Business processing method, device, and electronic equipment

Abstract

An embodiment of the present invention provides a method and device for business processing, and an electronic device. The method includes: clustering a feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into multiple clusters, bridge nodes, and outlier nodes; based on the file features of the file to be classified, filtering the decision tree in the order of outlier nodes, bridge nodes, nodes within the cluster, and core nodes to obtain a target decision tree; the first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening; classifying the file to be classified based on the target decision tree to determine the file type of the file to be classified; and initiating a first business of the corresponding file type when the file to be classified passes the verification strategy corresponding to the file type. In the embodiment of the present invention, not only does multiple screening reduce the number of decision trees for determining the file type of the file to be classified, but it also improves business processing efficiency.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a business processing method and device, and electronic equipment. Background Art

[0002] In the current annuity system, a large number of business documents for different businesses are circulated between various institutions. Even the form formats and attribute characteristics of documents of the same business type (such as transfer forms, payment business forms, and payment details forms) in different institutions are different. Such a large number of complex business documents initially required sales staff to manually classify each file category.

[0003] At present, in order to improve the classification efficiency of business documents, decision trees are used to automatically classify business documents. Figure 1 The following diagram illustrates the current business processing process. Business processing organizations typically transmit business documents requiring processing to designated sales representatives. These representatives then upload the received business documents to an automated system for automatic classification and file type determination. Parsing and verification are then performed. If verification succeeds, the business is initiated. If verification fails, the sales representative manually modifies the business document and re-parses and verifies the modified document.

[0004] However, as business continues to grow, the number of file types is also increasing. To avoid overloading a decision tree with features, it is usually necessary to train a large number of decision trees containing different features. Therefore, when determining the file type, all decision trees must be traversed, which is undoubtedly inefficient. Summary of the Invention

[0005] The embodiments of the present invention provide a method and apparatus for business processing, and an electronic device to solve the problem in the prior art that traversing all decision trees leads to low efficiency in file classification during business processing.

[0006] In a first aspect, an embodiment of the present invention provides a service processing method, the method comprising:

[0007] Clustering a feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into a plurality of clusters, bridge nodes that do not belong to the cluster and connect to more than one cluster, and outlier nodes that do not belong to the cluster and connect to less than two clusters, wherein the nodes in the cluster include core nodes and intra-cluster nodes, and the feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files;

[0008] Based on the file features in the file to be classified, the decision tree is screened in the order of the outlier nodes, the bridge nodes, the cluster nodes, and the core nodes to obtain a target decision tree, wherein the first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening;

[0009] Classifying the files to be classified based on the target decision tree to determine the file types of the files to be classified;

[0010] In a case where the file to be classified passes the verification policy corresponding to the file type, a first service corresponding to the file type is initiated.

[0011] Optionally, there are multiple file features, and the step of screening the decision tree to obtain a target decision tree based on the file features in the file to be classified in the order of the outlier nodes, the bridge nodes, the cluster nodes, and the core nodes includes:

[0012] Filtering the decision trees containing the target outlier node in the feature decision graph to obtain a first decision tree set, wherein the target outlier node is an outlier node having any of the file features among the outlier nodes in the feature decision graph;

[0013] Filtering the decision trees containing the target bridge node in the first decision tree set to obtain a second decision tree set, wherein the target bridge node is a bridge node having any of the file features among the bridge nodes corresponding to the decision trees in the first decision tree set;

[0014] Filtering the decision trees in the second decision tree set that contain target cluster nodes to obtain a third decision tree set, wherein the target cluster nodes are cluster nodes corresponding to the decision trees in the second decision tree set that have any of the file features;

[0015] Filtering the decision tree containing the target core node in the third decision tree set to obtain a fourth decision tree set, wherein the target core node is a core node having any of the file features among the core nodes corresponding to the decision trees in the third decision tree set;

[0016] A decision tree in the fourth decision tree set is determined as a target decision tree.

[0017] Optionally, the filtering of the decision trees including the target bridge node in the first decision tree set to obtain the second decision tree set includes:

[0018] When the number of decision trees in the first decision tree set is greater than or equal to a target threshold, screening the decision trees in the first decision tree set that include the target bridge node to obtain a second decision tree set;

[0019] The step of screening the second decision tree set for decision trees containing nodes in the target cluster to obtain a third decision tree set includes:

[0020] When the number of decision trees in the second decision tree set is greater than or equal to the target threshold, screening the decision trees in the second decision tree set that contain nodes in the target cluster to obtain a third decision tree set;

[0021] The step of screening the decision tree containing the target core node in the third decision tree set to obtain a fourth decision tree set includes:

[0022] When the number of decision trees in the third decision tree set is greater than or equal to the target threshold, screening the decision trees in the third decision tree set that contain the target core node to obtain a fourth decision tree set;

[0023] Accordingly, the method further includes:

[0024] When the number of decision trees in a target set is less than the target threshold, the decision tree in the target set is determined as a target decision tree, wherein the target set includes: the first decision tree set, the second decision tree set, or the third decision tree set.

[0025] Optionally, the filtering of the decision trees including the target bridge node in the first decision tree set to obtain the second decision tree set includes:

[0026] Deleting nodes corresponding to decision trees that do not include the target outlier node in the feature decision graph to obtain a first intermediate decision graph;

[0027] Filtering the decision tree including the target bridge node in the first intermediate decision graph to obtain the second decision tree set;

[0028] The step of screening the second decision tree set for decision trees containing nodes in the target cluster to obtain a third decision tree set includes:

[0029] Deleting nodes corresponding to the decision tree that do not include the target bridge node in the first intermediate decision graph to obtain a second intermediate decision graph;

[0030] Filtering the decision trees containing the nodes in the target cluster in the second intermediate decision graph to obtain the third decision tree set;

[0031] The step of screening the decision tree containing the target core node in the third decision tree set to obtain a fourth decision tree set includes:

[0032] Deleting nodes corresponding to decision trees that do not include nodes in the target cluster in the second intermediate decision graph to obtain a third intermediate decision graph;

[0033] The decision tree including the target core node is screened in the third intermediate decision graph to obtain the fourth decision tree set.

[0034] Optionally, after classifying the to-be-classified files based on the target decision tree and determining the file types of the to-be-classified files, the method further includes:

[0035] When the file to be classified fails to pass the verification strategy corresponding to the file type, obtaining multiple business files including the core node of the cluster;

[0036] Comparing the file features of the file to be classified with the file features of each of the business files in the order of outlier nodes, bridge nodes, and nodes within the cluster, and determining a score for each of the business files, wherein for each of the business files, the greater the score of the business file, the higher the similarity between the business file and the file to be classified;

[0037] Re-determining the file type of the target business file with the largest score as the file type of the file to be classified;

[0038] In a case where the file to be classified passes the verification policy corresponding to the re-determined file type, a second service corresponding to the re-determined file type is initiated.

[0039] Optionally, if the file to be classified fails to pass the verification policy corresponding to the re-determined file type, repairing the file features of the file to be classified based on the file feature template of the target business file;

[0040] In a case where the repaired file to be classified passes the verification policy corresponding to the re-determined file type, the second service is initiated.

[0041] Optionally, before clustering the feature decision graph using a graph clustering method and clustering the nodes in the feature decision graph into a plurality of clusters, bridge nodes that do not belong to the cluster and connect to more than one cluster, and outlier nodes that do not belong to the cluster and connect to less than two clusters, the method further includes:

[0042] Determine multiple file types based on the files corresponding to different businesses in different business organizations;

[0043] For each of the file types, a decision tree is obtained by training using the training data, wherein each decision tree includes a plurality of file features for distinguishing the file types;

[0044] The nodes corresponding to the same file features in different decision trees are overlapped into the same node to obtain the feature decision graph.

[0045] In a second aspect, an embodiment of the present invention further provides a service processing device, the device comprising:

[0046] A clustering module is used to cluster the feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into multiple clusters, bridge nodes that do not belong to the cluster and connect to more than one cluster, and outlier nodes that do not belong to the cluster and connect to less than two clusters, wherein the nodes in the cluster include core nodes and intra-cluster nodes, and the feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files;

[0047] A screening module is configured to screen the decision tree to obtain a target decision tree in the order of the outlier nodes, the bridge nodes, the cluster nodes, and the core nodes based on the file features in the file to be classified, wherein the first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening;

[0048] A determination module, configured to classify the to-be-classified files based on the target decision tree and determine the file type of the to-be-classified files;

[0049] The service initiation module is configured to initiate a first service corresponding to the file type if the file to be classified passes the verification policy corresponding to the file type.

[0050] In a third aspect, an embodiment of the present invention further provides an electronic device comprising a memory, a processor, a bus, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above-mentioned business processing method when executing the program.

[0051] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the business processing method described above are implemented.

[0052] In an embodiment of the present invention, a feature decision graph formed by connecting multiple decision trees for classifying files is clustered using a graph clustering method, thereby clustering the nodes in the feature decision graph into multiple clusters, bridge nodes, and outlier nodes. The structural similarity between the nodes is used to characterize the information contained in the nodes or the degree of influence on distinguishing file types. The decision trees are then screened in the order of outlier nodes, bridge nodes, cluster nodes, and core nodes to obtain a target decision tree. Since each screening, except for the initial screening, is screened from the screening results of the previous screening, the number of target decision trees obtained by the final screening is relatively small, so that the classification of the files to be classified can be achieved using fewer target decision trees, thereby initiating the corresponding first service. In an embodiment of the present invention, after clustering the feature decision graph, not only is the number of decision trees used to determine the file type of the files to be classified reduced by multiple screenings, thereby improving service processing efficiency, but also, by screening in descending order of influence on distinguishing file types, it is possible to avoid missing decision trees with greater influence, thereby ensuring the accuracy of file classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0054] Figure 1 A flowchart of the current business process;

[0055] Figure 2 A flowchart of the steps of the business processing method provided by an embodiment of the present invention;

[0056] Figure 3 This is a flowchart of the current processing method for verification failure;

[0057] Figure 4 A schematic diagram showing a feature decision diagram provided in an embodiment of the present invention;

[0058] Figure 5 This is the file verification and repair process in the embodiment of the present invention;

[0059] Figure 6 A flowchart of the actual application of the business processing method provided by an embodiment of the present invention;

[0060] Figure 7 This is one of the decision tree diagrams of an embodiment of the present invention;

[0061] Figure 8This is a second schematic diagram showing a decision tree according to an embodiment of the present invention;

[0062] Figure 9 A structural block diagram of a device for service processing according to an embodiment of the present invention;

[0063] Figure 10 This is a structural block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0065] It should be understood that references throughout this specification to "one embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the present invention. Therefore, the appearances of "in one embodiment" or "in an embodiment" throughout this specification do not necessarily refer to the same embodiment. Furthermore, these particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0066] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0067] See also Figure 2 , an embodiment of the present invention provides a service processing method, the method comprising:

[0068] Step 201: clustering the feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into multiple clusters, bridge nodes that do not belong to a cluster and are connected to more than one cluster, and outlier nodes that do not belong to a cluster and are connected to less than two clusters.

[0069] It should be noted that graph clustering refers to graph-based clustering, which clusters the nodes in the graph into different clusters. Among them, some nodes may not belong to any cluster. The graph clustering process can also be understood as dividing closely connected nodes together. For example, if a certain number of nodes are closely connected, then after graph clustering, these nodes are located in the same cluster. When a node is closely connected to nodes in at least two different clusters, and is connected to these at least two clusters, then the node is a bridge node, which does not belong to any cluster. When a node is connected to only one cluster, and is not closely connected to the nodes in the cluster or the node is not connected to any cluster, then the node is an outlier node.

[0070] It is understood that the nodes in a cluster include core nodes and intra-cluster nodes, where intra-cluster nodes are nodes in the cluster other than the core nodes. Each cluster has one and only one core node. A feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files. Each node in a decision tree corresponds to a file feature, which can be used to classify files. Multiple decision trees contain nodes corresponding to the same file feature and nodes corresponding to different file features. Based on the nodes corresponding to the same file feature, different decision trees can be connected to form a feature decision graph.

[0071] Step 202: Based on the file features in the file to be classified, the decision tree is screened in the order of outlier nodes, bridge nodes, in-cluster nodes, and core nodes to obtain a target decision tree.

[0072] It should be noted that in the feature decision graph after clustering, the structural similarity between nodes can represent the information contained in the nodes or the degree of influence on file type differentiation. Outlier nodes have lower structural similarity with other nodes, so they contain less information and have a greater impact on file type differentiation. The order of outlier nodes, bridge nodes, cluster nodes, and core nodes is the order of greatest influence on file type differentiation.

[0073] Here, the first screening is performed from the feature decision diagram, and each screening except the first screening is performed from the screening results of the previous screening, so that a smaller number of target decision trees can be obtained through layer-by-layer screening.

[0074] Screening in the order of outlier nodes, bridge nodes, in-cluster nodes, and core nodes can be understood as using a greedy algorithm to classify the files being classified, meaning that the optimal and fastest classification is always made at the moment. Alternatively, the algorithm doesn't consider the optimal result for overall file classification, but rather seeks the local optimal solution for the current file classification. Outlier nodes in the clustering results are considered the nodes most capable of determining the file type. The order of these nodes is bridge nodes, in-cluster nodes, and core nodes. This order is determined because core nodes have the most complex connections with many nodes, meaning they possess the highest information entropy and are the least likely to be the truth, making them the most complex. Outlier nodes, on the other hand, can be considered the nodes with the least information entropy and are the most capable of determining the file type. Therefore, using a greedy algorithm, all nodes in the clustering results are divided into four categories.

[0075] Step 203: Classify the files to be classified based on the target decision tree and determine the file type of the files to be classified.

[0076] It should be noted that a voting mechanism can be adopted, which can be various voting mechanisms such as minority obeys majority or weighted majority, and then the judgment results of each target decision tree are statistically analyzed using the determined voting mechanism to obtain the classification probability of various file types, and the file type corresponding to the classification probability with the largest value is determined as the file type of the file to be classified.

[0077] Step 204: When the file to be classified passes the verification policy corresponding to the file type, a first service corresponding to the file type is initiated.

[0078] It should be noted that each file type corresponds to a verification strategy, which can verify whether the file of that file type has errors. If there are errors, the verification fails, otherwise the verification passes.

[0079] In an embodiment of the present invention, a feature decision graph formed by connecting multiple decision trees for classifying files is clustered using a graph clustering method, thereby clustering the nodes in the feature decision graph into multiple clusters, bridge nodes, and outlier nodes. The structural similarity between the nodes is used to characterize the information contained in the nodes or the degree of influence on distinguishing file types. The decision trees are then screened in the order of outlier nodes, bridge nodes, cluster nodes, and core nodes to obtain a target decision tree. Since each screening, except for the initial screening, is screened from the screening results of the previous screening, the number of target decision trees obtained by the final screening is relatively small, so that the classification of the files to be classified can be achieved using fewer target decision trees, thereby initiating the corresponding first service. In an embodiment of the present invention, after clustering the feature decision graph, not only is the number of decision trees used to determine the file type of the files to be classified reduced by multiple screenings, thereby improving service processing efficiency, but also, by screening in descending order of influence on distinguishing file types, it is possible to avoid missing decision trees with greater influence, thereby ensuring the accuracy of file classification.

[0080] Optionally, the graph clustering method may be pSCAN (Fast and exact structural graph clustering).

[0081] It should be noted that pSCAN makes three observations about graph clustering: first, clusters in a graph cluster may overlap; second, core node clusters are disjoint; and third, clusters of non-core nodes are uniquely determined by core nodes. Therefore, pSCAN develops a two-step graph clustering paradigm that first partitions all cluster core nodes into clusters and then clusters non-core nodes into the same cluster by assigning each non-core node to its adjacent core nodes.

[0082] Specifically, for each node v∈V, where v represents a node in the graph structure and V represents the set of all nodes in the graph structure, pSCAN incrementally maintains an effective degree ed(v) and a similarity sd(v), with ed(v)≥sd(v). A node v with sd(v)≥μ is considered a core node, and ed(v)<μ is considered a non-core node, where μ is the minimum number of neighbors required to become a core node. Therefore, the set of core nodes can be effectively identified. Then, the cluster of core nodes is calculated based on the transitive property. If (u,v)∈E, where u and v represent two different core nodes, and E represents an edge in the graph structure, (u,v)∈E indicates that core node u and core node v are connected in the graph structure, and (u,v) is considered an edge in the graph structure. If two core nodes u and v are in the same cluster and have similar structures, they are assigned to the same cluster. In addition, if two core nodes u and v have been assigned to the same cluster, pSCAN clustering will not calculate the structural similarity between them. Therefore, pSCAN clustering saves a lot of structural similarity calculation times.

[0083] Node structure:

[0084] Let v∈V, the node structure of node v is defined by the node v and its neighbor nodes, denoted by Γ(v), that is, the node structure of node v is represented by the set consisting of the node v and its neighbor nodes.

[0085] Γ(v)={w∈V|(v,w)∈E}∪{v};

[0086] Structural similarity:

[0087] Structural similarity is defined as the ratio of the number of common neighbors of two nodes to the geometric mean of the number of neighbor nodes of the two nodes (the neighbor nodes here include the node itself).

[0088]

[0089] Here, Γ(v) represents the set consisting of node v and its adjacent nodes; Γ(w) represents the set consisting of node w and its adjacent nodes.

[0090] e-node:

[0091] The e-neighborhood of a node is defined as the set of nodes whose structural similarity to it is no less than e.

[0092] N e (v)={w∈Γ(v)|σ(v,w)≥e};

[0093] Core nodes:

[0094] A core node is a node whose number of e-neighbors is greater than or equal to the threshold μ;

[0095]

[0096] μ is the minimum number of neighbors that a node needs to have to become a core node. It is set by the system and can also be manually input to adjust the value. e When (v)≥μ, the node is a core node.

[0097] Bridge node:

[0098] If an isolated node v has neighbors that belong to two or more different clusters, it is a bridge node (node ​​v is not a member of any cluster, and node v connects different clusters).

[0099] Outlier nodes:

[0100] An isolated node v is an outlier if and only if all of its neighbors belong to only one cluster or do not belong to any cluster (node ​​v is not a member of any cluster and node v does not connect different clusters).

[0101] For example, in the figure above, given a threshold μ of 4 and a structural similarity threshold of 0.7, after the pSCAN clustering algorithm, the clustering results are: nodes 0 and 11 are core nodes, node 6 is a bridge node, nodes 0, 1, 2, 3, 4, and 5 are clustered into one cluster, and nodes 7, 8, 9, 10, and 11 are clustered into another cluster.

[0102] For example, the business files of Bank A form a cluster, and the business files of Pension Company B form another cluster. Each cluster contains all the business files. The present invention stores the structural similarity between all nodes in a HashMap for easy access. The HashMap stores the key in the form of {node m, node n}, and the value is the structural similarity (to two decimal places).

[0103] In the embodiment of the present invention, pSCAN is used to cluster the feature decision graph, which can improve the efficiency of graph clustering.

[0104] Optionally, there are multiple file features. File features are extracted from the file to be classified through feature extraction. Preferably, when extracting file features, file features in a feature set can be extracted. The feature set is a set consisting of all file features involved in the feature decision graph.

[0105] Based on the file features in the files to be classified, the decision tree is screened in the order of outlier nodes, bridge nodes, cluster nodes, and core nodes to obtain the target decision tree, including:

[0106] Decision trees containing target outlier nodes in the feature decision graph are screened to obtain a first decision tree set, wherein the target outlier node is an outlier node having any file feature among the outlier nodes in the feature decision graph.

[0107] It should be noted that, based on multiple file features of the file to be classified, the target outlier node can be determined from all outlier nodes in the feature decision graph. After the target outlier node is determined, the decision tree to which the target outlier node belongs, i.e., the decision tree containing the target outlier node, is determined. The first decision tree set is a set of decision trees containing the target outlier node, and each decision tree in the first decision tree set includes the target outlier node. The process of determining the target outlier node can be: each file feature of the file to be classified is matched with the file features corresponding to all outlier nodes, and the outlier node that successfully matches is determined as the target outlier node. Here, a successful match can be when the two file features are identical.

[0108] Decision trees including a target bridge node in the first decision tree set are screened to obtain a second decision tree set, wherein the target bridge node is a bridge node having any file feature among bridge nodes corresponding to the decision trees in the first decision tree set.

[0109] It should be noted that the first decision tree set is the result of the first screening. Based on multiple file features of the file to be classified, the target bridge node can be determined from all bridge nodes corresponding to the first screening result. After determining the target bridge node, the decision tree to which the target bridge node belongs is determined, i.e., the decision tree in the first decision tree set that includes the target bridge node. The second decision tree set is the set of decision trees in the first decision tree set that include the target bridge node, and each decision tree in the second decision tree set includes the target bridge node. The process of determining the target bridge node is similar to that of determining the target outlier node and will not be repeated here.

[0110] The decision trees in the second decision tree set that contain the target cluster node are screened to obtain a third decision tree set, wherein the target cluster node is the cluster node having any file feature among the cluster nodes corresponding to the decision trees in the second decision tree set.

[0111] It should be noted that the second decision tree set is the screening result of the second screening. Based on multiple file features of the file to be classified, the target in-cluster node can be determined from all in-cluster nodes corresponding to the screening result of the second screening. After determining the target in-cluster node, the decision tree to which the target in-cluster node belongs is determined, i.e., the decision tree in the second decision tree set that includes the target in-cluster node. The third decision tree set is the set of decision trees in the second decision tree set that include the target in-cluster node, and each decision tree in the third decision tree set includes the target in-cluster node. The process of determining the target in-cluster node is similar to the process of determining the target outlier node and will not be repeated here.

[0112] The decision tree containing the target core node in the third decision tree set is screened to obtain a fourth decision tree set, wherein the target core node is a core node having any file feature among the core nodes corresponding to the decision trees in the third decision tree set.

[0113] It should be noted that the third decision tree set is the screening result of the third screening. Based on multiple file features of the file to be classified, the target core node can be determined from all core nodes corresponding to the screening result of the third screening. After determining the target core node, the decision tree to which the target core node belongs is determined, that is, the decision tree in the third decision tree set that includes the target core node. The fourth decision tree set is the set of decision trees in the third decision tree set that include the target core node, and each decision tree in the fourth decision tree set includes the target core node. The process of determining the target core node is similar to the process of determining the target outlier node and will not be repeated here.

[0114] A decision tree in the fourth decision tree set is determined as a target decision tree.

[0115] In an embodiment of the present invention, the next screening is performed from the screening results of the previous screening, so that the number of decision trees in the screening results changes with a decreasing trend, and finally a smaller number of target decision trees are obtained, thereby improving the classification efficiency and further improving the business processing efficiency.

[0116] Optionally, the decision tree containing the target bridge node in the first decision tree set is filtered to obtain a second decision tree set, including:

[0117] When the number of decision trees in the first decision tree set is greater than or equal to the target threshold, screening the decision trees in the first decision tree set that include the target bridge node to obtain a second decision tree set;

[0118] Filter the decision trees that contain nodes in the target cluster in the second decision tree set to obtain a third decision tree set, including:

[0119] When the number of decision trees in the second decision tree set is greater than or equal to the target threshold, screening the decision trees in the second decision tree set that contain nodes in the target cluster to obtain a third decision tree set;

[0120] Filter the decision trees containing the target core node in the third decision tree set to obtain a fourth decision tree set, including:

[0121] When the number of decision trees in the third decision tree set is greater than or equal to the target threshold, screening the decision trees in the third decision tree set that contain the target core node to obtain a fourth decision tree set;

[0122] Accordingly, the method further includes:

[0123] When the number of decision trees in the target set is less than a target threshold, the decision tree in the target set is determined as a target decision tree, wherein the target set includes: a first decision tree set, a second decision tree set, or a third decision tree set.

[0124] It should be noted that, for any one of the first decision tree set, the second decision tree set, and the third decision tree set, there is a first situation in which the number of decision trees in the set is greater than or equal to the target threshold value, and a second situation in which the number of decision trees is less than the target threshold value. Therefore, different processing methods can be adopted for different situations. In the first case, it means that the number of decision trees obtained by screening is large. At this time, if the file type of the file to be classified is determined based on the decision trees obtained by screening, it is relatively time-consuming. Therefore, the next round of screening can be carried out from the results of this round of screening, and so on until the screening is carried out by core nodes or until the number of decision trees in the screening results is small. Here, until the number of decision trees in the screening results is small, that is, until the second case. The target threshold value can be set according to demand, wherein the target threshold value is usually small, for example, it can be 1 or 2, but is not limited to this.

[0125] In an embodiment of the present invention, a target threshold is set for the screening results of each screening. When the number of decision trees in the screening results is less than the target threshold, it means that the number of target decision trees is small, and the file type of the file to be classified can be determined in a shorter time. At this time, there is no need to continue screening, thereby saving business processing time.

[0126] Optionally, the decision tree containing the target bridge node in the first decision tree set is filtered to obtain a second decision tree set, including:

[0127] Delete the nodes corresponding to the decision trees that do not contain the target outlier node in the feature decision graph to obtain a first intermediate decision graph;

[0128] The decision tree including the target bridge node is screened in the first intermediate decision graph to obtain a second decision tree set.

[0129] It should be noted that by deleting the nodes corresponding to the decision trees that do not contain the target outlier node in the feature decision graph, the number of nodes in the feature decision graph can be reduced. The deleted nodes are the nodes that do not need to be considered, and the retained nodes are the nodes corresponding to the decision trees in the screening results of the first screening.

[0130] Filter the decision trees that contain nodes in the target cluster in the second decision tree set to obtain a third decision tree set, including:

[0131] Deleting the nodes corresponding to the decision trees that do not include the target bridge node in the first intermediate decision graph to obtain a second intermediate decision graph;

[0132] Decision trees containing nodes in the target cluster are screened in the second intermediate decision graph to obtain a third decision tree set.

[0133] It should be noted that by deleting the nodes corresponding to the decision trees that do not contain the target bridge node in the first intermediate decision graph, the number of nodes in the feature decision graph can be further reduced. The deleted nodes are the nodes that do not need to be considered, and the retained nodes are the nodes corresponding to the decision trees in the second screening results.

[0134] Filter the decision trees containing the target core node in the third decision tree set to obtain a fourth decision tree set, including:

[0135] Delete the nodes corresponding to the decision trees that do not contain the nodes in the target cluster in the second intermediate decision graph to obtain a third intermediate decision graph;

[0136] The decision tree including the target core node is screened in the third intermediate decision graph to obtain a fourth decision tree set.

[0137] It should be noted that by deleting the nodes corresponding to the decision trees that do not contain nodes in the target cluster in the second intermediate decision graph, the number of nodes in the feature decision graph can be further reduced. The deleted nodes are the nodes that do not need to be considered, and the retained nodes are the nodes corresponding to the decision trees in the screening results of the third screening.

[0138] In an embodiment of the present invention, during the layer-by-layer screening process, the nodes in the feature decision graph are gradually deleted, so that each time screening is performed, the screening is performed based on the feature decision graph after the nodes are deleted in the previous screening, thereby improving the screening efficiency.

[0139] Optionally, after classifying the files to be classified based on the target decision tree and determining the file types of the files to be classified, the method further includes:

[0140] When the file to be classified fails to pass the verification strategy corresponding to the file type, multiple business files including the core node of the cluster are obtained.

[0141] It should be noted that there is a possibility of error in the results of file classification, that is, the file type of the file to be classified determined may be different from its actual file type. The error here is not limited to incorrect classification, it may also be erroneous data in the file to be classified, or it may be a business logic error. For example, a file to be classified that should belong to the first file type cannot pass the verification because of a slight error in the format feature (for example, accidentally typing an extra .). Although the probability of this situation is low, once it occurs, it may cause unpredictable losses. Therefore, it is necessary to verify the file to be classified. For each cluster, multiple business files of the core node of the cluster are included. Preferably, each business file represents a file type.

[0142] The file features of the file to be classified are compared with the file features of each business file in the order of outlier nodes, bridge nodes, and nodes within the cluster to determine the score of each business file.

[0143] It should be noted that for each business file, the larger the score of the business file, the higher the similarity between the business file and the file to be classified. Therefore, the similarity between each business file and the file to be classified can be determined based on the business file score, where the higher the similarity, the more likely it is that the two files are of the same file type.

[0144] The file type of the target business file with the largest score is re-determined as the file type of the file to be classified;

[0145] If the file to be classified passes the verification policy corresponding to the re-determined file type, a second service corresponding to the re-determined file type is initiated.

[0146] It should be noted that the current processing methods for verification failure are usually as follows: Figure 3 As shown, after the intelligent classification system determines the file type, the file is parsed by the file parsing system, and the parsing results are then verified. If a classification error occurs, a salesperson will manually modify the file and upload it directly to the business initiation system. This information is also collected to adjust the intelligent classification system. If a data error occurs, the salesperson will also modify the incorrect data in the file and re-upload it to the intelligent classification system. The entire process requires human interaction.

[0147] In the embodiment of the present invention, when the verification fails, the file type of the file to be classified can be re-determined, which solves the problem that some files to be classified cannot pass the verification and saves the manual modification process.

[0148] Optionally, if the file to be classified fails to pass the verification policy corresponding to the re-determined file type, the file features of the file to be classified are repaired based on the file feature template of the target business file;

[0149] When the repaired file to be classified passes the verification policy corresponding to the re-determined file type, the second service is initiated.

[0150] It should be noted that if the re-determined file type still fails to pass verification, the problem may lie within the file itself, i.e., there is erroneous data in the file. In this case, you will need to repair the erroneous data. For example, if a position in the file should contain "-", but it actually contains ",", you can simply change the "," to "-".

[0151] The following is a specific example to illustrate the process of checking and repairing in the embodiment of the present invention. Figure 4 As shown in the figure, assume that node v6 is a file to be classified that has failed verification. Normally, it should be included in the cluster of node v1, but now it does not appear in the cluster. The problem is only a format issue rather than a business logic issue. Since the file features in the file can be used as an outlier node, the file cannot be used as an outlier. In this example, node v6 is used as a file for ease of understanding.

[0152] like Figure 5 As shown, after importing the file to be classified, the file type is determined and verified. If it can be parsed and used, the verification passes and is automatically transferred to the corresponding business node. If it cannot be parsed, the verification fails, meaning the file to be classified fails verification. This indicates that there may be a prediction error or a file feature error. In this case, it is determined whether it is a business logic error. If so, no automatic repair is performed, and the cause of the error can be prompted for manual repair. If it is not a business logic error, the core node file in each cluster is identified and the two files are compared in the order of outlier nodes, bridge nodes, and cluster nodes. If they do not match, the business organization represented by the cluster is excluded. Because core nodes are closely connected to every node in the cluster and contain the most information, comparing the file to be classified with the core node files is the fastest. After the comparison is completed, the file type with the highest score is extracted and reclassified as that type for parsing and verification. If the verification passes, it is automatically transferred to the corresponding business node. If it still fails, a new cluster is selected for reclassification of the error file.

[0153] If all the checks determined by all clusters fail, there may be a problem with the file characteristics of the file to be classified. The file to be classified will be compared with the template. When the error rate is above 5%, it is considered that the file cannot be automatically repaired, and automatic repair will not be performed. The cause of the error may be prompted to be repaired manually. When the error rate is within 5%, the file characteristics of the file to be classified will be modified to the file characteristics of the core node file with the highest clustering score, and the parsing and verification will be re-performed. If it fails, the file to be classified will be rolled back and changed to the file characteristics of the core node file with the second highest clustering score, and the verification will be performed again. This process will be repeated until the end. If all of them fail, the cause of the error will be prompted to be repaired manually.

[0154] In the embodiment of the present invention, the process of manually opening the files to be classified for comparison can be reduced to the greatest extent possible, thereby reducing the manpower and material costs and simultaneously reducing the probability of the files failing the verification.

[0155] Optionally, before clustering the feature decision graph using a graph clustering method and clustering the nodes in the feature decision graph into a plurality of clusters, bridge nodes that do not belong to a cluster and are connected to more than one cluster, and outlier nodes that do not belong to a cluster and are connected to less than two clusters, the method further comprises:

[0156] Based on the files corresponding to different businesses in different business organizations, multiple file types are determined.

[0157] For each file type, a decision tree is obtained by training using the training data, wherein each decision tree includes a plurality of file features for distinguishing file types;

[0158] The nodes corresponding to the same file features in different decision trees are overlapped into the same node to obtain a feature decision graph.

[0159] In the embodiment of the present invention, a decision tree for distinguishing file types is obtained based on training for each file type, and a feature decision graph can be quickly obtained by using nodes corresponding to the same file features in different decision trees.

[0160] Optionally, during the training of the decision tree, you can customize the selection of file features. This can be done based on the size of the feature subset, prediction accuracy, and the class distribution of the original data. The feature subset selected should be as small as possible and effective in identifying the target. It should not reduce, or even improve, the accuracy of the prediction model. The resulting class distribution should be as close as possible to that of the original data.

[0161] The selection of custom file features here can be regarded as a search and optimization problem, the purpose of which is to select file features that have the ability to distinguish file types. If a certain file feature will not have an impact on the current file type distinction or the impact it has is very low, then this file feature can be discarded by default. Specifically, the file feature in the embodiment of the present invention can be a file style feature associated with the file style. When selecting the file style feature, the information gain rate of the file style feature can be calculated, and then the size of the information gain rate can be compared. The larger the information gain rate, the more likely this file style feature is to have an impact on the current file type distinction, and the smaller the information gain rate, the lower the possibility. The C4.5 algorithm can be used here, and the information gain rate is used to replace the information gain in the ID3 algorithm. Pruning optimization will be performed during the construction of the decision tree, because some nodes with very little information may make the constructed decision tree too complex. These nodes will make the space complexity of distinguishing file types higher and consume more space. It is actually a better choice to ignore these nodes.

[0162] Specifically, we can categorize files based on two aspects: the first is the institution name, such as First Bank, Second Bank, First Pension Institution, Third Bank, etc. The second is the business category, such as business forms like employee resignation details, benefits payment details, member payment summary, member annuity payment details, and personnel change tracking forms. We search for file style features based on these two aspects. In the annuity scenario, the file style features are as follows:

[0163] 1. File name format: If the file name format appears in the form of "+", "," and "-", you can split the file name and regard the name before "+", "," and "-" as the organization name, and the name after "+", "," and "-" as the business name. For classified files, different organizations and different businesses can be classified by searching the name format.

[0164] 2. Whether there is a watermark in the file: Different institutions have different file styles. Some institutions will print watermarks on the files to protect their forms. The watermarks on the forms of the same business of the same institution are basically the same. Therefore, whether there is a watermark in the file can be used as a style feature to distinguish the file category.

[0165] 3. Whether the first three lines in the file are editable: The header design of different organizations' forms may be different. For some forms, the first few lines of the file may be locked and cannot be edited. However, the design of the first three lines of the same business form of the same organization is basically the same. Therefore, whether the first three lines of the file are editable can be used as a style feature to distinguish file categories.

[0166] 4. Date format in files: The date format of files in the same organization and the same business category is basically the same, such as dd-mm-yyyy|dd / mm / yyyy; yyyy-mm-dd|yyyy / mm / dd. However, the date formats required by files of different organizations may differ. Therefore, the date format type in the file can be used as a feature to identify the file category.

[0167] 5. Sheet page name: The business name may be written directly on the sheet page of some files to be classified. In this way, the business category of the file can be quickly located based on the sheet page name. However, sheet pages are generally used to locate business categories, not organization categories, because sheet pages generally do not contain organization names.

[0168] 6. Is there a business name or organization name in the remarks column? Most documents will have a remarks column. If the remarks column includes the business name or organization name, you can directly locate the corresponding business or organization for quick classification.

[0169] 7. Check if there is any abnormal information in the file, mark it in red and bold: Similar abnormal information may appear in files of the same business category of the same organization. For example, on a payment statement, if an employee has resigned, the information of the resigned employee will be marked in red. You can quickly locate the file category by checking if there is any abnormal information in the file.

[0170] 8. Background color of the first three rows of the sheet: The background colors of the first three rows of different business forms of the same organization may be different, such as yellow and purple. The background colors of the first three rows of forms of different organizations may also be different, but the background colors of the first three rows of the sheet for the same business of the same organization are basically the same. Therefore, the background colors of the first three rows of the sheet can be used as a style feature to distinguish files of different businesses and different organizations.

[0171] 9. Is there an asterisk (*) in the header of the file (required field): For some business requirement forms, required fields will appear in the file, such as whether the applicant is employed or has an ID number. In this case, an asterisk (*) may appear in the header of the file, indicating that the field is required. Otherwise, the file cannot be parsed. Therefore, the asterisk (*) can be used as a feature to identify the file category.

[0172] 10. End of line marker: The end of line marker of the same organization generally does not change. The end of line marker can be used as a feature to identify the file category.

[0173] It's understandable that a decision tree is a tree-like structure. Files to be classified are segmented step by step along the tree's root node (features), into child nodes, ultimately reaching leaf nodes. Leaf nodes represent the classification to which a data point belongs. Information gain is calculated by subtracting the information before and after the file's style features are segmented. When selecting a segmentation attribute, the attribute with the highest information gain is chosen, rather than the attribute with the highest information gain. This approach is more efficient and scientific, avoiding the inefficiency caused by excessive useless features.

[0174] Here are a few definitions:

[0175] Entropy is defined as the average amount of information contained in each message received:

[0176] l(x l )=-log2p(x l );

[0177] Among them, p(x l ) is the probability of selecting the file category, l(x l ) indicates the corresponding p(x l )’s entropy.

[0178] When calculating information entropy, we calculate the expected value of all attributes that may be classified as features under all file categories:

[0179]

[0180] When the probability in entropy is the result of data estimation, it is called empirical entropy. For example, suppose there are 20 data, divided into three categories, class x, class y and class z. Among the 20 data, 10 data belong to class x, then the probability that the data belongs to class x is one-half; among them, 5 data belong to class y, then the probability that the data belongs to class y is one-quarter; the probability that the data belongs to class z is also one-quarter. The probability of belonging is calculated based on the classification feature data provided by the file. Suppose there is a training data set D, the empirical entropy of the training data set D is H(D), |D| represents its sample capacity. Suppose there are M classes K M ,M=1,2,3,···,M,|K M | belongs to class K M The empirical entropy formula can be written as:

[0181]

[0182] Information gain: Information gain is calculated for each feature. The information gain g(D,A) of feature A with respect to training dataset D can be defined as the difference between the empirical entropy H(D) of set D and the empirical conditional entropy H(D|A) of D given the condition of feature A, i.e., g(D,A) = H(D) - H(D|A).

[0183] Information gain rate: It represents the ratio of the node information to the node split information metric. The gain rate is usually used as one of the attribute selection methods. The method used is to calculate the information gain rate based on information gain. The information gain rate formula is as follows: information gain divided by information entropy:

[0184]

[0185] Among them, g r (D, A) represents the information gain rate of feature A to training data set D, g(D, A) represents the information gain of feature A to training data set D; H A (D) represents the information entropy of the training dataset D under the given condition of feature A.

[0186] In an embodiment of the present invention, the C4.5 algorithm is used to generate a decision tree. Compared with the general ID3 decision tree construction algorithm, a pruning operation is performed for optimization during the tree construction process. When selecting the partitioning attributes, the C4.5 algorithm selects file features from the attributes with the largest information gain rate and calculates the information entropy of each file feature separately. Simply put, the smaller the information gain rate of each file feature, the greater the possibility that this file feature can confirm the file type. The file feature with the smallest information gain rate is selected as the root node of the decision tree.

[0187] In another embodiment of the present invention, since there are many file style features of a file, if only one decision tree is constructed to include all the file style features for classification, the constructed decision tree will be very complex and have many layers, which will result in a very low efficiency in distinguishing file types, and a complex and unique decision tree will need to be traversed each time. In the embodiment of the present invention, based on the generation of a graph by multiple decision trees, if each style feature of the file to be classified is compared and screened, and then voted and scored, this will cause the entire decision graph to be traversed when retrieving the file style features of a certain file. This is not only very inefficient, but also causes a waste of system resources. Therefore, the optimized pSCAN algorithm is adopted, and each parent node and leaf node in the decision tree is regarded as a node of the decision graph, and the connection between the parent and child nodes is regarded as the edge of the node in the graph, so that the decision tree is constructed into a graph structure. Next, according to the pSCAN algorithm, the file style features of the file are used as thresholds based on the structural similarity between each node and the minimum number of connected nodes of the node. The entire decision graph is clustered into different clusters, and all nodes in the graph are divided into core nodes, intra-cluster nodes, bridge nodes and outlier nodes. This makes the nodes in the same cluster (same organization or same business) have denser connections, and the nodes between different structures and different businesses have sparser connections, which is convenient for the subsequent screening of the greedy algorithm.

[0188] Based on the clustering results of pSCAN, a greedy algorithm is used to ensure that when classifying files by organization and business (determining file type), the optimal and fastest classification is always made. Specifically, the file style features corresponding to the node with the smallest inter-node structural similarity are first considered, that is, the file style features corresponding to the outlier node. This is because the file style features corresponding to the outlier node contain the least information (the fewest edges connecting to other nodes) and are closer to the facts. In the file to be classified, if the file style features corresponding to the target outlier node are included, all decision trees containing the target outlier node are retained, and all decision trees that do not contain the target outlier node are deleted, avoiding traversal of the decision graph. This process is repeated and the number of decision trees is reduced.

[0189] Preferably, in order to avoid repeated screening of files of the same organization, a HashMap is used to record all nodes of the files that have been fully screened once. If the next file to be classified contains all the nodes stored in the HashMap, it can be used directly to achieve fast classification.

[0190] The following is a specific example to illustrate the actual application process of the business processing method provided by the embodiment of the present invention. Figure 6 As shown, the steps after the parsing verification fails are the same as Figure 5 The steps after the parsing verification fails are the same as Figure 5The steps for parsable verification are the same as those in the previous section and will not be repeated here. The process of the embodiment of the present invention includes: generating a decision tree based on the training sample, generating a decision graph based on the generated multiple decision trees, performing graph clustering on the decision graph, screening the target decision tree based on the greedy algorithm, determining the file type based on the target decision tree and performing parsing verification.

[0191] The process of generating a decision tree is similar to that of generating a decision tree in the above-mentioned embodiment of the invention, and will not be described in detail here, but will be illustrated with a specific example. For example, to generate a decision tree corresponding to the employee transfer details table of the first organization, several file features from the preset feature set are randomly selected, such as whether the form has a drop-down box and whether the form is editable, and the information entropy is calculated for each. Assume that the data of 10 samples is obtained: 4 of the samples are not the employee transfer details table of the first organization, and 6 samples are the employee transfer details table of the first organization. Therefore, the information entropy is calculated as: -(4 / 10×log(4 / 10,2)+6 / 10×log(6 / 10,2))=0.9702.

[0192] If we add another file feature: whether the form is editable and then calculate the information entropy:

[0193] When the form is editable: 1 sample is the employee transfer details list of the first organization, and 5 samples are not the employee transfer details list of the first organization. Information entropy = -(1 / 6×log(1 / 6,2)+5 / 6×log(5 / 6,2)) = 0.65.

[0194] When the form is not editable: 2 samples are employee transfer details of the first organization, and 2 samples are not employee transfer details of the first organization. Information entropy = -(2 / 4×log(2 / 4,2)+2 / 4×log(2 / 4,2)) = 0.5.

[0195] The probability of a form being editable or not is 3 / 10 and 7 / 10 respectively. Therefore, the entropy after division is: 3 / 10×0.65+7 / 10×0.5, which is 0.545. The information gain rate is (0.9702-0.545) / 0.545. Similarly, the information gain rate of whether the form has a drop-down box feature can be calculated. The information gain rate of the editable form here is the largest, so it is selected as the initial basis for division. After selecting editable form as the basis for segmentation features, if the correct classification is achieved, it can be ended. If the classification is not ended, it is necessary to calculate the information gain rate of the remaining features that meet the above classification at the same time, and continue the feature separation operation (calculate the information gain rate of other separation features, use the feature with the largest information gain rate as the basis for separation, and then determine whether the classification is successful). The results are as follows Figure 7As shown in the figure, the first decision tree is successfully generated. Next, we need to calculate the information entropy and information gain rate of other features in the file to be classified. Finally, we can get multiple decision trees separated by multiple features. By combining multiple decision trees together, we can synthesize a decision graph.

[0196] Continuing with the above example, after obtaining the decision graph, the pSCAN algorithm is used to perform clustering operations. First, each leaf node in the decision tree is numbered according to the tree hierarchy.

[0197] Assume a complete decision tree such as Figure 8 As shown, the numbers are numbered in sequence. If there is a second decision tree, the numbers continue to increase and cannot be reset to 1.

[0198] Then, consider the text box as a node. You can see that the neighbors of node 1 are nodes 2 and 3, and node 1 is connected by two lines. The neighbors of node 2 are nodes 1, 4, and 5, and are connected by three lines. The neighbor of node 4 is node 2, and is connected by only one line. So, we can get:

[0199] Node 2 has 4 neighbors (node ​​1, node 2, node 4, node 5), and node 4 has 2 neighbors (node ​​2, node 4). Therefore, the structural similarity between nodes 2 and 4 is the ratio of the number of common neighbors between nodes 2 and 4 to the geometric mean of their neighbors: 2 / √2 × √4 = 0.707. The structural similarity between nodes 2 and 5 is also 2 / √2 × √4 = 0.707. All structural similarities are stored in a HashMap, where the key is {node m, node n} (e.g., {node 2, node 5}), and the value is the structural similarity, for the fastest access.

[0200] Assuming the decision tree is a complete one, the bottom layer of the decision tree can determine the file type. In this embodiment, the threshold is set to 2 because the bottom nodes of the decision graph are connected by only one line, and the bottom leaf nodes cannot be used as core nodes. Therefore, the threshold cannot be set to 1, and the minimum is set to 2. Similarly, the extreme case of structural similarity can be understood as node 2, with only node 4. In this case, the structural similarity is: 2 / √2×√3=0.816. Therefore, the structural similarity cannot be higher than 0.816, so it can be set to 0.7. After the pSCAN algorithm, the clusters are obtained: (node ​​2, node 4, node 5), (node ​​3, node 6, node 7), and node 1 is a bridge node connecting two different clusters.

[0201] By analogy with this method, the structural similarity parameter is set to 0.7, the minimum connection number threshold is set to 2, and the pSCAN clustering operation is performed on the entire decision graph. This will cluster the nodes with dense relationships into a cluster, and obtain core nodes, in-cluster nodes, bridge nodes, and outlier nodes.

[0202] In the embodiment of the present invention, the consumption of computer memory and resources is greatly reduced through efficient algorithms. At the same time, the function of automatically repairing incorrectly classified files also reduces development and maintenance costs, reduces the risks caused by file classification errors, and enables the entire business process to achieve further automated import, classification, verification, repair, and initiation of operations.

[0203] The above describes the service processing method provided by the embodiment of the present invention. The following describes the service processing device provided by the embodiment of the present invention with reference to the accompanying drawings.

[0204] See also Figure 9 , an embodiment of the present invention further provides a service processing device, the device comprising:

[0205] Clustering module 91 is used to cluster the feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into multiple clusters, bridge nodes that do not belong to a cluster and are connected to more than one cluster, and outlier nodes that do not belong to a cluster and are connected to less than two clusters. The nodes in the cluster include core nodes and intra-cluster nodes. The feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files.

[0206] A screening module 92 is configured to screen the decision tree based on the file features of the file to be classified, in the order of outlier nodes, bridge nodes, cluster nodes, and core nodes, to obtain a target decision tree. The first screening is performed from the feature decision graph, and each subsequent screening, except the first screening, is performed from the screening results of the previous screening.

[0207] A determination module 93 is configured to classify the files to be classified based on the target decision tree and determine the file type of the files to be classified;

[0208] The service initiating module 94 is configured to initiate a first service of a corresponding file type when the file to be classified passes the verification policy corresponding to the file type.

[0209] Optionally, there are multiple file features, and the screening module 92 includes:

[0210] A first screening unit is configured to screen decision trees containing target outlier nodes in the feature decision graph to obtain a first decision tree set, wherein the target outlier node is an outlier node having any file feature among the outlier nodes in the feature decision graph;

[0211] a second screening unit, configured to screen the decision trees containing the target bridge node in the first decision tree set to obtain a second decision tree set, wherein the target bridge node is a bridge node having any file feature among the bridge nodes corresponding to the decision trees in the first decision tree set;

[0212] a third screening unit, configured to screen the decision trees in the second decision tree set that contain the target cluster node to obtain a third decision tree set, wherein the target cluster node is the cluster node corresponding to the decision trees in the second decision tree set that has any file feature;

[0213] a fourth screening unit, configured to screen the decision trees containing the target core node in the third decision tree set to obtain a fourth decision tree set, wherein the target core node is a core node having any file feature among the core nodes corresponding to the decision trees in the third decision tree set;

[0214] A determining unit is configured to determine a decision tree in the fourth decision tree set as a target decision tree.

[0215] Optionally, the second screening unit is specifically configured to screen the decision trees in the first decision tree set that include the target bridge node when the number of decision trees in the first decision tree set is greater than or equal to a target threshold, to obtain the second decision tree set;

[0216] a third screening unit, specifically configured to, when the number of decision trees in the second decision tree set is greater than or equal to a target threshold, screen the decision trees in the second decision tree set that contain nodes in the target cluster to obtain a third decision tree set;

[0217] a fourth screening unit, specifically configured to, when the number of decision trees in the third decision tree set is greater than or equal to a target threshold, screen the decision trees in the third decision tree set that contain the target core node, to obtain a fourth decision tree set;

[0218] Accordingly, the device further comprises:

[0219] The set module is used to determine the decision tree in the target set as the target decision tree when the number of decision trees in the target set is less than the target threshold, wherein the target set includes: the first decision tree set, the second decision tree set or the third decision tree set.

[0220] Optionally, the second screening unit is specifically configured to delete nodes corresponding to decision trees that do not include the target outlier node in the feature decision graph to obtain a first intermediate decision graph; and screen the decision trees that include the target bridge node in the first intermediate decision graph to obtain a second decision tree set;

[0221] The third screening unit is specifically configured to delete nodes corresponding to decision trees that do not include the target bridge node in the first intermediate decision graph to obtain a second intermediate decision graph; and to screen decision trees that include nodes in the target cluster in the second intermediate decision graph to obtain a third decision tree set;

[0222] The fourth screening unit is specifically used to delete the nodes corresponding to the decision trees that do not contain the nodes in the target cluster in the second intermediate decision graph to obtain the third intermediate decision graph; and screen the decision trees that contain the target core nodes in the third intermediate decision graph to obtain the fourth decision tree set.

[0223] Optionally, the device further comprises:

[0224] A first repair module is used to obtain multiple business files including the core node of the cluster when the file to be classified fails to pass the verification strategy corresponding to the file type;

[0225] The second repair module is used to compare the file features of the file to be classified with the file features of each business file in the order of outlier nodes, bridge nodes, and nodes within the cluster to determine the score of each business file. For each business file, the larger the score of the business file, the higher the similarity between the business file and the file to be classified;

[0226] A third repair module is used to re-determine the file type of the target business file with the largest score as the file type of the file to be classified;

[0227] The fourth repairing module is configured to initiate a second service corresponding to the re-determined file type when the file to be classified passes the verification policy corresponding to the re-determined file type.

[0228] Optionally, the apparatus further comprises: a fourth repairing module, configured to repair the file features of the file to be classified based on the file feature template of the target business file if the file to be classified fails to pass the verification policy corresponding to the re-determined file type;

[0229] The fifth repairing module is configured to initiate a second service when the repaired file to be classified passes the verification policy corresponding to the re-determined file type.

[0230] Optionally, the device further comprises:

[0231] A first training module is used to determine multiple file types based on files corresponding to different businesses in different business organizations;

[0232] A second training module is used to train a decision tree for each file type using training data, wherein each decision tree includes multiple file features for distinguishing file types;

[0233] The third training module is used to overlap the nodes corresponding to the same file features in different decision trees into the same node to obtain a feature decision graph.

[0234] The service processing device provided by the embodiment of the present invention can realize Figures 1 to 8 To avoid repetition, the various processes of implementing the business processing method in the method embodiment are not described here.

[0235] In an embodiment of the present invention, a feature decision graph formed by connecting multiple decision trees for classifying files is clustered using a graph clustering method, thereby clustering the nodes in the feature decision graph into multiple clusters, bridge nodes, and outlier nodes. The structural similarity between the nodes is used to characterize the information contained in the nodes or the degree of influence on distinguishing file types. The decision trees are then screened in the order of outlier nodes, bridge nodes, cluster nodes, and core nodes to obtain a target decision tree. Since each screening, except for the initial screening, is screened from the screening results of the previous screening, the number of target decision trees obtained by the final screening is relatively small, so that the classification of the files to be classified can be achieved using fewer target decision trees, thereby initiating the corresponding first service. In an embodiment of the present invention, after clustering the feature decision graph, not only is the number of decision trees used to determine the file type of the files to be classified reduced by multiple screenings, thereby improving service processing efficiency, but also, by screening in descending order of influence on distinguishing file types, it is possible to avoid missing decision trees with greater influence, thereby ensuring the accuracy of file classification.

[0236] On the other hand, an embodiment of the present invention further provides an electronic device comprising a memory, a processor, a bus, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above-mentioned business processing method when executing the program.

[0237] Here is an example: Figure 10 A schematic diagram of the physical structure of an electronic device is shown.

[0238] like Figure 10 As shown, the electronic device may include: a processor (processor) 1010, a communication interface (Communications Interface) 1020, a memory (memory) 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute the following method:

[0239] A graph clustering method is used to cluster the feature decision graph, and the nodes in the feature decision graph are clustered into multiple clusters, bridge nodes that do not belong to a cluster and are connected to more than one cluster, and outlier nodes that do not belong to a cluster and are connected to less than two clusters. The nodes in the cluster include core nodes and intra-cluster nodes. The feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files.

[0240] Based on the file features in the files to be classified, the decision tree is screened in the order of outlier nodes, bridge nodes, cluster nodes and core nodes to obtain the target decision tree. The first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening.

[0241] Classify the files to be classified based on the target decision tree and determine the file type of the files to be classified;

[0242] When the file to be classified passes the verification policy corresponding to the file type, a first service corresponding to the file type is initiated.

[0243] In addition, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0244] In another aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for performing the business processing provided in the above embodiments is implemented, for example, including:

[0245] A graph clustering method is used to cluster the feature decision graph, and the nodes in the feature decision graph are clustered into multiple clusters, bridge nodes that do not belong to a cluster and are connected to more than one cluster, and outlier nodes that do not belong to a cluster and are connected to less than two clusters. The nodes in the cluster include core nodes and intra-cluster nodes. The feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files.

[0246] Based on the file features in the files to be classified, the decision tree is screened in the order of outlier nodes, bridge nodes, cluster nodes and core nodes to obtain the target decision tree. The first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening.

[0247] Classify the files to be classified based on the target decision tree and determine the file type of the files to be classified;

[0248] When the file to be classified passes the verification policy corresponding to the file type, a first service corresponding to the file type is initiated.

[0249] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0250] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0251] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A business processing method, characterized in that: The method comprises: Clustering a feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into a plurality of clusters, bridge nodes that do not belong to the cluster and connect to more than one cluster, and outlier nodes that do not belong to the cluster and connect to less than two clusters, wherein the nodes in the cluster include core nodes and intra-cluster nodes, and the feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files; Based on the file features in the file to be classified, the decision tree is screened in the order of the outlier nodes, the bridge nodes, the cluster nodes, and the core nodes to obtain a target decision tree, wherein the first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening; Classifying the files to be classified based on the target decision tree to determine the file types of the files to be classified; Initiating a first service corresponding to the file type when the file to be classified passes the verification policy corresponding to the file type; There are multiple file features, and based on the file features in the file to be classified, the decision tree is screened in the order of the outlier nodes, the bridge nodes, the cluster nodes, and the core nodes to obtain a target decision tree, including: Filtering the decision trees containing the target outlier node in the feature decision graph to obtain a first decision tree set, wherein the target outlier node is an outlier node having any of the file features among the outlier nodes in the feature decision graph; Filtering the decision trees containing the target bridge node in the first decision tree set to obtain a second decision tree set, wherein the target bridge node is a bridge node having any of the file features among the bridge nodes corresponding to the decision trees in the first decision tree set; Filtering the decision trees in the second decision tree set that contain target cluster nodes to obtain a third decision tree set, wherein the target cluster nodes are cluster nodes corresponding to the decision trees in the second decision tree set that have any of the file features; Filtering the decision tree containing the target core node in the third decision tree set to obtain a fourth decision tree set, wherein the target core node is a core node having any of the file features among the core nodes corresponding to the decision trees in the third decision tree set; A decision tree in the fourth decision tree set is determined as a target decision tree.

2. The method according to claim 1, characterized in that The step of screening the decision tree containing the target bridge node in the first decision tree set to obtain the second decision tree set includes: When the number of decision trees in the first decision tree set is greater than or equal to a target threshold, screening the decision trees in the first decision tree set that include the target bridge node to obtain a second decision tree set; The step of screening the second decision tree set for decision trees containing nodes in the target cluster to obtain a third decision tree set includes: When the number of decision trees in the second decision tree set is greater than or equal to the target threshold, screening the decision trees in the second decision tree set that contain nodes in the target cluster to obtain a third decision tree set; The step of screening the decision tree containing the target core node in the third decision tree set to obtain a fourth decision tree set includes: When the number of decision trees in the third decision tree set is greater than or equal to the target threshold, screening the decision trees in the third decision tree set that contain the target core node to obtain a fourth decision tree set; Accordingly, the method further includes: When the number of decision trees in a target set is less than the target threshold, the decision tree in the target set is determined as a target decision tree, wherein the target set includes: the first decision tree set, the second decision tree set, or the third decision tree set.

3. The method according to claim 1, characterized in that The step of screening the decision tree containing the target bridge node in the first decision tree set to obtain the second decision tree set includes: Deleting nodes corresponding to decision trees that do not include the target outlier node in the feature decision graph to obtain a first intermediate decision graph; Filtering the decision tree including the target bridge node in the first intermediate decision graph to obtain the second decision tree set; The step of screening the second decision tree set for decision trees containing nodes in the target cluster to obtain a third decision tree set includes: Deleting nodes corresponding to the decision tree that do not include the target bridge node in the first intermediate decision graph to obtain a second intermediate decision graph; Filtering the decision trees containing the nodes in the target cluster in the second intermediate decision graph to obtain the third decision tree set; The step of screening the decision tree containing the target core node in the third decision tree set to obtain a fourth decision tree set includes: Deleting nodes corresponding to decision trees that do not include nodes in the target cluster in the second intermediate decision graph to obtain a third intermediate decision graph; The decision tree including the target core node is screened in the third intermediate decision graph to obtain the fourth decision tree set.

4. The method according to claim 1, wherein After classifying the to-be-classified files based on the target decision tree and determining the file types of the to-be-classified files, the method further includes: When the file to be classified fails to pass the verification strategy corresponding to the file type, obtaining multiple business files including the core node of the cluster; Comparing the file features of the file to be classified with the file features of each of the business files in the order of outlier nodes, bridge nodes, and nodes within the cluster, and determining a score for each of the business files, wherein for each of the business files, the greater the score of the business file, the higher the similarity between the business file and the file to be classified; Re-determining the file type of the target business file with the largest score as the file type of the file to be classified; In a case where the file to be classified passes the verification policy corresponding to the re-determined file type, a second service corresponding to the re-determined file type is initiated.

5. The method according to claim 4, characterized in that If the file to be classified fails to pass the verification policy corresponding to the re-determined file type, repairing the file features of the file to be classified based on the file feature template of the target business file; In a case where the repaired file to be classified passes the verification policy corresponding to the re-determined file type, the second service is initiated.

6. The method according to claim 1, wherein Before clustering the feature decision graph in a graph clustering manner, clustering the nodes in the feature decision graph into a plurality of clusters, bridge nodes that do not belong to the clusters and are connected to more than one cluster, and outlier nodes that do not belong to the clusters and are connected to less than two clusters, the method further includes: Determine multiple file types based on the files corresponding to different businesses in different business organizations; For each of the file types, a decision tree is obtained by training using the training data, wherein each decision tree includes a plurality of file features for distinguishing the file types; The nodes corresponding to the same file features in different decision trees are overlapped into the same node to obtain the feature decision graph.

7. A business processing device, characterized in that: The device comprises: A clustering module is used to cluster the feature decision graph using a graph clustering method, clustering the nodes in the feature decision graph into multiple clusters, bridge nodes that do not belong to the cluster and connect to more than one cluster, and outlier nodes that do not belong to the cluster and connect to less than two clusters, wherein the nodes in the cluster include core nodes and intra-cluster nodes, and the feature decision graph is a decision graph formed by connecting multiple decision trees for classifying files; A screening module is configured to screen the decision tree to obtain a target decision tree in the order of the outlier nodes, the bridge nodes, the cluster nodes, and the core nodes based on the file features in the file to be classified, wherein the first screening is performed from the feature decision graph, and each screening except the first screening is performed from the screening results of the previous screening; A determination module, configured to classify the to-be-classified files based on the target decision tree and determine the file type of the to-be-classified files; A service initiation module, configured to initiate a first service corresponding to the file type if the file to be classified passes the verification policy corresponding to the file type; The screening module comprises: A first screening unit is configured to screen decision trees containing target outlier nodes in the feature decision graph to obtain a first decision tree set, wherein the target outlier node is an outlier node having any file feature among the outlier nodes in the feature decision graph; a second screening unit, configured to screen the decision trees containing the target bridge node in the first decision tree set to obtain a second decision tree set, wherein the target bridge node is a bridge node having any file feature among the bridge nodes corresponding to the decision trees in the first decision tree set; a third screening unit, configured to screen the decision trees in the second decision tree set that contain the target cluster node to obtain a third decision tree set, wherein the target cluster node is the cluster node corresponding to the decision trees in the second decision tree set that has any file feature; a fourth screening unit, configured to screen the decision trees containing the target core node in the third decision tree set to obtain a fourth decision tree set, wherein the target core node is a core node having any file feature among the core nodes corresponding to the decision trees in the third decision tree set; A determining unit is configured to determine a decision tree in the fourth decision tree set as a target decision tree.

8. An electronic device, characterized in that: include: A processor, a communication interface, a memory, and a communication bus; wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the steps of the business processing method according to any one of claims 1 to 6 when executing the program stored in the memory.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the business processing method according to any one of claims 1 to 6 are implemented.

Images