Group division method and device, equipment, medium and program product

Abstract

The invention discloses a group division method and device, equipment, a medium and a program product, and relates to the technical field of computers. The method comprises the steps of obtaining an entity relation graph; obtaining a first matrix and a second matrix based on an entity association relationship represented in the entity relationship graph and the m candidate groups; performing matrix updating on the preset matrix by taking reduction of the statistical magnitude mean square error as a target to obtain an updated matrix; and determining classification results respectively corresponding to the m candidate groups based on the updated matrix elements. Through the above mode, the association condition between different candidate groups is analyzed through the first matrix, and the problem of analysis limitation caused by neglecting the association between the candidate groups is avoided, so that the accuracy of knowing the classification result through updating the matrix is improved through minimizing the statistical magnitude mean square error corresponding to the statistical magnitude. The method can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic and auxiliary driving.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a group division method, device, equipment, medium and program product. Background Art

[0002] Suppose you want to compare which strategy B is better than the control strategy A. The common approach is to randomly divide the experimental subjects into the experimental group and the control group. The experimental strategy is applied to the subjects in the experimental group, and the control strategy is applied to the subjects in the control group. Finally, the experimental data of all subjects in each group are collected, and the sample means of the data collected from the two groups are compared to perform hypothesis testing to determine which strategy is more effective.

[0003] In the related art, when assigning objects to different groups, a large number of objects are usually randomly divided into two groups equally according to an average distribution method. Objects in one group adopt experimental strategy A to carry out the experimental process, and objects in the other group adopt experimental strategy B to carry out the experimental process, so as to obtain experimental results after the experimental process.

[0004] However, the average distribution method ignores the relationship between objects in different groups. Since the social network is a relatively dense network, it is impossible to divide different groups completely cleanly. Therefore, the groups obtained by dividing using this method have great limitations and are not conducive to obtaining good experimental results. Summary of the invention

[0005] The embodiments of the present application provide a grouping method, apparatus, device, medium and program product, which can analyze the association between different candidate groups through a first matrix, avoid ignoring the association between candidate groups and causing analysis limitations, thereby minimizing the statistical mean square error corresponding to the statistic, and improving the accuracy of the classification result obtained by updating the matrix. The technical solution is as follows.

[0006] In one aspect, a method for grouping is provided, the method comprising:

[0007] Obtain an entity relationship graph; the entity relationship graph is composed of a plurality of entity nodes and entity relationship edges, the entity relationship edges are used to describe the entity association relationship between the entity nodes; wherein the plurality of entity nodes are divided into m candidate groups, where m is an integer greater than 1;

[0008] Based on the entity association relationship represented in the entity relationship graph and the m candidate groups, a first matrix and a second matrix are obtained; wherein the first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; and the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group;

[0009] The preset matrix is ​​updated with the goal of reducing the statistical mean square error to obtain an updated matrix, wherein the statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix; the statistical mean square error is a mean square error expression of the statistic when performing the estimation process, the statistic is a parameter of the estimation process result through the first matrix and the second matrix, and the updated matrix is ​​composed of updated matrix elements;

[0010] Based on the update matrix elements, classification results corresponding to the m candidate groups are determined respectively, and the classification results are used to represent the category groups for classifying the candidate groups, and different category groups are used to execute different processing methods.

[0011] On the other hand, a group division device is provided, the device comprising:

[0012] A graph acquisition module, used to acquire an entity relationship graph; the entity relationship graph is composed of a plurality of entity nodes and entity relationship edges, and the entity relationship edges are used to describe the entity association relationship between the entity nodes; wherein the plurality of entity nodes are divided into m candidate groups, where m is an integer greater than 1;

[0013] A matrix acquisition module, used to obtain a first matrix and a second matrix based on the entity association relationship represented in the entity relationship graph and the m candidate groups; wherein the first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; and the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group;

[0014] An updating module, configured to update the preset matrix with the goal of reducing a statistical mean square error to obtain an updated matrix, wherein the statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix; the statistical mean square error is a mean square error expression of a statistic during a pre-estimation process, the statistic is a parameter of a pre-estimation process result obtained by the first matrix and the second matrix, and the update matrix is ​​composed of update matrix elements;

[0015] The classification module is used to determine the classification results corresponding to the m candidate groups respectively based on the update matrix elements, and the classification results are used to represent the category groups for classifying the candidate groups, and different category groups are used to execute different processing methods.

[0016] On the other hand, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement a group division method as described in any of the above-mentioned embodiments of the present application.

[0017] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction, at least one program, a code set or an instruction set is stored in the storage medium, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the group division method as described in any of the above-mentioned embodiments of the present application.

[0018] On the other hand, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes any of the group division methods described in the above embodiments.

[0019] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0020] Based on the entity association relationship in the entity relationship graph and the m candidate groups after the entity nodes are divided, a first matrix describing the entity association relationship between the candidate groups is obtained, and a second matrix describing the entity association relationship corresponding to the entity nodes in the same candidate group is obtained, and then the preset matrix is ​​updated with the goal of reducing the statistical mean square error, so as to determine the classification results corresponding to the m candidate groups respectively according to the updated matrix elements, so as to achieve the purpose of targeted classification of the candidate groups. In the above process, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between the candidate groups and causing the problem of analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, the preset matrix can be updated in a more targeted manner, which is conducive to improving the accuracy of the classification results known by updating the matrix, and then more accurate experimental data can be obtained with the help of more accurate classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 It is a schematic diagram of an implementation environment provided by an exemplary embodiment of the present application;

[0023] Figure 2 is a flow chart of a group division method provided by an exemplary embodiment of the present application;

[0024] Figure 3 is a flow chart of a group division method provided by another exemplary embodiment of the present application;

[0025] Figure 4 is a schematic diagram of determining a first matrix and a second matrix based on a candidate group provided by an exemplary embodiment of the present application;

[0026] Figure 5 is a flow chart of a group division method provided by another exemplary embodiment of the present application;

[0027] Figure 6 is a flow chart of a group division method provided by another exemplary embodiment of the present application;

[0028] Figure 7 It is an overall schematic diagram of a group division method provided by an exemplary embodiment of the present application;

[0029] Figure 8 is a structural block diagram of a group division device provided by an exemplary embodiment of the present application;

[0030] Fig. 9 It is a structural block diagram of a server provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0032] In the related art, when assigning objects to different groups, the objects are usually randomly divided into two groups according to the average distribution method. The objects in one group adopt experimental strategy A to conduct the experiment, and the objects in the other group adopt experimental strategy B to conduct the experiment, so as to obtain the experimental results after the experiment. However, the average distribution method ignores the relationship between objects in different groups. Since the social network is a relatively dense network, it is impossible to completely divide different groups. Therefore, the groups obtained by dividing in this way have great limitations and are not conducive to obtaining good experimental results.

[0033] In an embodiment of the present application, a group division method is introduced, which can analyze the correlation between different candidate groups through a first matrix, avoid ignoring the correlation between candidate groups and causing analysis limitations, thereby minimizing the statistical mean square error corresponding to the statistic, and improving the accuracy of the classification results learned by updating the matrix. The group division method can be applied to various scenarios such as game data analysis scenarios, environmental data analysis scenarios, medical data analysis scenarios, social data analysis scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc., and the embodiments of the present application are not limited to this.

[0034] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards in the relevant regions. For example, the entity relationship diagrams, candidate categories and other contents involved in this application are all obtained with full authorization.

[0035] Secondly, the implementation environment involved in the embodiment of the present application is described. The group division method provided in the embodiment of the present application can be implemented by the terminal alone, or by the server, or by the terminal and the server through data interaction, which is not limited in the embodiment of the present application. Optionally, the group division method executed by the terminal and the server interactively is used as an example for description.

[0036] For illustration, please refer to Figure 1 The implementation environment involves a terminal 110 and a server 120 , and the terminal 110 and the server 120 are connected via a communication network 130 .

[0037] In some embodiments, the terminal 110 has a relationship graph acquisition function, which is used to acquire an entity relationship graph. The entity relationship graph is composed of multiple entity nodes and entity relationship edges, and the entity relationship edges are used to describe the entity association relationship between entity nodes.

[0038] Optionally, the terminal 110 has a relationship graph construction function, which is used to construct an entity relationship graph according to legally acquired entities and entity association relationships, wherein the entities are entity nodes in the entity relationship graph, and the entity association relationships between entity nodes are represented by entity relationship edges.

[0039] Optionally, the terminal 110 sends the entity relationship diagram to the server 120 via the communication network 130; or, the terminal 110 sends the entities and entity associations used to construct the entity relationship diagram to the server 120 via the communication network 130; thereby, the server 120 obtains the entity relationship diagram.

[0040] The multiple entity nodes are divided into m candidate groups, where m is an integer greater than 1.

[0041] Illustratively, based on a preset number of group divisions, multiple entity nodes in the entity relationship graph are divided into m candidate groups, and each candidate group includes at least one entity node.

[0042] In some embodiments, the server 120 obtains the first matrix and the second matrix based on the entity association relationships represented in the entity relationship graph and the m candidate groups.

[0043] The first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group.

[0044] In some embodiments, the server 120 updates the preset matrix with the goal of reducing the statistical mean square error to obtain an updated matrix. The statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix. That is, the statistical mean square error can not only reflect the difference between the groups and the preset matrix, but also reflect the difference between the group and the preset matrix, thereby improving the comprehensiveness of the expression of the statistical mean square error. The statistical mean square error is the mean square error expression of the statistic when performing the estimation process, and the statistic is a parameter of the estimated processing result through the first matrix and the second matrix; the update matrix is ​​composed of the update matrix elements.

[0045] In some embodiments, the server 120 determines classification results corresponding to the m candidate groups respectively based on the updated matrix elements.

[0046] The classification result is used to characterize the category group for classifying the candidate group, and different category groups are used to perform different processing methods. For example: if the classification result corresponding to the first candidate group indicates that the first candidate group is classified into category 1, the first candidate group is processed and analyzed using the processing method corresponding to category 1; if the classification result corresponding to the second candidate group indicates that the second candidate group is classified into category 2, the second candidate group is processed and analyzed using the processing method corresponding to category 2, etc.

[0047] In some embodiments, the server 120 sends the classification result to the terminal 110 via the communication network 130 , and the terminal 110 performs an analysis and processing process on the entity nodes in the candidate group using a corresponding processing method based on the classification result.

[0048] It is worth noting that the above-mentioned terminals include but are not limited to mobile terminals such as mobile phones, tablet computers, portable laptops, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc., and can also be implemented as desktop computers, etc.; the above-mentioned servers can be independent physical servers, or they can be server clusters or distributed systems composed of multiple physical servers, or they can be cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), as well as big data and artificial intelligence platforms.

[0049] Among them, cloud technology refers to a hosting technology that unifies hardware, application programs, networks and other resources within a wide area network or local area network to achieve data computing, storage, processing and sharing. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool that can be used on demand and is flexible and convenient.

[0050] In some embodiments, the above server can also be implemented as a node in a blockchain system.

[0051] Combined with the above-mentioned noun introduction and application scenarios, the group division method provided by this application is explained, and the application of this method to a server is used as an example for explanation. Figure 2 As shown, the method includes the following steps 210 to 240.

[0052] Step 210, obtaining an entity relationship diagram.

[0053] Among them, the entity relationship graph consists of multiple entity nodes and entity relationship edges.

[0054] Schematically, the entity relationship diagram includes multiple entity nodes, each entity node represents an entity information, and different entity nodes represent different entity information.

[0055] For example: each entity node represents a user object, and different entity nodes represent different user objects; or, each entity node represents an enterprise, and different entity nodes represent different enterprises; or, each entity node represents a piece of knowledge information, and different entity nodes represent different knowledge information, etc.

[0056] Among them, the entity relationship edge is used to describe the entity association relationship between entity nodes.

[0057] Schematically, entity nodes having entity association relationships among a plurality of entity nodes are associated through entity relationship edges, thereby obtaining an entity relationship graph consisting of a plurality of entity nodes and entity relationship edges.

[0058] For example: multiple entity nodes include entity node 1 representing object A, entity node 2 representing object B, and entity node 3 representing object C; if there is an entity association relationship between object A and object B, entity node 1 and entity node 2 are connected through an entity relationship edge; similarly, if there is an entity association relationship between object A and object C, entity node 1 and entity node 3 are connected through an entity relationship edge, etc.

[0059] Optionally, the property of the entity association relationship representation is content determined based on the form of the entity node representation.

[0060] Schematically, the entity node represents the user object, and the entity association relationship is implemented as at least one of the multiple relationships such as friend relationship, interactive relationship, team relationship, etc. between the user objects. For example: multiple user objects such as user object A and user object B are registered objects in a social application, and user object A and user object B are friends in the social application, then it is considered that there is an entity association relationship between user object A and user object B; or, if the number of interactions (such as likes, comments, favorites, forwarding, etc.) between user object A and user object B in the social application reaches a preset number (such as: 1 time, 3 times, 5 times, etc.), then it is considered that there is an entity association relationship between user object A and user object B, etc.

[0061] In some embodiments, the entity relationship graph corresponds to the collection source of the entity nodes and the properties of the entity association relationships.

[0062] Optionally, the entity node is a node representing a user object, and the multiple user objects are a set of registered objects registered in a social application; based on the authorization of the multiple user objects, the multiple user objects are collected, and the friend relationships (an entity association relationship) between the multiple user objects are analyzed to construct the entity relationship graph P1; or, the interactive relationship (another entity association relationship) between the multiple user objects is analyzed to construct the entity relationship graph P2, etc.

[0063] Schematically, when the entity relationship diagram is a relationship diagram constructed based on the use of objects, the entity relationship diagram can be a relationship diagram constructed based on social scenarios (such as social applications, social websites, etc.), or a relationship diagram constructed based on game scenarios (such as online games, cloud games, etc.), etc.

[0064] Illustratively, when the entity relationship diagram is a relationship diagram constructed based on content such as enterprises and knowledge information, the entity relationship diagram can be a relationship diagram constructed based on the medical field, or a relationship diagram constructed based on the transportation field, etc., and the embodiments of the present application do not limit this.

[0065] The multiple entity nodes are divided into m candidate groups, where m is an integer greater than 1.

[0066] Schematically, multiple nodes in the entity relationship diagram are pre-divided into m candidate groups.

[0067] Optionally, m is a pre-selected division value, and the m candidate groups can be used to analyze the relationship between multiple entity nodes in the form of groups.

[0068] In some embodiments, after obtaining a plurality of entity nodes, the plurality of entity nodes are randomly divided into m candidate groups, each candidate group including at least one entity node.

[0069] In some embodiments, after obtaining an entity relationship graph consisting of multiple entity nodes, the multiple entity nodes are divided into m candidate groups according to the arrangement of the multiple entity nodes in the entity relationship graph, and each candidate group includes at least one entity node.

[0070] Schematically, three entity nodes on the periphery of the entity relationship graph are divided into one candidate group; five entity nodes in the center of the entity relationship graph are divided into another candidate group, and so on, thereby obtaining m candidate groups after grouping multiple entity nodes.

[0071] In some embodiments, one entity node corresponds to one candidate group. Illustratively, if entity node 1 is classified into candidate group A, entity node 1 is no longer classified into candidate group B, so that there is no intersection between the m candidate groups.

[0072] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0073] Step 220 , obtaining a first matrix and a second matrix based on the entity association relationships represented in the entity relationship graph and the m candidate groups.

[0074] Illustratively, after obtaining m candidate groups, based on the entity association relationships represented by the entity relationship graph, the entity association relationships between entity nodes within the m candidate groups are analyzed, and the entity association relationships between the m candidate groups are analyzed, etc.

[0075] In some embodiments, the entity association relationships among the m candidate groups are analyzed, and the analysis results among the m candidate groups are integrated to obtain a first matrix.

[0076] Illustratively, the first matrix includes a plurality of first matrix elements, that is, the first matrix is ​​composed of a plurality of first matrix elements, wherein the first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups.

[0077] Optionally, each first matrix element represents a group association status between two of the m candidate groups, and the group association status is determined by an entity association relationship.

[0078] For example, m candidate groups include candidate group m1, candidate group m2, and candidate group m3. By analyzing the entity association relationship between the entity nodes in candidate group m1 and the entity nodes in candidate group m2, the group association between candidate group m1 and candidate group m2 is obtained to obtain the first matrix element N 12 Similarly, by analyzing the entity association relationship between the entity nodes in the candidate group m1 and the entity nodes in the candidate group m3, the group association between the candidate group m1 and the candidate group m2 is obtained to obtain the first matrix element N 13 Similarly, by analyzing the entity association relationship between the entity nodes in the candidate group m2 and the entity nodes in the candidate group m3, the group association between the candidate group m1 and the candidate group m2 is obtained to obtain the first matrix element N 23 wait.

[0079] Optionally, the first matrix generated based on the m candidate groups is an m*m matrix, that is, the number of matrix rows is m and the number of matrix columns is also m.

[0080] In some embodiments, the entity association relationships between the entity nodes in the m candidate groups are analyzed, and the analysis results corresponding to the m candidate groups are integrated to obtain a second matrix.

[0081] Illustratively, the second matrix includes a plurality of second matrix elements, that is, the second matrix is ​​composed of a plurality of second matrix elements, wherein the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group.

[0082] Optionally, each second matrix element corresponds to a candidate group, and the second matrix element is determined based on the entity association relationship corresponding to the entity node in the candidate group; m second matrix elements are determined based on m candidate groups.

[0083] For example, the m candidate groups include candidate group m1, candidate group m2, and candidate group m3. The second matrix element D corresponding to the candidate group m1 is determined by the entity association relationship corresponding to the entity node in the candidate group m1. 1 Similarly, the second matrix element D corresponding to the candidate group m2 is determined through the entity association relationship corresponding to the entity node in the candidate group m2. 2 Similarly, the second matrix element D corresponding to the candidate group m3 is determined through the entity association relationship corresponding to the entity node in the candidate group m3. 3 wait.

[0084] Optionally, the second matrix generated based on the m candidate groups is an m*1 matrix, that is, the number of matrix rows is m, the number of matrix columns is 1, and so on.

[0085] Step 230 , updating the preset matrix with the goal of reducing the statistical mean square error to obtain an updated matrix.

[0086] Illustratively, the preset matrix is ​​a pre-set matrix, for example, the preset matrix is ​​a unit matrix, etc.

[0087] The statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix. That is, when determining the statistical mean square error, not only the loss between the first matrix representing the entity association relationship between groups and the preset matrix is ​​considered, but also the loss between the second matrix representing the entity association relationship within the group and the preset matrix is ​​considered.

[0088] Optionally, when determining the first error based on the first matrix and the preset matrix, a matrix operation is performed on the first matrix and the preset matrix using a first error function to obtain the first error.

[0089] Illustratively, the first error is obtained based on the sum of matrix elements in the matrix product between the first matrix and the preset matrix; or, the first error is obtained based on the sum of diagonal matrix elements in the matrix product between the first matrix and the preset matrix, etc.

[0090] Optionally, when determining the second error based on the second matrix and the preset matrix, a matrix operation is performed on the second matrix and the preset matrix using a second error function to obtain the second error.

[0091] Illustratively, the second error is obtained based on the sum of matrix elements in the matrix product between the second matrix and the preset matrix; or, the second error is obtained based on the sum of diagonal matrix elements in the matrix product between the second matrix and the preset matrix, etc.

[0092] In some embodiments, the sum of the first error and the second error is taken as the statistical mean square error.

[0093] It is worth noting that the above process of obtaining the first error through the first error function and obtaining the second error through the second error function is only an illustrative example, and the transformation of the formula is within the protection scope and is not limited here.

[0094] In some embodiments, after the first matrix, the second matrix and the preset matrix are obtained, the statistical mean square error is calculated by using the first matrix, the second matrix and the preset matrix.

[0095] The statistical mean square error is an expression of the mean square error of the statistic when performing the estimation process, and the statistic is a parameter of the estimation process result through the first matrix and the second matrix.

[0096] Indicatively, statistics, also known as estimators, are a term in statistics, which refers to the value calculated when estimating parameters. Among them, the process of estimating parameters involves parameter estimation in statistics. Statistics can be roughly divided into descriptive statistics and inferential statistics. Parameter estimation is one of the important contents of inferential statistics.

[0097] Descriptive statistics refers to various activities that describe data characteristics using summary data methods such as tabulation, classification, graphics, and calculations; inferential statistics is based on descriptive statistics, and summarizes regularities based on sample data (such as the entity relationship diagram mentioned above), so as to infer and predict the population (such as the multiple entity nodes mentioned above) in order to better understand the population.

[0098] In some embodiments, when statistics are obtained through parameter estimation, it is hoped that a more global grasp of multiple entity nodes can be obtained through a more accurate statistics acquisition process, thereby facilitating subsequent analysis of the multiple entity nodes with the help of more accurate statistics.

[0099] Schematically, parameter estimation represents estimating a population using a statistic (estimator). For example, estimating a population mean using a sample mean, where the sample mean represents a statistic used to estimate a population mean, and the population mean represents the mean of a large number of samples (such as multiple entity nodes); or estimating a population variance using a sample variance, where the sample variance represents a statistic used to estimate a population variance, and the population variance represents the variance of a large number of samples (such as multiple entity nodes), etc.

[0100] Optionally, the above population mean, population variance and other parameters are generally referred to as population parameters θ, and the statistics used to estimate the population parameters θ are referred to as The purpose of parameter estimation is to determine the statistic more accurately. So that we can use statistics A more accurate statistical process is performed on the population parameter θ that characterizes the population.

[0101] In some embodiments, a preset function is obtained, in which function variables include a first variable indicating a first matrix, a second variable indicating a second matrix, and a third variable indicating a preset matrix; the first matrix, the second matrix, and the preset matrix are substituted into the preset function to calculate the statistical mean square error.

[0102] Optionally, the parameter estimated by the statistic is called the population parameter, and the mean square error of the statistic is determined based on the statistic and the population parameter. Therefore, the smaller the mean square error of the statistic, the better the estimation effect of the statistic. The statistic can be used to express multiple entity nodes more comprehensively and balanced from multiple angles (the population parameter represents multiple parameters).

[0103] Illustratively, based on the above characteristics of the statistical mean square error, with the goal of reducing the statistical mean square error, on the premise that the first matrix and the second matrix are known, the preset matrix is ​​updated to obtain an updated matrix.

[0104] Optionally, with the goal of reducing a statistical mean square error, the preset matrix is ​​updated at least once using the first matrix and the second matrix to obtain an updated matrix.

[0105] Illustratively, the number of updates is preselected to reduce the statistical mean square error, the preset matrix is ​​updated multiple times through the first matrix and the second matrix, and after the preset matrix is ​​updated the number of times reaching the update number, an updated matrix is ​​obtained.

[0106] Among them, the update matrix is ​​composed of update matrix elements.

[0107] Schematically, the update matrix includes a plurality of matrix elements, which are referred to as update matrix elements, that is, the update matrix is ​​composed of a plurality of update matrix elements.

[0108] The updated matrix is ​​a matrix obtained by updating the first matrix and the second matrix on the basis of the preset matrix, and can more comprehensively display the entity association relationship within the candidate groups represented by the first matrix and the second matrix and between the candidate groups.

[0109] Step 240: Determine the classification results corresponding to the m candidate groups respectively based on the updated matrix elements.

[0110] Illustratively, after the update matrix elements are obtained, the classification results corresponding to the m candidate groups are determined according to the values ​​of the update matrix elements.

[0111] In some embodiments, a matrix transformation is performed on the update matrix, so as to obtain matrix elements corresponding to the m candidate groups respectively based on multiple update matrix elements in the update matrix, that is, to obtain m matrix elements, and then determine the one-to-one correspondence between the m matrix elements and the m candidate groups according to the positions of the matrix elements, and then determine the classification results corresponding to the m candidate groups respectively according to the values ​​of the matrix elements corresponding to the candidate groups.

[0112] The classification result is used to represent the category group into which the candidate group is classified.

[0113] Schematically, the category group is a pre-set group content used to classify the candidate groups, and different category groups are used to perform different processing methods; therefore, based on the classification result, it is possible to determine the category group into which the candidate group needs to be classified, and then use the processing method corresponding to the category group to analyze and process the candidate group classified into the category group.

[0114] Optionally, the pre-set category groups include experimental groups and control groups. The experimental groups are used to perform data analysis and processing in the set experimental form; the control groups are used not to perform the data analysis and processing, or to perform the data analysis and processing in a form different from the set experimental form, etc.

[0115] Schematically, the classification result corresponding to the candidate group m1 is the experimental group, which means that the candidate group m1 is classified into the experimental group, so that the entity nodes in the candidate group m1 are subjected to data analysis and processing through the set experimental form; or, the classification result corresponding to the candidate group m2 is the control group, which means that the candidate group m1 is classified into the control group, so that the entity nodes in the candidate group m2 are subjected to data analysis and processing through other experimental forms, etc.

[0116] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0117] In summary, based on the entity association relationship in the entity relationship graph and the m candidate groups after the entity nodes are divided, a first matrix describing the entity association relationship between the candidate groups is obtained, and a second matrix describing the entity association relationship corresponding to the entity nodes in the same candidate group is obtained, and then the preset matrix is ​​updated with the goal of reducing the statistical mean square error, so as to determine the classification results corresponding to the m candidate groups respectively according to the updated matrix elements therein, so as to achieve the purpose of targeted classification of the candidate groups. In the above process, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between the candidate groups and causing the problem of analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, the preset matrix can be updated in a more targeted manner, which is conducive to improving the accuracy of the classification results known by the updated matrix, and then more accurate experimental data can be obtained with the help of more accurate classification results.

[0118] In an optional embodiment, a first matrix is ​​generated based on the entity association relationships between candidate groups; and a second matrix is ​​generated based on the entity association relationships within the m candidate groups. Figure 3 As shown above Figure 2 The illustrated step 220 may also be implemented as the following steps 310 to 320 .

[0119] Step 310: Generate a first matrix based on the entity association relationships among the m candidate groups.

[0120] Wherein, m is an integer greater than 1, and the m candidate groups are obtained by dividing the multiple entity nodes.

[0121] Optionally, the entity association relationship between the m candidate groups is determined according to the entity association relationship between the entity nodes corresponding to different candidate groups, that is, the entity association relationship between the m candidate groups is used to refer to the ability to establish an association relationship between the candidate groups.

[0122] In an optional embodiment, the m candidate groups are divided into pairs to obtain multiple group division results.

[0123] In an illustrative manner, after obtaining m candidate groups, the m candidate groups are divided in pairs to obtain a plurality of group division results, each of which includes two candidate groups.

[0124] like Figure 4As shown, it is a schematic entity relationship diagram, wherein the multiple entity nodes include entity nodes 1 to 10; the multiple entity nodes are divided into four candidate groups, namely candidate group 410, candidate group 420, candidate group 430 and candidate group 440. Candidate group 410 and candidate group 420 are combined into a group division result M1; candidate group 410 and candidate group 430 are combined into a group division result M2; candidate group 410 and candidate group 440 are combined into a group division result M3; candidate group 420 and candidate group 430 are combined into a group division result M4, etc.

[0125] In an optional embodiment, the number of entity relationship edges between two candidate groups in each group division result is determined to obtain the number of relationship edges corresponding to the multiple group division results.

[0126] Illustratively, after obtaining multiple grouping results, each grouping result is analyzed separately to determine two candidate groups in one grouping result, and determine the number of entity relationship edges between the two candidate groups, which corresponds to the grouping result.

[0127] The entity relationship edge is used to connect two entity nodes having an entity association relationship, so the entity relationship edge between two candidate groups is used to represent that the entity relationship edge spans the two candidate groups.

[0128] like Figure 4 As shown, the group division result M1 composed of candidate group 410 and candidate group 420 is taken as an example for explanation. For candidate group 410 and candidate group 420 in group division result M1, it is determined that the entity relationship edge between the two candidate groups includes: the entity relationship edge connecting entity node 1 and entity node 3, and the entity relationship edge connecting entity node 2 and entity node 4; based on this, it is determined that the number of entity relationship edges between candidate group 410 and candidate group 420 is 2, and 2 is used as the number of relationship edges corresponding to group division result M1; similarly, for candidate group 420 and candidate group 430 in group division result M4, it is determined that the entity relationship edge between the two candidate groups is: the entity relationship edge connecting entity node 4 and entity node 6; based on this, it is determined that the number of entity relationship edges between candidate group 420 and candidate group 430 is 1, and 1 is used as the number of relationship edges corresponding to group division result M4, etc.

[0129] Based on the above process, multiple group division results are analyzed respectively to determine the number of relationship edges corresponding to the multiple group division results, which will not be elaborated here.

[0130] In an optional embodiment, a first matrix is ​​generated with a plurality of relationship edge numbers as first matrix elements.

[0131] Optionally, when constructing the first matrix, the number of candidate groups m is used as the number of matrix columns and matrix rows of the first matrix, so as to construct an m*m first matrix, and the first matrix is ​​composed of first matrix elements, and the first matrix elements are determined by filling in the number of relationship edges of the group division results obtained by combining the candidate groups pairwise.

[0132] Illustratively, based on the existence of first matrix elements consisting of the same candidate group in the first matrix, such as the first matrix element representing the number of relationship edges between candidate group m1 and candidate group m2, when filling the first matrix element, the first matrix element is set to 0; similarly, other first matrix elements representing the same candidate group are filled with 0.

[0133] Indicatively, for m candidate groups, an m*m first matrix is ​​constructed, and C represents the first matrix, and C represents ij represents the first matrix element constituting the first matrix, where i is any candidate group; j is also any candidate group. When the first matrix element is C ii , then C ii =0; then the first matrix element C ij The first matrix C is determined by the number of relationship edges between candidate group i and candidate group j, and is expressed as the following formula 1.

[0134] Formula 1:

[0135]

[0136] Among them, c 11 The first matrix element corresponding to the group division result consisting of candidate group 1 and candidate group 1 is represented based on c 11 is the first matrix element of the same candidate group, so c 11 Fill with value 0; c 12 The first matrix element corresponding to the group division result consisting of candidate group 1 and candidate group 2 is represented by the number of relationship edges corresponding to the group division result. 12 Filling; c 1m The first matrix element corresponding to the group division result consisting of candidate group 1 and candidate group m is represented by the number of relationship edges corresponding to the group division result. 1m Filling, etc.

[0137] Among them, the first matrix element c 21 and the first matrix element c 12 For example, based on the first matrix element c 21 and the first matrix element c 12Both represent the same group division result composed of candidate group 1 and candidate group 2, so the number of relationship edges corresponding to the group division result is used to fill the first matrix element c 21 and the first matrix element c 12 ; Similarly, the first matrix element c 13 and the first matrix element c 31 The same number of relationship edges is also used for filling, etc., which is not limited here.

[0138] Indicatively, for example Figure 4 As shown in the entity relationship diagram, the number of relationship edges of the group division result M1 composed of the candidate group 410 and the candidate group 420 is determined to be 2; the number of relationship edges of the group division result M2 composed of the candidate group 410 and the candidate group 430 is determined to be 1; the number of relationship edges of the group division result M3 composed of the candidate group 410 and the candidate group 440 is determined to be 1; the number of relationship edges of the group division result M4 composed of the candidate group 420 and the candidate group 430 is determined to be 1; the number of relationship edges of the group division result M5 composed of the candidate group 420 and the candidate group 440 is determined to be 2; the number of relationship edges of the group division result M6 composed of the candidate group 430 and the candidate group 440 is determined to be 1.

[0139] Based on Figure 4 The number of candidate groups shown is 4, so the number of matrix rows and matrix columns of the first matrix are both 4; based on the above Figure 4 The determined numbers of multiple relationship edges are used to numerically fill in the above formula 1 with the numbers of multiple relationship edges, thereby obtaining the following formula 2 after numerical filling.

[0140] Formula 2:

[0141]

[0142] Formula 2 is based on Figure 4 The entity relationship diagram shown in the figure obtains a first matrix, and the number of matrix rows and matrix columns of the first matrix are the number of candidate groups.

[0143] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0144] Step 320 : generating a second matrix based on at least one entity node respectively included in the m candidate groups and the entity association relationship.

[0145] Wherein, m is an integer greater than 1, and the m candidate groups are obtained by dividing the multiple entity nodes.

[0146] Optionally, each of the m candidate groups obtained by division is analyzed separately, and analysis results corresponding to the m candidate groups are obtained according to the number of entity nodes included therein and the number of entity nodes that have entity association relationships with the entity nodes, and then the second matrix is ​​generated by integrating the m analysis results.

[0147] In an optional embodiment, for a plurality of entity nodes in the entity relationship graph, the number of entity nodes having an entity association relationship with each entity node is determined to obtain the number of associated nodes corresponding to the plurality of entity nodes.

[0148] Schematically, multiple entity nodes in the entity relationship diagram are respectively taken as an analysis object, and the multiple entity nodes are analyzed separately.

[0149] For any entity node, other entity nodes that have entity association relationships with the entity node are determined, and the number of nodes of the other entity nodes is determined as the number of associated nodes corresponding to the entity node.

[0150] Based on the fact that the process can be performed through an entity relationship graph, when determining other entity nodes that have an entity association relationship with the entity node, determining other entity nodes connected to the entity node through entity relationship edges, and determining the number of other entity nodes as the number of association nodes corresponding to the entity node;

[0151] Alternatively, when determining other entity nodes that have entity association relationships with the entity node, the number of entity relationship edges on the entity node is determined, that is, the number of associated nodes corresponding to the entity node can be determined.

[0152] like Figure 4 As shown, for entity node 1 in the entity relationship graph, there are 2 entity nodes that have entity association relationships with entity node 1, namely entity node 2 and entity node 3, so the number of associated nodes corresponding to entity node 1 is 2; similarly, for entity node 2 in the entity relationship graph, there are 4 entity nodes that have entity association relationships with entity node 2, namely entity node 1, entity node 4, entity node 6 and entity node 10, so the number of associated nodes corresponding to entity node 2 is 4, and so on.

[0153] Based on the above process, the number of associated nodes corresponding to multiple entity nodes is determined respectively, and the acquisition process is not repeated here; the final result is: the number of associated nodes of entity node 1 is 2; the number of associated nodes of entity node 2 is 4; the number of associated nodes of entity node 3 is 3; the number of associated nodes of entity node 4 is 5; the number of associated nodes of entity node 5 is 3; the number of associated nodes of entity node 6 is 4; the number of associated nodes of entity node 7 is 1; the number of associated nodes of entity node 8 is 2; the number of associated nodes of entity node 9 is 3; the number of associated nodes of entity node 10 is 3.

[0154] In an optional embodiment, within the same candidate group, the sum of the numbers of associated nodes corresponding to the entity nodes is determined to obtain the sum of the numbers of nodes corresponding to the m candidate groups.

[0155] Illustratively, after knowing the number of associated nodes corresponding to the multiple entity nodes, the multiple entity nodes are divided into m candidate groups, so that the number of nodes corresponding to the entity nodes in the same candidate group can be determined.

[0156] like Figure 4 As shown, for candidate group 410, it includes entity node 1 and entity node 2; based on the number of associated nodes of entity node 1 being 2 and the number of associated nodes of entity node 2 being 4, the numbers of associated nodes corresponding to the two entity nodes are summed up, and the sum of the number of nodes corresponding to candidate group 410 is 6; similarly, for candidate group 420, it includes entity node 3, entity node 4 and entity node 5; based on the number of associated nodes of entity node 3 being 3, the number of associated nodes of entity node 4 being 5, and the number of associated nodes of entity node 5 being 3, the numbers of associated nodes corresponding to the three entity nodes are summed up, and the sum of the number of nodes corresponding to candidate group 420 is 11, and so on.

[0157] Based on the above process, the sum of the number of nodes corresponding to multiple candidate groups is determined, and the acquisition process is not repeated here; the final result is: the sum of the number of nodes corresponding to candidate group 410 is 6; the sum of the number of nodes corresponding to candidate group 420 is 11; the sum of the number of nodes corresponding to candidate group 430 is 7; the sum of the number of nodes corresponding to candidate group 440 is 6.

[0158] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0159] In an optional embodiment, the second matrix is ​​generated with the sum of multiple node numbers as the second matrix elements.

[0160] Optionally, when constructing the second matrix, the number of candidate groups m is taken as the number of matrix rows of the second matrix, and the number of matrix columns of the second matrix is ​​set to 1, so as to construct an m*1 second matrix, and the second matrix is ​​composed of second matrix elements, and the second matrix elements are filled and determined by the sum of the number of nodes corresponding to the entity nodes in the same candidate group.

[0161] Indicatively, for m candidate groups, an m*1 second matrix is ​​constructed, and D represents the second matrix. i represents the second matrix elements constituting the second matrix, where i is any candidate group; then the second matrix element D i The second matrix D is determined by the sum of the number of nodes corresponding to the candidate group i and is expressed as the following formula three.

[0162] Formula 3:

[0163] D=(D 1 ,D 2 ,…,D m ) T

[0164] Where T represents the transposition relationship; (D 1 ,D 2 ,…,D m ) is a 1*m matrix; based on the transposition relationship, determine that D is an m*1 matrix; D 1 represents the second matrix element corresponding to candidate group 1, which is filled by the sum of the number of nodes corresponding to candidate group 1; D 2 represents the second matrix element corresponding to candidate group 2, which is filled by the sum of the number of nodes corresponding to candidate group 2; D m The second matrix element corresponding to the candidate group m is represented and filled by the sum of the number of nodes corresponding to the candidate group m.

[0165] Indicatively, for example Figure 4 According to the entity relationship diagram shown, the sum of the number of nodes corresponding to candidate group 410 is determined to be 6; the sum of the number of nodes corresponding to candidate group 420 is determined to be 11; the sum of the number of nodes corresponding to candidate group 430 is determined to be 7; and the sum of the number of nodes corresponding to candidate group 440 is determined to be 6.

[0166] Based on Figure 4 The number of candidate groups shown is 4, so the number of rows of the second matrix is ​​m, and the number of columns is 1; based on the above Figure 4 The determined 4 node numbers and the 4 relationship edge numbers are used to fill in the above formula 3, thereby obtaining the following formula 4 after the numerical filling.

[0167] Formula 4:

[0168]

[0169] Formula 4 is based on Figure 4 The second matrix is ​​obtained by obtaining the entity relationship diagram shown.

[0170] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0171] In summary, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between candidate groups and causing analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, a more targeted matrix update process can be performed on the preset matrix, which is conducive to improving the accuracy of the classification results known by updating the matrix, and then using more accurate classification results to obtain more precise experimental data.

[0172] In an embodiment of the present application, a method for generating a first matrix and a second matrix is ​​introduced. By dividing the m candidate groups into two groups, the number of relationship edges of the entity relationship edges between two candidate groups is determined, so that the first matrix is ​​generated with the number of relationship edges as the first matrix element, which fully characterizes the group association between different candidate groups; in addition, the number of associated nodes of the entity nodes that have an entity association relationship with each entity node is determined, so that according to the sum of the number of associated nodes of the entity nodes in the same candidate group, the sum of the number of nodes corresponding to the m candidate groups is obtained, and the second matrix is ​​generated with the sum of the number of nodes as the second matrix element, which fully characterizes the entity association within the candidate group. Avoiding the limitation of only analyzing multiple candidate groups separately, using the first matrix as a reference for subsequent analysis is conducive to improving the accuracy of obtaining the subsequent update matrix.

[0173] In an optional embodiment, when updating the preset matrix to obtain the updated matrix, a pre-function for updating the matrix through the first matrix, the second matrix and the preset matrix is ​​obtained to perform the updating process. Figure 5 As shown above Figure 2 The illustrated step 230 may also be implemented as the following steps 510 to 530 .

[0174] Step 510, obtaining a preset function.

[0175] The preset function is used to perform matrix update through the first matrix, the second matrix and the preset matrix.

[0176] Illustratively, the preset function is a pre-set function, and the function variables in the preset function include a first variable indicating a first matrix, a second variable indicating a second matrix, and a third variable indicating a preset matrix.

[0177] Step 520: Perform matrix operations on the first matrix, the second matrix, and the preset matrix using a preset function to obtain a statistical mean square error.

[0178] In an illustrative manner, based on the first matrix, the second matrix and the preset matrix, the first matrix, the second matrix and the preset matrix are comprehensively analyzed by the obtained preset function. For example, the first matrix, the second matrix and the preset matrix are substituted into the preset function for the solution process to obtain the statistical mean square error.

[0179] Schematically, the following formula 5 is a preset function expression.

[0180] Formula 5:

[0181] M(R)=B(X(R)) 2 +8(ω 2 +4)V(X(R))

[0182] Among them, M(R) represents the mean square error of the calculated statistic; R represents the statistic (or estimator), which is the content obtained based on the preset matrix update, so the statistics in the initial stage or the intermediate stage are still in the updating process; ω is a preset parameter, such as 1, 0.9 and other values; X(R) is the result obtained after an operation process (such as through the first operation function) is performed on the statistic R; B(X) is the result obtained after another operation process (such as through the second operation function) is performed on X(R); V(X) is the result obtained after another operation process (such as through the third operation function) is performed on X(R).

[0183] In an optional embodiment, the preset function includes a first error function and a second error function, the first error function is used to obtain a first error constituting a mean square error of a statistic based on a first matrix and a preset matrix; the second error function is used to obtain a second error constituting a mean square error of a statistic based on a second matrix and a preset matrix.

[0184] Optionally, the first error function is implemented as B(X(R)) 2 , the second error function is realized as 8(ω 2 +4)V(X(R)); or, the first error function is realized as 8(ω 2 +4)V(X(R)), the second error function is implemented as B(X(R)) 2 . Then the preset function is the sum of the first error function and the second error function.

[0185] Schematically, if the first error function is expressed as B(X(R)) 2, it means that the first error function is a function based on the joint representation of the first operation function and the second operation function; if the second error function is expressed as 8(ω 2 +4)V(X(R)), then it means that the second error function is a function based on the joint representation of the above-mentioned first operation function and the third operation function, etc.

[0186] It is worth noting that the division process of the above preset function into the first error function and the second error function is only an illustrative example. The formula transformation contents represented by the preset function, the first error function, the second error function and other functions are all contents protected by the embodiments of the present application and are not limited here.

[0187] In some embodiments, the first operation function represents a processing function of a preset matrix, and the second operation function represents a processing function between the first operation function and the first matrix, then the first error function represents the error relationship between the preset matrix and the first matrix; the second operation function represents a processing function between the first operation function and the second matrix, then the second error function represents the error relationship between the preset matrix and the second matrix.

[0188] Illustratively, the first operation function, the second operation function and the third operation function are explained in the following process.

[0189] In an optional embodiment, a preset matrix and a first transposed matrix of the preset matrix are obtained.

[0190] Illustratively, the preset matrix is ​​a pre-set matrix.

[0191] In some embodiments, the preset matrix is ​​a pre-set unit matrix, which is a square matrix, that is, the number of matrix rows and matrix columns of the unit matrix are the same, which is the number m of candidate groups; in addition, based on the properties of the unit matrix, the diagonal of the unit matrix is ​​determined to be 1, and the other matrix elements are 0.

[0192] The preset matrix is ​​transposed to obtain a first transposed matrix corresponding to the preset matrix.

[0193] Optionally, if the preset matrix is ​​a unit matrix in which both the number of matrix rows and the number of matrix columns are m, after the unit matrix is ​​transposed, the first transposed matrix is ​​the unit matrix.

[0194] In some embodiments, the preset matrix is ​​an intermediate matrix updated based on the unit matrix, and the unit matrix is ​​a square matrix with m matrix rows and m matrix columns. The preset matrix is ​​an intermediate matrix updated based on the unit matrix, and its number of matrix rows and m matrix columns are also m, but the matrix elements therein are intermediate contents obtained based on the update.

[0195] The preset matrix is ​​transposed to obtain a first transposed matrix corresponding to the preset matrix.

[0196] Optionally, if the preset matrix is ​​an intermediate matrix, after the intermediate matrix is ​​transposed, the rows and columns of the matrix elements in the first transposed matrix are swapped, so that the matrix elements therein may change.

[0197] In an optional embodiment, an arcsine process is performed on the matrix product between the preset matrix and the first transposed matrix to obtain a first intermediate matrix.

[0198] Illustratively, after obtaining the preset matrix and the corresponding first transposed matrix, matrix multiplication is performed on the preset matrix and the first transposed matrix to obtain a matrix product, and then an inverse sine transform is performed on the matrix product to obtain a first intermediate matrix.

[0199] Illustratively, the process of obtaining the first intermediate matrix through the preset matrix and the first transposed matrix is ​​shown in the following formula 6, where formula 6 is the above-mentioned first operation function.

[0200] Formula 6:

[0201]

[0202] Wherein, X(R) is used to represent the first intermediate matrix obtained by the first operation function in the process of obtaining the statistical mean square error; R is a preset matrix (the unit matrix in the initial stage or the intermediate matrix in the intermediate stage); R T is the first transposed matrix corresponding to the preset matrix; arcsin is the inverse sine function. That is, the first operation function is a function that uses the preset matrix as an independent variable to obtain the corresponding dependent variable, and the value of the dependent variable X(R) can be called the first dependent variable.

[0203] Schematically, for the preset matrix R and the first transposed matrix R T Perform matrix product to get matrix RR T ; For the matrix RR T When performing the arcsine transform, the matrix RR T Perform an inverse sine transformation on each matrix element in to obtain the matrix arcsin(RR T ); then use the matrix arcsin(RR T ) is the dividend, and 2π is the divisor. For the matrix arcsin(RR T ) is divided by each matrix element in the formula to realize the first intermediate matrix X(R) obtained by the first operation function operation shown in formula six.

[0204] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0205] In an optional embodiment, matrix product is performed on the first matrix and the first intermediate matrix to obtain the second intermediate matrix.

[0206] Illustratively, based on the preset matrix and the first transposed matrix being m*m matrices, the first intermediate matrix obtained is also an m*m matrix; the first matrix, as a matrix determined by m candidate groups, is also an m*m matrix; the second intermediate matrix obtained by matrix multiplication of the first matrix and the first intermediate matrix is ​​also an m*m matrix.

[0207] In an optional embodiment, a sum operation is performed on matrix elements on the matrix diagonal corresponding to the second intermediate matrix to obtain the sum of the first elements.

[0208] Illustratively, the second intermediate matrix is ​​an m*m matrix, a matrix diagonal corresponding to the second intermediate matrix is ​​determined, and a plurality of matrix elements on the matrix diagonal are determined, and then a sum operation is performed on the plurality of matrix elements to obtain the sum of the first elements.

[0209] For example: the second intermediate matrix is ​​a 3*3 matrix, and the multiple matrix elements on the matrix diagonal corresponding to the second intermediate matrix are determined to include: the matrix elements in the first row and first column, the matrix elements in the second row and second column, and the matrix elements in the third row and third column; the matrix elements in the first row and first column, the matrix elements in the second row and second column, and the matrix elements in the third row and third column are summed to obtain the sum of the first elements.

[0210] In an optional embodiment, based on the sum of the first elements and the second matrix, a statistical mean square error is obtained using a preset function.

[0211] Illustratively, after obtaining the sum of the first elements, the sum of the first elements corresponding to the first matrix and the second matrix are combined to obtain the statistical mean square error through a preset function.

[0212] In some embodiments, a sum operation is performed on the first matrix elements in the first matrix to obtain the sum of the second elements.

[0213] Illustratively, the first matrix includes a plurality of first matrix elements, and a sum operation is performed on the plurality of first matrix elements to obtain the sum of the second elements.

[0214] For example, the first matrix is ​​a 3*3 matrix including 9 first matrix elements, and a sum operation is performed on the 9 first matrix elements to obtain the sum of the second elements.

[0215] In some embodiments, the adjustment parameter is obtained based on the sum of the first element and the sum of the second element.

[0216] Illustratively, after obtaining the sum of the first element and the sum of the second element, the sum of the first element and the sum of the second element are processed by the following formula 7 to obtain the adjustment parameter, and formula 7 is the second operation function mentioned above.

[0217] Formula 7:

[0218]

[0219] Wherein, B(X) is used to represent the adjustment parameter obtained by the second operation function in the process of obtaining the mean square error of the statistic; C is the first matrix; X is the first intermediate matrix; CX is the second intermediate matrix; trace is used to represent the sum of the matrix elements on the matrix diagonal of the matrix, then trace(CX) means the sum of the matrix elements on the matrix diagonal of the matrix CX, and the result obtained is the sum of the first element; ∑ 1≤i,j≤K C ij It is used to represent the sum of the second elements obtained by summing multiple first matrix elements in the first matrix. That is, the second operation function is a function that uses the first matrix and the first dependent variable as independent variables to obtain the corresponding dependent variable. The value of the dependent variable B(X) can be called the second dependent variable, that is, the adjustment parameter represented here.

[0220] The adjustment parameter is used to adjust the statistical mean square error according to the preset matrix.

[0221] Schematically, as shown in Formula 7, the adjustment parameters are obtained through the first matrix and the preset matrix. The adjustment parameters can better reflect the entity association relationship between the m candidate groups represented by the first matrix, avoiding the limitation of analyzing the m candidate groups as independent groups separately; therefore, by adjusting the second matrix with the help of the adjustment parameters, it is possible to perform a more comprehensive adjustment process on the second matrix with the help of the adjustment parameters that represent the group association situation on the basis of maintaining the entity association relationship represented by the second matrix and determined by comprehensively integrating multiple identical candidate groups.

[0222] In some embodiments, under a preset function, the second matrix is ​​adjusted with an adjustment parameter to obtain a statistical mean square error.

[0223] In an illustrative manner, after obtaining the adjustment parameters, based on the preset function shown in Formula 5, the second matrix is ​​adjusted by adjusting the parameter B(X) to obtain the statistical mean square error M(R); wherein the information of the second matrix is ​​contained in V(X(R)0 obtained by operating the third operation function in the preset function.

[0224] Optionally, the content of the second matrix expressed by the third operation function V(X(R)) is as follows.

[0225] In some embodiments, a second transposed matrix of the second matrix is ​​obtained.

[0226] Illustratively, the second matrix is ​​transposed to obtain a second transposed matrix corresponding to the second matrix.

[0227] Optionally, if the second matrix is ​​a matrix with m matrix rows and 1 matrix columns, after the second matrix is ​​transposed, the second transposed matrix is ​​a matrix with 1 matrix row and m matrix columns.

[0228] In some embodiments, the first intermediate matrix is ​​matrix adjusted by matrix product between the second matrix and the second transposed matrix to obtain a third intermediate matrix.

[0229] Optionally, an adjustment matrix is ​​obtained, and the adjustment matrix is ​​a column matrix with m matrix rows and 1 matrix element, that is, the number of matrix columns of the adjustment matrix is ​​1, which can be expressed as 1=(1,1,…,1) T , the length of the adjustment matrix is ​​m, and each matrix element is 1.

[0230] Optionally, a third transposed matrix corresponding to the adjustment matrix is ​​obtained. Schematically, the adjustment matrix is ​​transposed to obtain the third transposed matrix. Based on the adjustment matrix being 1=(1,1,…,1) T , then the third transposed matrix is ​​1 T =(1,1,…,1).

[0231] Optionally, the matrix product between the adjustment matrix and the third transposed matrix is ​​adjusted with the first intermediate matrix to obtain an intermediate adjustment matrix; and the matrix product is performed on the intermediate adjustment matrix, the second matrix and the second transposed matrix to obtain a third intermediate matrix.

[0232] In some embodiments, a sum operation is performed on matrix update elements on the matrix diagonal corresponding to the third intermediate matrix to obtain the sum of the third elements.

[0233] Schematically, the first intermediate matrix is ​​an m*m matrix; the matrix product between the adjustment matrix and the third transposed matrix is ​​an m*m matrix, that is, the matrix product between the column matrix with m matrix rows and the row matrix with m matrix columns is an m*m matrix; therefore, the intermediate adjustment matrix is ​​an m*m matrix; the second matrix is ​​a column matrix with m matrix rows, and the second transposed matrix is ​​a row matrix with m matrix columns, therefore, the matrix product of the second matrix and the second transposed matrix is ​​also an m*m matrix, therefore, the third intermediate matrix is ​​also an m*m matrix.

[0234] Optionally, based on the fact that the third intermediate matrix is ​​also an m*m matrix, a matrix diagonal corresponding to the third intermediate matrix is ​​determined, and multiple matrix elements on the matrix diagonal are determined, and then the multiple matrix elements are summed to obtain the sum of the third elements.

[0235] Illustratively, the sum of the third element is obtained using the following formula 8, which is the third operation function.

[0236] Formula 8:

[0237]

[0238] Where V(X) is the sum of the third element; D is the second matrix; D T is the second transposed matrix corresponding to the second matrix; X is the first intermediate matrix mentioned above; 1 is the adjustment matrix; 1 T is the third transposed matrix corresponding to the adjustment matrix; It is used to represent the third intermediate matrix; trace is used to represent the sum of the matrix elements on the matrix diagonal of the matrix. It is used to represent the sum of the matrix elements on the matrix diagonal of the third intermediate matrix, and the result is the sum of the third elements. That is, the third operation function is a function that uses the second matrix and the first dependent variable as independent variables to obtain the corresponding dependent variable. The value of the dependent variable V(X) can be called the third dependent variable, that is, the sum of the third elements represented here.

[0239] In some embodiments, under a preset function, the sum of the third elements is adjusted by adjusting parameters to obtain a statistical mean square error.

[0240] Schematically, as shown in Formula 5 above, after obtaining the adjustment parameter B(X), i.e., B(X(R)), and obtaining the sum of the third elements V(X) determined by the second matrix, i.e., V(X(R)), V(X) is adjusted by the square of the adjustment parameter B(X), thereby obtaining the statistical mean square error based on the first matrix, the second matrix and the preset matrix.

[0241] In some embodiments, if the first matrix, the second matrix and the preset matrix are known, that is, the values ​​of the matrix elements in the matrices are known, the statistical mean square error can be calculated by formula five as above; based on this process, the first matrix element in the first matrix and the second matrix element in the second matrix are contents determined by the entity relationship diagram, and the preset matrix is ​​the content that needs to be updated, so it can be regarded that the statistical mean square error obtained by solving the first matrix, the second matrix and the preset matrix has a corresponding relationship with the current preset matrix. If the preset matrix is ​​updated, the corresponding calculated statistical mean square error may also change.

[0242] Step 530 , updating the preset matrix with the goal of minimizing the statistical mean square error to obtain an updated matrix.

[0243] Schematically, the error between the statistic used to estimate the overall parameter and the overall parameter is characterized based on the statistical mean square error. Therefore, if we want to more accurately analyze multiple entity nodes in the entity relationship diagram through the statistic corresponding to the statistical mean square error, with the goal of minimizing the statistical mean square error, we can make the obtained statistics more conducive to accurately estimating the overall parameter as much as possible.

[0244] In an optional embodiment, a partial derivative operation is performed on a preset function and a preset matrix to obtain a partial derivative matrix.

[0245] Illustratively, in addition to obtaining the statistical mean square error by operating the first matrix, the second matrix and the preset matrix through a preset function, partial derivative operations are performed on the preset function and the preset matrix to obtain a partial derivative matrix.

[0246] Indicatively, the preset function (abbreviated as M(R)) formula shown in Formula 5 above is used to obtain the partial derivative of the preset matrix R, and the partial derivative matrix is ​​obtained, which is expressed as

[0247] Optionally, the partial derivative matrix is ​​a Jacobian matrix. Based on the preset matrix R being an m*m matrix, the partial derivative matrix is ​​also an m*m matrix.

[0248] In an optional embodiment, based on the partial derivative matrix, a gradient descent algorithm is used to update the preset matrix with the goal of minimizing the statistical mean square error to obtain an updated matrix.

[0249] Optionally, an adaptive moment estimation (Adam) method is used as a gradient descent algorithm. Adam is an advanced gradient descent optimization algorithm and can also be called an adaptive optimization operator.

[0250] In schematic form, the Adam operator is an iterative algorithm for finding the optimal solution to a function. For example, given a convex function f(X), where X is a vector or matrix, the Adam operator can iteratively find the value of X when f(X) reaches the minimum value.

[0251] In some embodiments, based on the partial derivative matrix, a preset matrix is ​​updated by a gradient descent algorithm with the goal of minimizing the statistical mean square error to obtain a preliminary updated matrix.

[0252] In schematic form, the Adam operator is used to optimize the statistical mean square error M(R), with the goal of minimizing the statistical mean square error M(R). The iterative solution is used to obtain the value of the preliminary update matrix when the statistical mean square error M(R) is minimized. The preliminary update matrix corresponds to the minimum statistical mean square error; the preset matrix corresponds to the statistical mean square error before iterative optimization.

[0253] In some embodiments, a regularization adjustment is performed on each matrix row of the preliminary update matrix to obtain an update matrix.

[0254] Schematically, for each matrix row of the preliminary update matrix determined based on the gradient descent algorithm, L2-norm is used to regularize it to obtain the update matrix.

[0255] Among them, the condition for updating the matrix needs to satisfy RR T The diagonal elements of are all 1, and the remaining elements are between -1 and 1; therefore, the purpose of the L2-norm method is to make the update matrix R meet the above conditions.

[0256] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0257] In summary, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between candidate groups and causing analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, a more targeted matrix update process can be performed on the preset matrix, which is conducive to improving the accuracy of the classification results known by updating the matrix, and then using more accurate classification results to obtain more precise experimental data.

[0258] In an embodiment of the present application, the content of obtaining an updated matrix by minimizing the statistical mean square error is introduced. The first matrix, the second matrix and the preset matrix are subjected to matrix operations by a preset function to obtain the statistical mean square error. The preset function can determine the upper bound of the statistical mean square error corresponding to the preset matrix, and then the preset matrix is ​​updated with the goal of reducing the statistical mean square error, until the statistical mean square error optimization is completed, and the updated matrix corresponding to the currently optimized statistical mean square error is obtained. The updated matrix is ​​used as a statistic that can better estimate the processing results, which is conducive to more accurately knowing the classification results corresponding to multiple candidate groups, avoiding the problem of large deviations in experimental results caused by randomly dividing candidate groups into category groups, thereby greatly improving the accuracy of the experimental results.

[0259] In an optional embodiment, after the update matrix is ​​obtained, the update matrix is ​​adjusted in rows and columns by a preset adjustment matrix, and the classification results corresponding to the m candidate groups are determined based on the row and column adjustment matrix. Figure 6 As shown above Figure 2 The embodiment shown can also be implemented as the following steps 610 to 650; Figure 2 The illustrated step 240 may also be implemented as the following steps 640 to 650 .

[0260] Step 610, obtaining an entity relationship diagram.

[0261] Among them, the entity relationship graph is composed of multiple entity nodes and entity relationship edges, and the entity relationship edges are used to describe the entity association relationship between entity nodes.

[0262] Schematically, entity nodes having entity association relationships among a plurality of entity nodes are associated through entity relationship edges, thereby obtaining an entity relationship graph consisting of a plurality of entity nodes and entity relationship edges.

[0263] The multiple entity nodes are divided into m candidate groups, where m is an integer greater than 1.

[0264] Schematically, multiple nodes in the entity relationship graph are pre-divided into m candidate groups. Optionally, m is a pre-selected division value, and the m candidate groups can analyze the relationship between multiple entity nodes in the form of groups.

[0265] In some embodiments, after obtaining a plurality of entity nodes, the plurality of entity nodes are randomly divided into m candidate groups, each candidate group including at least one entity node.

[0266] Step 620 , obtaining a first matrix and a second matrix based on the entity association relationships represented in the entity relationship graph and the m candidate groups.

[0267] The first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group.

[0268] In some optional embodiments, the m candidate groups are divided into pairs to obtain multiple group division results.

[0269] Each group division result includes two candidate groups.

[0270] In an optional embodiment, the number of entity relationship edges between two candidate groups in each group division result is determined to obtain the number of relationship edges corresponding to the multiple group division results.

[0271] Illustratively, after obtaining multiple grouping results, each grouping result is analyzed separately to determine two candidate groups in one grouping result, and determine the number of entity relationship edges between the two candidate groups, which corresponds to the grouping result.

[0272] The entity relationship edge is used to connect two entity nodes having an entity association relationship, so the entity relationship edge between two candidate groups is used to represent that the entity relationship edge spans the two candidate groups.

[0273] In an optional embodiment, a first matrix is ​​generated with a plurality of relationship edge numbers as first matrix elements.

[0274] Indicatively, for m candidate groups, an m*m first matrix is ​​constructed, and C represents the first matrix, and C represents ij represents the first matrix element constituting the first matrix, where i is any candidate group; j is also any candidate group. When the first matrix element is C ii , then C ii =0; then the first matrix element C ij The first matrix C is determined by the number of relationship edges between candidate group i and candidate group j, and is expressed as the above formula 1.

[0275] In an optional embodiment, for a plurality of entity nodes in the entity relationship graph, the number of entity nodes having an entity association relationship with each entity node is determined to obtain the number of associated nodes corresponding to the plurality of entity nodes.

[0276] Schematically, multiple entity nodes in the entity relationship diagram are taken as analysis objects respectively, and the multiple entity nodes are analyzed respectively; for any entity node, other entity nodes that have entity association relationships with the entity node are determined, and the number of nodes of the other entity nodes is determined as the number of associated nodes corresponding to the entity node.

[0277] In an optional embodiment, within the same candidate group, the sum of the numbers of associated nodes corresponding to the entity nodes is determined to obtain the sum of the numbers of nodes corresponding to the m candidate groups.

[0278] In an optional embodiment, the second matrix is ​​generated with the sum of the number of nodes as the second matrix elements.

[0279] Indicatively, for m candidate groups, an m*1 second matrix is ​​constructed, and D represents the second matrix. i represents the second matrix elements constituting the second matrix, where i is any candidate group; then the second matrix element D i The second matrix D is determined by the sum of the number of nodes corresponding to the candidate group i, and is expressed as shown in the above formula three.

[0280] Step 630 , updating the preset matrix with the goal of reducing the statistical mean square error to obtain an updated matrix.

[0281] Among them, the statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix; the statistical mean square error is the mean square error expression of the statistic when performing the estimation processing, the statistic is a parameter of the estimation processing result through the first matrix and the second matrix, and the update matrix is ​​composed of the update matrix elements.

[0282] In an optional embodiment, a preset function is obtained.

[0283] Illustratively, the preset function is a pre-set function, and the function variables in the preset function include a first variable indicating a first matrix, a second variable indicating a second matrix, and a third variable indicating a preset matrix.

[0284] In an optional embodiment, a matrix operation is performed on the first matrix, the second matrix, and the preset matrix using a preset function to obtain a statistical mean square error.

[0285] Illustratively, when a preset matrix is ​​obtained, and the first matrix and the second matrix are known, a preset function as shown in Formula 5 above is used to perform matrix operations to obtain a statistical mean square error.

[0286] In an optional embodiment, a unit matrix is ​​obtained as a preset matrix, and the number of matrix rows and the number of matrix columns of the unit matrix are m.

[0287] Schematically, the number of matrix rows and the number of matrix columns of the identity matrix are the same, both are m, so the identity matrix is ​​a square matrix; based on the properties of the identity matrix, the diagonal of the identity matrix is ​​determined to be 1, and the other matrix elements are 0.

[0288] For example, if the number m of candidate groups is 3, the number of matrix rows and the number of matrix columns of the unit matrix are both 3, and the unit matrix is ​​expressed as the following formula 9.

[0289] Formula 9:

[0290]

[0291] In an optional embodiment, the preset matrix is ​​iterated n times to update the matrix with the goal of reducing the statistical mean square error, to obtain an updated matrix.

[0292] Wherein, n is a positive integer and n is the preset number of iterative updates.

[0293] Optionally, when n is 1, it means that the preset matrix is ​​updated once with the goal of reducing the statistical mean square error to obtain an updated matrix.

[0294] Optionally, when n is greater than 1, it means that the preset matrix is ​​iteratively updated multiple times to obtain an updated matrix with the goal of reducing the statistical mean square error.

[0295] In some embodiments, during the s+1th iterative matrix update process for the preset matrix, a statistical mean square error corresponding to the s+1th update process is obtained, where s is a positive integer less than the preset number of iterative updates.

[0296] The statistical mean square error corresponding to the s+1th update process is determined based on the first matrix, the second matrix and the preset matrix after the sth update; wherein s is a positive integer less than n.

[0297] Illustratively, during the s+1 iterative matrix update process for a preset matrix, the statistical mean square error corresponding to the s-th update process is determined; with the help of a preset function, the first matrix, the second matrix and the statistical mean square error corresponding to the s-th update process are solved to obtain the statistical mean square error corresponding to the s+1-th update process.

[0298] Optionally, when obtaining the statistical mean square error corresponding to the s+1th update process through the preset function, the first error function in the preset function is used to perform matrix processing on the first matrix and the preset matrix after the s-th update to obtain the first error corresponding to the s+1th update process; in addition, the second error function in the preset function is used to perform matrix processing on the second matrix and the preset matrix after the s-th update to obtain the second error corresponding to the s+1th update process, thereby combining the first error corresponding to the s+1th update process and the second error corresponding to the s+1th update process to obtain the statistical mean square error corresponding to the s+1th update process.

[0299] Schematically, the first error corresponding to the first update process is obtained by performing matrix processing on the first matrix and the preset matrix (such as the unit matrix) obtained directly through the first error function in the preset function; in addition, the second error corresponding to the first update process is obtained by performing matrix processing on the second matrix and the preset matrix (such as the unit matrix) obtained directly through the second error function in the preset function, and the sum of the first error and the second error is used as the statistical mean square error corresponding to the first update process. Thereafter, in the second update (i.e., obtaining the statistical mean square error corresponding to the s+1th update process), the first error corresponding to the second update process is obtained by performing matrix processing on the first matrix and the statistical mean square error corresponding to the first update process through the first error function in the preset function; in addition, the second error corresponding to the second update process is obtained by performing matrix processing on the second matrix and the statistical mean square error corresponding to the first update process through the second error function in the preset function, and the sum of the first error and the second error is used as the statistical mean square error corresponding to the second update process, etc.

[0300] That is, in the process of updating the preset matrix, the preset matrix and the preset matrix after the intermediate stage update can be regarded as the statistics of the intermediate stage, and the statistics of the intermediate stage are the statistics that have not been optimized; the updated matrix obtained after the final update can be regarded as the optimized statistics. Therefore, the preset matrix is ​​used to obtain the updated matrix through the iterative update process, and the updated matrix is ​​used as the matrix expression of the representative statistics obtained after optimization, which is conducive to better predicting the overall node information represented by multiple entity nodes, thereby more accurately predicting the correlation between entity nodes.

[0301] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0302] In some embodiments, with the goal of reducing the statistical mean square error corresponding to the s+1th update process, the preset matrix after the sth update is updated s+1 times; in response to performing n iterative matrix updates on the preset matrix, an updated matrix is ​​obtained.

[0303] In an illustrative example, a matrix updating process is performed based on a unit matrix as a preset matrix. In a first updating process, the unit matrix is ​​updated once with the goal of reducing a statistical mean square error obtained through the first matrix, the second matrix, and the preset matrix, thereby obtaining a matrix after the first update.

[0304] Afterwards, based on the matrix after the first update, the first matrix and the second matrix, the statistical mean square error corresponding to the second update process is obtained, and then with the goal of reducing the statistical mean square error, the matrix after the first update is updated, etc., until the preset number of iterations n is reached, and the updated matrix is ​​obtained.

[0305] That is, the statistical mean square error is calculated based on the first matrix, the second matrix and the preset matrix or the matrix after the most recent update, so as to reduce the statistical mean square error as the goal, and execute the current update process until the updated matrix is ​​obtained after n iterative updates.

[0306] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0307] Step 640: Perform row and column adjustment on the update matrix using a preset adjustment matrix to obtain a row and column adjustment matrix.

[0308] Schematically, R* represents the update matrix, which is an m*m matrix. In order to facilitate the determination of classification results corresponding to multiple candidate groups through the update matrix elements in the update matrix, the update matrix is ​​adjusted in rows and columns through a preset adjustment matrix to obtain an m*1 matrix for analysis. The matrix obtained by adjusting the rows and columns of the update matrix through the preset adjustment matrix is ​​called a row and column adjustment matrix.

[0309] In some embodiments, a preset adjustment matrix is ​​obtained by random sampling from a normal distribution, such as a matrix U=(U 1 ,U 2 ,…,U m ) T , is an m*1 matrix.

[0310] The update matrix R* is adjusted in rows and columns by using the preset adjustment matrix U to obtain a row and column adjustment matrix, that is, the matrix R*U. Based on the row and column adjustment process, the row and column adjustment matrix R*U becomes an m*1 matrix.

[0311] Step 650 : determining classification results corresponding to the m candidate groups respectively based on the adjustment matrix elements in the row and column adjustment matrix.

[0312] Schematically, the matrix elements in the row-column adjustment matrix are called adjustment matrix elements. When the classification results respectively corresponding to the m candidate groups are determined by the adjustment matrix elements, the row-column adjustment matrix is ​​first mapped by a sign function.

[0313] In an optional embodiment, a sign function is used to perform sign analysis on the adjustment matrix elements in the row and column adjustment matrix to obtain sign judgment results corresponding to the plurality of adjustment matrix elements.

[0314] In an illustrative manner, for the row-column adjustment matrix R*U, the row-column adjustment matrix R*U is mapped to sgn(R*U), that is, each adjustment matrix element x in the row-column adjustment matrix is ​​mapped to sgn(x); in addition, sgn(x) is adjusted using the following formula 10 to obtain a sign judgment result.

[0315] Formula 10:

[0316]

[0317] Wherein, T is the sign judgment result; sgn is the sign function, if the adjustment matrix element x is greater than 0, then sgn(x) is 1, if the adjustment matrix element x is less than 0, then sgn(x) is 0.

[0318] For example, the row-column adjustment matrix R*U includes m adjustment matrix elements x, and for each adjustment matrix element x, m symbol judgment results T=(T 1 ,T 2 ,…,T m ).

[0319] In an optional embodiment, the classification results corresponding to the m candidate groups are determined based on the symbol judgment results.

[0320] In an illustrative manner, after obtaining the symbol judgment result, the symbol judgment result is the content obtained by the adjustment matrix elements in the row and column adjustment matrix with a matrix size of m*1, so the m adjustment matrix elements correspond one-to-one to the m candidate groups; the symbol judgment results corresponding one-to-one to the m adjustment matrix elements also correspond one-to-one to the m candidate groups.

[0321] Optionally, the adjustment matrix elements within the row and column adjustment matrix are arranged in sequence according to the order of the candidate groups, so that the corresponding multiple symbol judgment results are arranged in sequence according to the order of the candidate groups; the classification result corresponding to the first candidate group is determined based on the symbol judgment result corresponding to the first candidate group; similarly, the classification result corresponding to the p-th candidate group is determined based on the symbol judgment result corresponding to the p-th candidate group, and so on, where p is a positive integer.

[0322] Indicatively, based on the symbol judgment result T=(T 1 ,T 2 ,…,T m ) classify the m candidate groups respectively, that is, determine the classification results corresponding to the m candidate groups respectively.

[0323] The classification result is used to represent the category group for classifying the candidate group, and different category groups are used to perform different processing methods.

[0324] Optionally, the pre-set category groups include experimental groups and control groups. The experimental groups are used to perform data analysis and processing in the set experimental form; the control groups are used not to perform the data analysis and processing, or to perform the data analysis and processing in a form different from the set experimental form, etc.

[0325] Schematically, the classification result corresponding to the candidate group m1 is the experimental group, which means that the candidate group m1 is classified into the experimental group, so that the entity nodes in the candidate group m1 are subjected to data analysis and processing through the set experimental form; or, the classification result corresponding to the candidate group m2 is the control group, which means that the candidate group m1 is classified into the control group, so that the entity nodes in the candidate group m2 are subjected to data analysis and processing through other experimental forms, etc.

[0326] In some embodiments, in response to the symbol determination result being a first value, the candidate group corresponding to the symbol determination result is classified into the first group.

[0327] Illustratively, taking the first value of 1 as an example, when the symbol judgment result is 1, the candidate group corresponding to the symbol judgment result is determined, and the candidate group is classified into the first group, and the classification result corresponding to the candidate group is the "first group".

[0328] For example: The symbol judgment result T = (T 1,T 2 ,…,T m ) indicates T 2 =1, it means that the classification result of candidate group 2 is "first group", so candidate group 2 is classified into the first group, such as the first group is the experimental group.

[0329] In some embodiments, in response to the symbol determination result being a second value, the candidate group corresponding to the symbol determination result is classified into the second group.

[0330] The first value is different from the second value, the first group is different from the second group, and the first group and the second group are used to characterize groups that conduct experimental processes through different experimental methods.

[0331] Illustratively, taking the second value of 0 as an example, when the symbol judgment result is 0, the candidate group corresponding to the symbol judgment result is determined, and the candidate group is classified into the second group. The classification result corresponding to the candidate group is the "second group".

[0332] For example: The symbol judgment result T = (T 1 ,T 2 ,…,T m ) indicates T 1 =0, it means that the classification result of candidate group 1 is "the second group", so candidate group 1 is classified into the second group, such as the second group is the control group.

[0333] Compared with the random diversion method, the above division method fully takes the statistics into consideration and makes the determination of the statistics more accurate by effectively reducing the statistical mean square error of the statistics, which is conducive to making the process of group division based on statistics more accurate and improving the authenticity of the analysis data.

[0334] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0335] In an optional embodiment, if Figure 7 As shown in the figure, the group division method is illustrated.

[0336] (1) Based on multiple entity nodes and entity association relationships between entity nodes, an entity relationship graph 710 is constructed, wherein the entity relationship edges describe the entity association relationships between entity nodes.

[0337] (2) Divide multiple entity nodes into m candidate groups, where m is an integer greater than 1. Figure 7 Taking m=2 as an example, a plurality of entity nodes (shown as 11 entity nodes in the figure) are divided into two candidate groups, namely candidate group 711 and candidate group 712.

[0338] (3) Based on a method of dividing the plurality of candidate groups into two groups, a group division result is obtained, which indicates that the candidate group 711 and the candidate group 712 are divided into one group division result; the number of entity relationship edges between the two candidate groups in the group division result is determined, and the number of relationship edges corresponding to the group division result is obtained; Figure 7 The number of relationship edges corresponding to the group division results is 4.

[0339] In addition, the first matrix is ​​generated with the number of relationship edges as the first matrix elements. Figure 7 As shown, based on the number of candidate groups being 2, the first matrix is ​​set to be a 2*2 matrix, so it includes four first matrix elements; based on the above formula 1, the first matrix element c is determined 11 =0; the first matrix element c 12 =4; the first matrix element c 21 =4; the first matrix element c 22 =0.

[0340] (4) For multiple entity nodes in the entity relationship graph 710, determine the number of entity nodes that have an entity association relationship with each entity node, and obtain the number of associated nodes corresponding to the multiple entity nodes; within the same candidate group, determine the sum of the number of associated nodes corresponding to the entity nodes, and obtain the sum of the number of nodes corresponding to the two candidate groups. Among them, the sum of the number of nodes corresponding to the candidate group 711 is 1+3+3+1+2=10; the sum of the number of nodes corresponding to the candidate group 712 is 2+1+2+2+2+1=10.

[0341] Then, a second matrix is ​​generated with the sum of multiple node numbers as the second matrix elements. Figure 7 As shown, based on the fact that the sum of the number of nodes corresponding to candidate group 711 is 10 and the sum of the number of nodes corresponding to candidate group 712 is 10, the second matrix element D is determined by the above formula 3. 1 =10; the second matrix element D 2 =10.

[0342] (5) Obtaining the statistical mean square error between the first matrix, the second matrix and the preset matrix through a preset function. Schematically, the first error function in the preset function is used to perform matrix processing on the first matrix and the preset matrix to obtain a first error; the second error function in the preset function is used to perform matrix processing on the second matrix and the preset matrix to obtain a second error, thereby combining the first error and the second error to obtain the statistical mean square error. Afterwards, with the goal of reducing the statistical mean square error, the preset matrix is ​​iteratively updated until an updated matrix is ​​obtained.

[0343] Among them, the statistical mean square error is the mean square error expression of the statistic when performing the estimation process, and the statistic is a parameter of the estimation process result through the first matrix and the second matrix; the update matrix is ​​composed of update matrix elements.

[0344] (6) Determine the classification results corresponding to the m candidate groups based on the updated matrix elements.

[0345] The classification result is used to characterize the category group for classifying the candidate group, and different category groups are used to perform different processing methods. For example, if the classification result corresponding to the candidate group 711 indicates that the candidate group 711 is classified into the category experimental group, the candidate group 711 is processed and analyzed by the experimental processing method; if the classification result corresponding to the candidate group 712 indicates that the candidate group 712 is classified into the control group, the candidate group 712 is processed and analyzed by the control processing method, etc.

[0346] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0347] In summary, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between candidate groups and causing analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, a more targeted matrix update process can be performed on the preset matrix, which is conducive to improving the accuracy of the classification results known by updating the matrix, and then using more accurate classification results to obtain more precise experimental data.

[0348] In the embodiment of the present application, it is introduced that the update matrix is ​​adjusted in rows and columns by a preset adjustment matrix, so as to determine the content of the classification results corresponding to the m candidate groups respectively based on the row and column adjustment matrix. By adjusting the rows and columns of the preset adjustment matrix, the adjustment matrix elements in the row and column adjustment matrix correspond to the m candidate groups one by one, so as to facilitate the value of the result according to the symbol corresponding to the m adjustment matrix elements one by one, and more quickly know the classification results corresponding to the m candidate groups respectively, and improve the efficiency of obtaining the classification results.

[0349] In an optional embodiment, the group division method is commonly used in AB testing experiments.

[0350] AB testing is an emerging method for optimizing products by collecting and analyzing feedback data from users through authorization. Schematically, first set up experimental group A and experimental group B, which differ in a parameter that needs to be optimized (e.g., the product icon color of experimental group A is green, and the icon color of experimental group B is red). The users of the product are randomly directed to experimental group A or experimental group B. The data generated by the object operation is recorded after authorization in this process, and the data is analyzed accordingly. According to the results of the analysis, it is determined which experimental group A and experimental group B use better parameters, so as to achieve the purpose of optimizing the product. The users (traffic) are randomly directed to two different groups, A and B, whose product characteristics are somewhat different. After a period of time, if it is found that the product effect of experimental group B (measured by specific indicators, such as click-through rate, conversion rate, etc.) is higher, it means that the product characteristics of experimental group B are better.

[0351] Suppose you need to compare which strategy B or control strategy A is better, the common approach is to randomly divide the experimental subjects into the experimental group and the control group. The experimental strategy is applied to the users in the experimental group, and the control strategy is applied to the users in the control group. Finally, the data of all the subjects in each group are collected, and the sample means of the data collected from the two groups are compared to perform hypothesis testing and determine which strategy is more effective.

[0352] The above process has an assumption, called the Stable Unit Treatment Value Assumption (SUTVA). That is, the behavior of an object is only related to the strategy it receives, and has nothing to do with the strategies received by other objects. However, in a social scenario, this assumption is usually not met, and an object is likely to be affected by the behavior of his social neighbors (such as relatives and friends).

[0353] For example: In the experimental group, if the experimental strategy can make the subject forward more videos to his friends, then his friends, even in the control group, will receive more videos forwarded by their friends and click to watch them, which will be indirectly affected by the experimental strategy and will forward more videos to their friends' friends. At this time, the SUTVA assumption no longer holds. This phenomenon is generally called the network effect. When the assumption is not true, the effect of the experimental strategy will be misestimated according to the commonly used experimental method of randomly grouping objects. The common method to solve the network effect is to first define an entity relationship graph that describes the propagation path of the network effect, and then divide the entity relationship graph into many non-intersecting communities (clusters) through an algorithm, and then randomly divide the objects into the experimental group and the control group according to the community, and finally collect data for hypothesis testing.

[0354] In the related art, it is assumed that there is an undirected graph G = (V, E), where V = {1, 2, ..., n}, is a set of n entity nodes, and E = {(i, j)}, is a set of entity relationship edges in the undirected graph, that is: if (i, j) ∈ E, then it indicates that there is an entity relationship edge between entity node i and entity node j.

[0355] Optionally, the undirected graph represents the relationship between groups of people. Each entity node represents a user object, and each entity relationship edge represents the existence of a network effect propagation relationship between the user objects corresponding to the entity node.

[0356] Assume that the graph G can be divided into K non-overlapping clusters (such as the m candidate groups mentioned above) by a certain algorithm, where K is a positive integer; under the traditional cluster randomization scheme, each cluster is independently divided using the Bernoulli distribution (Bernoulli) 0.5 allocation strategy, that is, there is a 50% probability of being assigned to the experimental group and a 50% probability of being assigned to the control group. Among them, the objects in the cluster assigned to the experimental group accept the experimental strategy; the objects in the cluster assigned to the control group accept the control strategy.

[0357] Remember T j ∈{0,1} represents the strategy assigned to the jth cluster, i.e., T j =1 means the jth cluster is assigned to the experimental group, T j =0 means that the jth cluster is assigned to the control group. i Represents the experimental observation data of object i (any object). i ∈{0,1} represents the allocation strategy of object i, that is, Z i =1 means that subject i is assigned to the experimental group, Z j =0 means that subject j was assigned to the control group.

[0358] Therefore, the difference between the experimental strategy and the control strategy can be estimated using the following formula 11.

[0359] Formula 11:

[0360]

[0361] in, is an estimator, or statistic, used to estimate the difference between the experimental strategy and the control strategy.

[0362] However, the disadvantage of the related technology is that it relies on a strong assumption, that is, there is no network propagation path between different clusters, and all network effects are within the same cluster. In other words, after an entity relationship graph is divided into multiple clusters, there are no entity relationship edges between different clusters, and the two entity nodes of each entity relationship edge are within the same cluster. This assumption is usually not met in social network scenarios, because social networks are relatively dense networks, and it is impossible to divide clusters completely cleanly so that there are no entity relationship edges between clusters. For example: the ratio of the number of entity relationship edges between two entity nodes in the same cluster to the total number of entity relationship edges is the edge ratio. Only when the edge ratio is equal to 100% or close to 100%, the above estimate is In order to accurately estimate the real difference in strategies. Usually, after the clusters are divided, the edge proportion is only about 40%, so the parameter estimation based on the above Bernoulli 0.5 randomization method is inaccurate. The corresponding mean square error (MSE) is usually large and is not suitable for obtaining good experimental results.

[0363] In an optional embodiment, the above grouping method is applied to the social field, such as an entity relationship diagram is a social network diagram constructed based on social relationships, and the above grouping method is called "a method for optimizing random schemes to reduce MSE in social network experiments". This method converts MSE into a convex optimization problem, solves the convex optimization, and obtains a randomized scheme. Convex optimization problems are used to solve the problem of obtaining a global optimal solution within a domain of definition for convex functions. The random sampling steps of the convex optimization problem are given below, which can be implemented as follows.

[0364] (a) Obtain the entity relationship diagram.

[0365] The entity relationship graph can be expressed as an undirected graph G. The entity relationship graph consists of multiple entity nodes and entity relationship edges, and the entity relationship edges are used to describe the entity association relationship between entity nodes.

[0366] Indicatively, since the randomization solution process is performed on clusters (candidate groups), it is assumed that there are K clusters in total (such as the m candidate groups mentioned above), so only the randomization scheme of these K clusters needs to be designed.

[0367] (2) Obtain the first matrix and the second matrix.

[0368] Let C = (C ij) is a K×K symmetric matrix, that is, the first matrix mentioned above. ii =0, C ij is the number of entity relationship edges connecting clusteri and clusterj.

[0369] Remember i is the number of friends of object i (one object corresponds to one entity node, so object i is entity node i) (the number of objects that have entity association with object i), D j is the d of all objects in cluster j i The sum of D = (D 1 ,D 2 ,…,D k ) T , that is: the second matrix mentioned above.

[0370] (3) Perform matrix update on the preset matrix to obtain an updated matrix.

[0371] Assume that R is a K×K symmetric matrix (a preset matrix used to update the update matrix). Define the following functions:

[0372] (1) Record X That is: as above formula 6;

[0373] (2) Let B(X) = 4*trace(CX)-∑ 1≤i,j≤K C ij , that is: as above formula 7;

[0374] (3) Record Where 1=(1,1,…,1) T It means that each element of length K is 1, that is: as shown in formula 8 above.

[0375] The trace of the matrix is ​​used to represent the sum of the matrix elements on the matrix diagonal of the matrix. Finally, the function is defined as shown in Formula 5 above:

[0376] M(R)=B(X(R)) 2 +8(ω 2 +4)V(X(R))

[0377] In order to perform convex optimization solution through the above formulas 5 to 8 to update the preset matrix R and obtain the updated matrix R*, the following convex optimization expression is proposed:

[0378] where the matrix R satisfies RR T The diagonal elements of are all 1, and the rest of the elements are between -1 and 1.

[0379] In order to solve the above convex optimization, the following algorithm is proposed:

[0380] Input: graph G; K clusters; M (representing the number of iterations); ω (representing the preset parameters, such as 1 and other numerical values).

[0381] (1) Initialize R=I, that is, the preset matrix is ​​initialized to the unit matrix I, and i=0 (i is the number of loops).

[0382] (2) for i = 1, 2, …, M;

[0383] 2.1. Calculate M(R). M(R) is a numerical value, which is the statistical mean square error calculated based on the current R. When i=1, the current R is the unit matrix, and M(R) is calculated based on the unit matrix. When i=2, the current R is the intermediate matrix obtained after updating the unit matrix once, and M(R) is calculated based on the intermediate matrix, etc. That is, the statistical mean square error is calculated based on the intermediate matrix obtained after the most recent update.

[0384] 2.2. Optimization by Adaptive Optimization Operator Adam And update R, that is: in Represents the Jacobian matrix obtained by taking the partial derivative of the function M(R) with respect to R.

[0385] 2.3. Regularize each row of the updated R according to L2-norm.

[0386] Output: Update the matrix R*, that is, after updating M times, the optimized matrix R* is obtained.

[0387] After obtaining the update matrix R*, the grouping vectors of all clusters T = (T 1 ,…,T K ) T It can be obtained by the above formula + mapping sampling, that is:

[0388]

[0389] The grouping vector is the symbol judgment result mentioned above, which represents the classification results of k candidate groups. The classification result corresponding to each candidate group k is represented by T in the grouping vector T. k be confirmed.

[0390] Illustratively, the content of the obtained grouping vector is described as follows.

[0391] (1) First, from a standard multivariate normal distribution Randomly sample a vector U=(U 1 ,U2 ,…,U K ) T , that is: the preset adjustment matrix mentioned above;

[0392] (2) Calculate the vector R*U, that is, the row and column adjustment matrix mentioned above;

[0393] (3) Map the vector R*U to sgn(R*U), that is, map each element x in the vector to sgn(x), that is, map each adjustment matrix element in the above row and column adjustment matrix;

[0394] (4) The grouping vector T is calculated by the above formula 10.

[0395] After obtaining the grouping vector T, we can j Group each cluster. If T j =1, then the jth cluster is assigned to the experimental group; if T j =0, then the jth cluster is assigned to the control group.

[0396] This random diversion method can effectively reduce the amount of estimation The MSE of Also known as statistics, it is the update matrix R* determined based on the above process; the estimated quantity The MSE of is the M(R) corresponding to the update matrix, which can be expressed as M(R*), etc. That is, by effectively reducing the estimated MSE improves the accuracy of the estimate. The upper bound of the MSE is M(R). The principle of the above diversion method is to minimize an approximate solution of M(R).

[0397] The advantages of the embodiments of the present application are illustrated by numerical experiments below, considering the following two methods of generating simulation data.

[0398] (1) Model 1: where ε i is a standard normal distribution; Y i Represents the experimental observation data of object i; N i are all friends of object i; Z i is the allocation strategy of object i; Z j is the allocation strategy for object j, based on j∈N i , it means that objects i and j are friends; d i is the number of friends of object i; It is the average of the sum of the number of friends corresponding to multiple objects (multiple entity nodes).

[0399] (2) Model 2: where ε i is a standard normal distribution; Y i Represents the experimental observation data of object i; N i are all friends of object i; Z i is the allocation strategy of object i; Z j is the allocation strategy for object j, based on j∈N i , it means that objects i and j are friends; d i is the number of friends of object i; It is the average of the sum of the number of friends corresponding to multiple objects (multiple entity nodes).

[0400] The data generation process is simulated under the above two data generation models, and then experiments are performed according to the Bernoulli random scheme in the related art and the random scheme provided in the embodiment of the present application, and finally the calculation is performed. The MSE of , as shown in Table 1 below, is the simulation result.

[0401] Table 1

[0402]

[0403] From the above, we can see that under different data generation models, the randomization scheme proposed in this application makes the statistics The MSE is better than that of the traditional Bernoulli random scheme. There is a smaller MSE; based on the fact that MSE is used to measure the estimation of statistics, a smaller MSE represents a better estimation level. Therefore, the grouping results classified by a smaller MSE will make the experimental results closer to the actual strategy difference improvement and have a smaller error.

[0404] It is worth noting that the above are merely illustrative examples and are not limited to the embodiments of the present application.

[0405] In summary, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between candidate groups and causing analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, a more targeted matrix update process can be performed on the preset matrix, which is conducive to improving the accuracy of the classification results known by updating the matrix, and then using more accurate classification results to obtain more precise experimental data.

[0406] In the embodiment of the present application, the upper bound of the MSE of the statistic, M(R), is first estimated, and then the randomization scheme that makes M(R) reach the minimum value is solved by the convex optimization method; the given statistic is realized. A more conservative and smaller upper bound of the MSE is obtained, and then this upper bound is optimized, and the process of continuing to optimize the MSE of the statistic is achieved, so that the grouping process is carried out with the help of the grouping results classified by the smaller MSE, so that the experimental results are closer to the actual strategy difference improvement and the error is smaller.

[0407] Figure 8 is a structural block diagram of a group division device provided by an exemplary embodiment of the present application, such as Figure 8 As shown, the device includes the following parts:

[0408] A graph acquisition module 810 is used to acquire an entity relationship graph; the entity relationship graph is composed of a plurality of entity nodes and entity relationship edges, and the entity relationship edges are used to describe the entity association relationship between the entity nodes; wherein the plurality of entity nodes are divided into m candidate groups, where m is an integer greater than 1;

[0409] The matrix acquisition module 820 is used to obtain a first matrix and a second matrix based on the entity association relationship represented in the entity relationship graph and the m candidate groups; wherein the first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; and the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group;

[0410] An updating module 830 is used to perform matrix update on the preset matrix with the goal of reducing a statistical mean square error to obtain an updated matrix, wherein the statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix; the statistical mean square error is a mean square error expression of a statistic when performing a pre-estimation process, the statistic is a parameter of a pre-estimation process result by the first matrix and the second matrix, and the update matrix is ​​composed of update matrix elements;

[0411] The classification module 840 is used to determine the classification results corresponding to the m candidate groups respectively based on the update matrix elements, and the classification results are used to represent the category groups for classifying the candidate groups, and different category groups are used to perform different processing methods.

[0412] In an optional embodiment, the matrix acquisition module 820 is also used to divide the m candidate groups into pairs to obtain multiple group division results; determine the number of entity relationship edges between two candidate groups in each group division result to obtain the number of relationship edges corresponding to the multiple group division results; generate the first matrix using the multiple number of relationship edges as the first matrix elements; generate the second matrix based on at least one entity node respectively included in the m candidate groups and the entity association relationship.

[0413] In an optional embodiment, the matrix acquisition module 820 is also used to generate the first matrix based on the entity association relationship between the m candidate groups; for the multiple entity nodes in the entity relationship diagram, determine the number of entity nodes that have the entity association relationship with each entity node, and obtain the number of associated nodes corresponding to the multiple entity nodes; within the same candidate group, determine the sum of the number of associated nodes corresponding to at least one entity node, and obtain the sum of the number of nodes corresponding to the m candidate groups; and generate the second matrix using the sum of multiple node numbers as the second matrix elements.

[0414] In an optional embodiment, the update module 830 is also used to obtain a preset function, and the preset function is used to perform matrix update through the first matrix, the second matrix and the preset matrix; perform matrix operations on the first matrix, the second matrix and the preset matrix using the preset function to obtain the statistical mean square error; and perform matrix update on the preset matrix with the goal of minimizing the statistical mean square error to obtain the updated matrix.

[0415] In an optional embodiment, the update module 830 is also used to obtain the preset matrix and a first transposed matrix of the preset matrix; perform arcsine processing on the matrix product between the preset matrix and the first transposed matrix to obtain a first intermediate matrix; perform matrix product on the first matrix and the first intermediate matrix to obtain a second intermediate matrix; perform sum operation on the matrix elements on the matrix diagonal corresponding to the second intermediate matrix to obtain the sum of the first elements; based on the sum of the first elements and the second matrix, obtain the statistical mean square error using the preset function.

[0416] In an optional embodiment, the update module 830 is also used to perform a sum operation on the first matrix elements in the first matrix to obtain the sum of the second elements; based on the sum of the first elements and the sum of the second elements, an adjustment parameter is obtained, and the adjustment parameter is used to adjust the statistical mean square error according to the preset matrix; under the preset function, the second matrix is ​​adjusted with the adjustment parameter to obtain the statistical mean square error.

[0417] In an optional embodiment, the update module 830 is also used to obtain a second transposed matrix of the second matrix; perform matrix adjustment on the first intermediate matrix using the matrix product between the second matrix and the second transposed matrix to obtain a third intermediate matrix; perform sum operation on the matrix update elements on the matrix diagonal corresponding to the third intermediate matrix to obtain the sum of the third elements; under the preset function, adjust the sum of the third elements by the adjustment parameter to obtain the statistical mean square error.

[0418] In an optional embodiment, the update module 830 is also used to perform partial derivative operations on the preset function and the preset matrix to obtain a partial derivative matrix; based on the partial derivative matrix, the preset matrix is ​​updated through a gradient descent algorithm with the goal of minimizing the statistical mean square error to obtain a preliminary update matrix; regularization adjustment is performed on each matrix row of the preliminary update matrix to obtain the update matrix.

[0419] In an optional embodiment, the classification module 840 is also used to perform row and column adjustments on the update matrix using a preset adjustment matrix to obtain a row and column adjustment matrix, wherein the preset adjustment matrix is ​​used to adjust the m*m update matrix to an m*1 row and column adjustment matrix; based on the adjustment matrix elements in the row and column adjustment matrix, the classification results corresponding to the m candidate groups are determined respectively, and the adjustment matrix elements are the results obtained after adjusting the update matrix elements.

[0420] In an optional embodiment, the plurality of adjustment matrix elements correspond one-to-one to the m candidate groups;

[0421] The classification module 840 is also used to perform symbol analysis on the adjustment matrix elements in the row and column adjustment matrix respectively by using a symbol function to obtain symbol judgment results corresponding to multiple adjustment matrix elements respectively; in response to the symbol judgment result being a first value, the candidate group corresponding to the symbol judgment result is classified into the first group; or, in response to the symbol judgment result being a second value, the candidate group corresponding to the symbol judgment result is classified into the second group, the first value is different from the second value, the first group and the second group are different, and the first group and the second group are used to characterize groups that conduct experimental processes through different experimental methods.

[0422] In an optional embodiment, the update module 830 is also used to obtain a unit matrix as the preset matrix, the number of matrix rows of the unit matrix is ​​m, and the number of matrix columns is m; in the process of performing s+1 iterative matrix updates on the preset matrix, the statistical mean square error corresponding to the s+1th update process is obtained, and the statistical mean square error corresponding to the s+1th update process is determined based on the first matrix, the second matrix and the preset matrix after the sth update; wherein s is a positive integer less than the preset number of iterative updates; with the goal of reducing the statistical mean square error corresponding to the s+1th update process, the preset matrix after the sth update is updated s+1 times; in response to the number of iterative matrix updates on the preset matrix reaching the number of iterative updates, the updated matrix is ​​obtained.

[0423] In summary, in addition to considering the entity association relationship within the candidate group through the second matrix, the association between different candidate groups is also analyzed through the first matrix to avoid ignoring the association between candidate groups and causing analysis limitations; by minimizing the statistical mean square error corresponding to the statistic representing the estimated processing result, a more targeted matrix update process can be performed on the preset matrix, which is conducive to improving the accuracy of the classification results known by updating the matrix, and then using more accurate classification results to obtain more precise experimental data.

[0424] It should be noted that the group division device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the group division device provided in the above embodiment and the group division method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0425] Fig. 9 The schematic diagram of the structure of a server provided by an exemplary embodiment of the present application is shown. The server 900 includes a central processing unit (CPU) 901, a system memory 904 including a random access memory (RAM) 902 and a read-only memory (ROM) 903, and a system bus 905 connecting the system memory 904 and the central processing unit 901. The server 900 also includes a mass storage device 906 for storing an operating system 913, an application program 914, and other program modules 915.

[0426] The mass storage device 906 is connected to the central processing unit 901 through a mass storage controller (not shown) connected to the system bus 905. The mass storage device 906 and its associated computer readable media provide non-volatile storage for the server 900. That is, the mass storage device 906 may include a computer readable medium (not shown) such as a hard disk or a compact disc read only memory (CD-ROM) drive.

[0427] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instructions, data structures, program modules or other data. The above-mentioned system memory 904 and mass storage device 906 can be collectively referred to as memory.

[0428] According to various embodiments of the present application, the server 900 can also be connected to a remote computer on the network through a network such as the Internet. That is, the server 900 can be connected to the network 912 through the network interface unit 911 connected to the system bus 905, or the network interface unit 911 can be used to connect to other types of networks or remote computer systems (not shown).

[0429] The above-mentioned memory also includes one or more programs, and the one or more programs are stored in the memory and configured to be executed by the CPU.

[0430] An embodiment of the present application also provides a computer device, which includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the group division method provided by the above-mentioned method embodiments.

[0431] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the group division method provided by the above-mentioned method embodiments.

[0432] The embodiments of the present application also provide a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes any of the group division methods described in the above embodiments.

[0433] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A grouping method, characterized in that: The method comprises: Obtain an entity relationship graph; the entity relationship graph is composed of a plurality of entity nodes and entity relationship edges, the entity relationship edges are used to describe the entity association relationship between the entity nodes; wherein the plurality of entity nodes are divided into m candidate groups, where m is an integer greater than 1; Based on the entity association relationship represented in the entity relationship graph and the m candidate groups, a first matrix and a second matrix are obtained; wherein the first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; and the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group; The preset matrix is ​​updated with the goal of reducing the statistical mean square error to obtain an updated matrix, wherein the statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix; the statistical mean square error is a mean square error expression of the statistic when performing the estimation process, the statistic is a parameter of the estimation process result through the first matrix and the second matrix, and the updated matrix is ​​composed of updated matrix elements; Based on the update matrix elements, classification results corresponding to the m candidate groups are determined respectively, and the classification results are used to represent the category groups for classifying the candidate groups, and different category groups are used to execute different processing methods.

2. The method according to claim 1, characterized in that The first matrix and the second matrix are obtained based on the entity association relationship represented in the entity relationship graph and the m candidate groups, including: Divide the m candidate groups into pairs to obtain multiple group division results; Determine the number of entity relationship edges between two candidate groups in each group division result, and obtain the number of relationship edges corresponding to the plurality of group division results respectively; Using a plurality of relationship edge numbers as the first matrix elements, generating the first matrix; The second matrix is ​​generated based on at least one entity node respectively included in the m candidate groups and the entity association relationship.

3. The method according to claim 1, characterized in that The first matrix and the second matrix are obtained based on the entity association relationship represented in the entity relationship graph and the m candidate groups, including: Based on the entity association relationship between the m candidate groups, generating the first matrix; For the plurality of entity nodes in the entity relationship graph, determine the number of entity nodes having the entity association relationship with each entity node, and obtain the number of association nodes respectively corresponding to the plurality of entity nodes; In the same candidate group, determine the sum of the number of associated nodes corresponding to at least one entity node, and obtain the sum of the number of nodes corresponding to the m candidate groups; The second matrix is ​​generated by taking the sum of multiple node quantities as the second matrix elements.

4. The method according to any one of claims 1 to 3, characterized in that: The step of updating the preset matrix with the goal of reducing the statistical mean square error to obtain an updated matrix includes: Obtaining a preset function, where the preset function is used to perform matrix update using the first matrix, the second matrix, and the preset matrix; Performing matrix operations on the first matrix, the second matrix, and the preset matrix using the preset function to obtain the statistical mean square error; The preset matrix is ​​updated with the goal of minimizing the statistical mean square error to obtain the updated matrix.

5. The method according to claim 4, characterized in that The performing matrix operations on the first matrix, the second matrix, and the preset matrix using the preset function to obtain the statistical mean square error includes: Acquire the preset matrix and a first transposed matrix of the preset matrix; Performing arcsine processing on the matrix product between the preset matrix and the first transposed matrix to obtain a first intermediate matrix; Performing a matrix product on the first matrix and the first intermediate matrix to obtain a second intermediate matrix; Performing a sum operation on matrix elements on the matrix diagonal corresponding to the second intermediate matrix to obtain a sum of the first elements; Based on the sum of the first elements and the second matrix, the statistical mean square error is obtained using the preset function.

6. The method according to claim 5, characterized in that The obtaining the statistical mean square error based on the sum of the first elements and the second matrix by using the preset function includes: Performing a sum operation on the first matrix elements in the first matrix to obtain a sum of the second elements; Based on the sum of the first element and the sum of the second element, an adjustment parameter is obtained, where the adjustment parameter is used to obtain the statistical mean square error according to the preset matrix adjustment; Under the preset function, the second matrix is ​​adjusted with the adjustment parameter to obtain the statistical mean square error.

7. The method according to claim 6, characterized in that The step of adjusting the second matrix with the adjustment parameter under the preset function to obtain the statistical mean square error includes: Obtain a second transposed matrix of the second matrix; Performing matrix adjustment on the first intermediate matrix by using the matrix product between the second matrix and the second transposed matrix to obtain a third intermediate matrix; Performing a sum operation on matrix update elements on the matrix diagonal corresponding to the third intermediate matrix to obtain a sum of the third elements; Under the preset function, the sum of the third elements is adjusted by the adjustment parameter to obtain the statistical mean square error.

8. The method according to claim 4, characterized in that The step of updating the preset matrix with the goal of minimizing the statistical mean square error to obtain the updated matrix includes: Performing partial derivative operation on the preset function and the preset matrix to obtain a partial derivative matrix; Based on the partial derivative matrix, the preset matrix is ​​updated by a gradient descent algorithm with the goal of minimizing the statistical mean square error to obtain a preliminary updated matrix; Regularization adjustment is performed on each matrix row of the preliminary update matrix to obtain the update matrix.

9. The method according to any one of claims 1 to 3, characterized in that: The determining, based on the update matrix elements, the classification results respectively corresponding to the m candidate groups comprises: Performing row-column adjustment on the update matrix using a preset adjustment matrix to obtain a row-column adjustment matrix, wherein the preset adjustment matrix is ​​used to adjust the m*m update matrix to an m*1 row-column adjustment matrix; The classification results respectively corresponding to the m candidate groups are determined based on adjustment matrix elements in a row-column adjustment matrix, wherein the adjustment matrix elements are the results obtained after adjusting the update matrix elements.

10. The method according to claim 9, characterized in that A plurality of adjustment matrix elements correspond one-to-one to the m candidate groups; The determining the classification results respectively corresponding to the m candidate groups based on the adjustment matrix elements in the row and column adjustment matrix includes: Performing symbol analysis on the adjustment matrix elements in the row and column adjustment matrix respectively by using a symbol function to obtain symbol judgment results respectively corresponding to a plurality of adjustment matrix elements; In response to the symbol judgment result being a first value, the candidate group corresponding to the symbol judgment result is classified into the first group; or, in response to the symbol judgment result being a second value, the candidate group corresponding to the symbol judgment result is classified into the second group, the first value is different from the second value, the first group is different from the second group, and the first group and the second group are used to characterize groups that conduct experimental processes through different experimental methods.

11. The method according to any one of claims 1 to 3, characterized in that: The step of updating the preset matrix with the goal of reducing the statistical mean square error to obtain an updated matrix includes: Obtaining a unit matrix as the preset matrix, wherein the unit matrix has m matrix rows and m matrix columns; In the process of performing an s+1th iterative matrix update on the preset matrix, a statistical mean square error corresponding to the s+1th update process is obtained, where the statistical mean square error corresponding to the s+1th update process is determined based on the first matrix, the second matrix, and the preset matrix after the sth update; wherein s is a positive integer less than the preset number of iterative updates; With the goal of reducing the statistical mean square error corresponding to the s+1th updating process, performing s+1 matrix updates on the preset matrix after the sth update; In response to the number of iterative matrix updates performed on the preset matrix reaching the number of iterative updates, the updated matrix is ​​obtained.

12. A grouping device, characterized in that: The device comprises: A graph acquisition module, used to acquire an entity relationship graph; the entity relationship graph is composed of a plurality of entity nodes and entity relationship edges, and the entity relationship edges are used to describe the entity association relationship between the entity nodes; wherein the plurality of entity nodes are divided into m candidate groups, where m is an integer greater than 1; A matrix acquisition module, used to obtain a first matrix and a second matrix based on the entity association relationship represented in the entity relationship graph and the m candidate groups; wherein the first matrix elements constituting the first matrix are used to describe the entity association relationship between the candidate groups; and the second matrix elements constituting the second matrix are used to describe the entity association relationship corresponding to the entity nodes in the same candidate group; An updating module, configured to update the preset matrix with the goal of reducing a statistical mean square error to obtain an updated matrix, wherein the statistical mean square error includes a first error determined based on the first matrix and the preset matrix, and a second error determined based on the second matrix and the preset matrix; the statistical mean square error is a mean square error expression of a statistic during a pre-estimation process, the statistic is a parameter of a pre-estimation process result obtained by the first matrix and the second matrix, and the update matrix is ​​composed of update matrix elements; The classification module is used to determine the classification results corresponding to the m candidate groups respectively based on the update matrix elements, and the classification results are used to represent the category groups for classifying the candidate groups, and different category groups are used to execute different processing methods.

13. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one program, and the at least one program is loaded and executed by the processor to implement the group division method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that: The storage medium stores at least one program, and the at least one program is loaded and executed by the processor to implement the group division method as described in any one of claims 1 to 11.

15. A computer program product, characterized in that It comprises computer instructions, which, when executed by a processor, implement the grouping method as described in any one of claims 1 to 11.

Images