layer, E i represents the edge set in the i-th layer graph, G1 is the global epidemic network, G i , i∈{2,…,n} is a cross-border e-commerce network, and the data with real physical meaning is selected so that it can build an effective alignment relationship between network layer graphs (i.e., M={E ij}) algorithm in order to build a reasonable multi-layer network model;

[0049] Step S6: For the target layer G iThe abnormal feature value is not obvious or does not exist, which makes it impossible to detect anomalies. The steps of using anomaly detection algorithm to perform cross-layer anomaly detection on multi-layer complex networks include:

[0050] Anomaly detection algorithm is used to detect abnormal subgraphs of G1 layer in multi-layer complex networks;

[0051] Based on the abnormal subgraph of G1 layer, modeling is carried out according to the similarity law;

[0052] Apply the network alignment method and calculate the alignment matrix;

[0053] According to the feature transfer learning method, the target layer G is calculated i The abnormal subgraph of .

[0054] Among them, the objective function of the anomaly detection algorithm is:

[0055]

[0056]

[0057] Among them, α is the abnormal significance level, S1 is the abnormal subgraph of G1, and F α (S1) represents the outlier evaluation function of S1 for the outlier significance level α, Φ represents a parameter-free scanning statistic function, N α (S1) represents the number of abnormal nodes in S1, N(S1) represents the total number of nodes in S1, S i G i The abnormal subgraph, M 1i G1 and G i The alignment relationship matrix of , T represents the feature transfer function.

[0058] Step S7: Determine abnormal logistics information according to the abnormal detection result;

[0059] Step S8: Based on the abnormal logistics information, the epidemic spread risk assessment is performed, where the risk assessment function is A: The abnormal subgraph S i =(V i ,E i ), when A(V i )≥ξ, it is judged to have reached the risk warning level, where ξ is the risk warning threshold parameter.

[0060] Step S9: Tracing the historical transaction data of the commodity based on the abnormal logistics information;

[0061] Step S10: Tracking the logistics tracks of different batches based on the historical transaction data of the goods;

[0062] Step S11: Fuse the epidemic characteristic data with time series to detect the remaining logistics information overlapping with the abnormal logistics space.

[0063] The technical solution of the present invention makes the following research:

[0064] (1) Research on the construction method of multi-layer complex network based on cross-border e-commerce data

[0065] Since the destructive effect of high temperature on the lipid layer of coronavirus has been recognized by the industry, cold chain transportation has become an important way for virus transmission during logistics and transportation. The present invention will focus on collecting various transaction data related to cold chain transportation of cross-border e-commerce platforms, and build a network model based on this. According to network science theory, the networked expression of human society using real data will have the characteristics of a complex network, and the network constructed based on the data set of the cross-border e-commerce platform is no exception. Its complexity is mainly manifested in the huge number of nodes, the topological structure has different characteristics, and the node attributes are diversified. In addition to integrating the data collected by multiple cross-border e-commerce platforms (such as Amazon, Taobao, wish, eBay, etc.), the present invention also collects global epidemic-related data and builds a multi-layer complex network model on this basis.

[0066] (2) Research on cross-layer abnormal subgraph detection technology based on network alignment

[0067] Network alignment refers to finding the mapping relationship between nodes in different networks through similarity features. Using network alignment technology, we establish an alignment relationship between network nodes based on the similarity relationship between the cross-border e-commerce network and the global COVID-19 epidemic network. Then, based on the detection of abnormal subgraphs in the global COVID-19 epidemic network, we use feature migration technology to perform anomaly detection in the cross-border e-commerce network with weak abnormal features.

[0068] (3) Research on epidemic transmission risk assessment methods

[0069] According to the anomaly detection algorithm, the abnormal risk quantification value of the cross-border e-commerce network node (e-commerce or logistics) is obtained to evaluate the risk of imported (cold chain) logistics carrying the COVID-19 virus.

[0070] The research idea of ​​this embodiment is novel: existing anomaly detection algorithms all rely on strong anomaly features to obtain better detection results, while the present invention studies anomaly detection technology in a network with weak anomaly features, and specifically applies it to a cross-border e-commerce network to conduct risk assessment on the spread of the epidemic by imported goods, which meets social needs and has extremely strong practical significance.

[0071] The application scenario of this embodiment is novel: in the risk assessment of the spread of the epidemic by imported goods, the combination of cross-border e-commerce data will be more comprehensive than simply evaluating the epidemic data of the importing country. The present invention not only considers the epidemic situation in the importing country, but also traces the historical transaction data of the goods, tracks the logistics trajectories of different batches, and integrates the epidemic characteristic data with time series to detect goods that cause the risk of epidemic spread through other countries. The accuracy and recall rate of risk assessment of this applied research will be improved.

[0072] The above disclosures are only several specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for assessing the risk of epidemic spread in cross-border e-commerce networks, characterized in that: The following steps are involved: Obtain the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network; Based on the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network, a multi-layer complex network is constructed; Adopt anomaly detection algorithm to perform cross-layer anomaly detection on multi-layer complex networks; Determine abnormal logistics information based on abnormal detection results; Conduct epidemic spread risk assessment based on abnormal logistics information; The step of obtaining the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network includes: Collect e-commerce data and viral spread data; Build an e-commerce logistics topology network based on e-commerce data; Construct a topological network of epidemic transmission paths based on virus transmission data; The network features of the e-commerce logistics topology network and the epidemic transmission path topology network are screened respectively. According to the similarity of the network features, the mutual relationship between the nodes of the e-commerce logistics topology network and the epidemic transmission path topology network is obtained, and the alignment relationship between the e-commerce logistics topology network and the epidemic transmission path topology network is obtained. The network features include: node attribute features and mutual relationship features; The multi-layer complex network is C = (G, M), where G = {G i }, i∈{1,…,n} represents a set of layer graphs, G i =(V i ,E i ) is the i-th layer graph, E ij represents the aligned edge between the two layers of graph i and j, V i represents the set of graph nodes at the i-th layer, E i represents the edge set in the i-th layer graph, G1 is the global epidemic network; The step of using anomaly detection algorithm to perform cross-layer anomaly detection on a multi-layer complex network includes: Anomaly detection algorithm is used to detect abnormal subgraphs of G1 layer in multi-layer complex networks; Based on the abnormal subgraph of G1 layer, modeling is carried out according to the similarity law; Apply the network alignment method and calculate the alignment matrix; According to the feature transfer learning method, the target layer G is calculated i The abnormal subgraph of The objective function of the anomaly detection algorithm is: Among them, α is the abnormal significance level, S1 is the abnormal subgraph of G1, and F α (S1) represents the outlier evaluation function of S1 for the outlier significance level α, Φ represents a parameter-free scanning statistic function, N α (S1) represents the number of abnormal nodes in S1, N(S1) represents the total number of nodes in S1, S i G i The abnormal subgraph, M 1i G1 and G i The alignment relationship matrix, T represents the feature transfer function; The risk assessment function for the epidemic spread risk assessment is A: The abnormal subgraph S i =(V i ,E i ), when A(V i )≥ξ, it is judged to have reached the risk warning level, where ξ is the risk warning threshold parameter.

2. The method for assessing the risk of epidemic spread in a cross-border e-commerce network according to claim 1, characterized in that: Also includes: Trace historical commodity transaction data based on abnormal logistics information; Track the logistics tracks of different batches based on historical commodity transaction data; The epidemic characteristic data with time series are integrated to detect the remaining logistics information overlapping with the abnormal logistics space.

Images