CN111061624A - Policy execution effect determination method and device, electronic equipment and storage medium - Google Patents

Policy execution effect determination method and device, electronic equipment and storage medium Download PDF

Info

Publication number
CN111061624A
CN111061624A CN201911097186.6A CN201911097186A CN111061624A CN 111061624 A CN111061624 A CN 111061624A CN 201911097186 A CN201911097186 A CN 201911097186A CN 111061624 A CN111061624 A CN 111061624A
Authority
CN
China
Prior art keywords
cluster
vertex
objects
exchange
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911097186.6A
Other languages
Chinese (zh)
Inventor
李星
李邵明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Sankuai Online Technology Co Ltd
Original Assignee
Beijing Sankuai Online Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Sankuai Online Technology Co Ltd filed Critical Beijing Sankuai Online Technology Co Ltd
Priority to CN201911097186.6A priority Critical patent/CN111061624A/en
Publication of CN111061624A publication Critical patent/CN111061624A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/36—Preventing errors by testing or debugging software
    • G06F11/3668—Software testing
    • G06F11/3672—Test management
    • G06F11/3688—Test management for test execution, e.g. scheduling of test suites
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9024—Graphs; Linked lists
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/906—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • Finance (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Databases & Information Systems (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Quality & Reliability (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The embodiment of the application discloses a method for determining a policy execution effect, belongs to the technical field of computers, and is beneficial to improving the accuracy of determining the policy execution effect. The method comprises the following steps: constructing an object relationship network graph according to a preset network relationship between objects; presetting M cluster identifiers to be uniformly arranged on the vertex; determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition by exchanging the cluster identifiers of part or all vertexes of the object relationship network graph; the energy of the object relationship network graph is determined according to the number of edges connecting vertexes with different cluster identifications; dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster; randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity; and respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies.

Description

Policy execution effect determination method and device, electronic equipment and storage medium
Technical Field
Embodiments of the present application relate to the field of computer technologies, and in particular, to a method and an apparatus for determining a policy execution effect, an electronic device, and a computer-readable storage medium.
Background
Traditional strategies (such as models, algorithms and the like) generally adopt A/B testing mode to evaluate the operation effect. The basic idea of taking original data as object data for example is as follows: the objects are divided into two independent groups A/B, then different strategies are used for the objects in the two groups A/B, and then the effect of the different strategies applied to the objects is evaluated. Or respectively constructing experimental group data and comparison group data, and evaluating the effects of different strategies based on the performances of the experimental group data and the comparison group data under different strategies. The method is suitable for scenes without association among different groups of data, and can randomly group the data. For data with network relationship, if the a/B groups are still obtained in a random grouping manner, the evaluation result may be inaccurate due to the association between different groups of data.
It can be seen that there is still a need for improvement in the prior art for a grouping method of data for determining policy enforcement effects and a method for determining policy enforcement effects based on the grouping method.
Disclosure of Invention
Embodiments of the present application provide a method for determining a policy enforcement effect, which improves a grouping manner of data suitable for different policies, and can improve accuracy of determining the policy enforcement effect.
In order to solve the above problem, in a first aspect, an embodiment of the present application provides a policy enforcement effect determining method, including:
constructing an object relationship network graph according to a preset network relationship between objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship;
determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition by exchanging the cluster identifiers of part or all vertexes of the object relationship network graph; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications;
dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects;
randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity;
respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies;
wherein the preset network relationship is associated with the target policy.
In a second aspect, an embodiment of the present application provides a method and an apparatus for determining a policy execution effect, where the method includes:
the object relationship network graph building module is used for building an object relationship network graph according to the preset network relationship among the objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship;
the cluster identifier exchange module is used for exchanging the cluster identifiers of part or all of the vertexes of the object relationship network graph and determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications;
the first object grouping module is used for dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects;
the second object grouping module is used for randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity;
the strategy execution effect determining module is used for respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies;
wherein the preset network relationship is associated with the target policy, and M is a natural number greater than 2.
In a third aspect, an embodiment of the present application further discloses an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the policy execution effect determination method according to the embodiment of the present application.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium on which a computer program is stored, where the program is executed by a processor to determine the steps of the policy execution effect determination method disclosed in the embodiments of the present application.
The method for determining the strategy execution effect, disclosed by the embodiment of the application, comprises the steps of constructing an object relation network graph according to a preset network relation between objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship; determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition by exchanging the cluster identifiers of part or all vertexes of the object relationship network graph; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications; dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects; randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity; respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies; the preset network relation is associated with the target strategy, and accuracy of determining strategy execution effect is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a flowchart of a policy enforcement effect determination method according to a first embodiment of the present application;
FIG. 2 is a schematic diagram of a partial object relationship network in an embodiment of the present application;
FIG. 3 is a schematic diagram of a partial object relationship network in an embodiment of the present application;
FIG. 4 is a schematic diagram of a partial object relationship network in an embodiment of the present application;
fig. 5 is a schematic structural diagram of a policy enforcement effect determining apparatus according to a second embodiment of the present application;
fig. 6 is a second schematic structural diagram of a policy enforcement effect determination apparatus according to a second embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Example one
As shown in fig. 1, a method for determining a policy execution effect disclosed in an embodiment of the present application includes: step 110 to step 150.
Step 110, constructing an object relationship network graph according to the preset network relationship among the objects.
The top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; and the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship.
When the method is implemented specifically, firstly, an object relation network graph needs to be constructed, then, the preset network relation minimization among the objects in different categories is taken as a target, and the objects are clustered based on the preset network relation to obtain M cluster objects.
Wherein M is a natural number greater than 2, and the number of objects in each cluster is the same or similar;
the object in the embodiment of the application is an object targeted by a policy to be evaluated, the type of the object is determined according to an executed policy, and the object can be a merchant or an individual user. For example, when the policy is to perform information push, take-away product marketing to the user, the object is an individual user; and when the strategy is a ranking list for the merchant, the object is the merchant.
The preset network relationship described in the embodiment of the present application is used to represent data association existing in the target policy execution process between different objects. The preset network relationship is associated with a target policy. For example, when the target policy to be evaluated is a policy for sharing a product among users, and the operation effect of the target policy is represented as a click rate of the user on the shared product, the preset network relationship may include a relationship between product sharing and sharing acceptance; when the target strategy to be evaluated is a friend assistance strategy, and the operation effect of the target strategy is embodied as the click rate of a friend on a product to be assisted, the preset network relationship may include a friend relationship.
When the method is implemented specifically, an object relationship network graph is firstly constructed based on the selected objects and used for indicating whether a preset network relationship exists between different pairs and indicating which objects have the preset network relationship. And then, carrying out balance graph clustering on the basis of the constructed object relationship network graph, clustering the selected objects to obtain a plurality of clusters, so that the number of the objects in each cluster is the same or similar, and the users in each cluster have a relatively tight preset network relationship, while the preset network relationship among the users in different clusters is weakened as much as possible.
When the method is implemented specifically, each object corresponds to one vertex, and the vertices corresponding to the objects with the preset network relationship are connected through the edges to construct the object relationship network graph.
Taking the object as an example of the user, assuming that the target policy is a sharing click policy, the preset network relationship is as follows: the sharing and receiving relationship is connected through a side for two users (for example, user a shares a takeaway order, and user B clicks the takeaway order shared by user a, and then it is determined that the sharing and receiving relationship exists between user a and user B) having the sharing and receiving relationship with the selected users as the vertexes of the object relationship network graph. According to the method, an object relation network graph is constructed.
When the object relationship network graph is constructed, a cluster identifier is randomly set for each vertex, so that the cluster identifiers of M clusters are uniformly set in the vertices in the object relationship network graph.
In some embodiments of the present application, the selected users may be randomly and evenly divided into M clusters first, so that the number of users in each cluster is the same or similar (for example, 100000 users are randomly divided into 20 clusters, each cluster includes 5000 users). Each cluster is then represented by a unique cluster identifier, for example, the numbers 1 to M identifying M clusters. And then, for each vertex in the object relationship network graph, setting the cluster identifier of the cluster where the user corresponding to the vertex is located as the cluster identifier of the vertex. For example, if the user a is in the cluster identified by the number 1, the cluster identification of the vertex vertexA corresponding to the user a is set to the number 1.
And the cluster identifier is used for indicating that the object corresponding to the vertex is divided into the cluster to which the cluster identifier belongs. As described above, the cluster of vertexA is identified by the number 1, indicating that the user a corresponding to the vertex vertexA is divided into the cluster 1 identified by the number 1.
After a cluster identifier is set for each vertex in the object relationship network graph, each vertex will include user information (e.g., a user identifier) and information of a cluster in which a user is located (e.g., a cluster identifier). For the convenience of readers to understand the present application, fig. 2 schematically shows a schematic diagram of the object relationship network graph after the cluster identifiers are set, and in an actual implementation process, as described above, if 100000 users are randomly divided into 20 clusters, each cluster includes 5000 users, then the object relationship network graph includes 100000 vertexes, where each 5000 vertex is set with the same cluster identifier, and each 100000 vertex corresponds to 20 cluster identifiers.
As shown in fig. 2: vertex 201 corresponds to user a, and the cluster identifier set for vertex 201 is 1; the vertex 202 corresponds to the user B, and the cluster identifier set for the vertex 202 is 1; vertex 203 corresponds to user C, and the cluster identifier set for vertex 203 is 1; vertex 204 corresponds to user D, and the cluster identifier set for vertex 204 is 2; vertex 205 corresponds to user E, and the cluster identifier set for vertex 205 is 2; vertex 206 corresponds to user F, and the cluster identifier set for vertex 206 is 2; vertex 207 corresponds to user G, and the cluster set for vertex 207 is identified as 1.
And 120, exchanging the cluster identifiers of part or all of the vertexes of the object relationship network graph, and determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition.
Wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications. Because the vertexes of different cluster identifiers correspond to the objects divided in different clusters, and the edges connecting the vertexes represent the preset network relationship existing among the objects, the number of the edges connecting the vertexes with different cluster identifiers represents the number of the objects divided in different clusters and having the preset network relationship, that is, the energy of the object relationship network diagram represents the number of the objects divided in different clusters and having the preset network relationship. Therefore, the smaller the energy of the object relationship network graph is, the more the number of the objects of the preset network relationship are divided into the same cluster is, and the more reasonable the cluster division is.
In some embodiments of the present application, determining a cluster identifier setting result that minimizes energy of the object relationship network graph under a preset condition by exchanging cluster identifiers of part or all of vertices of the object relationship network graph includes: performing a preset number of coarse clustering, the coarse clustering comprising: based on the determined probability, carrying out random exchange on the cluster identifications of the partial vertexes; performing fine clustering on the coarse clustering result, wherein the fine clustering comprises the following steps: and exchanging the cluster identifications of part or all of the vertexes in the object relationship network graph by a clustering method of simulated annealing, and determining a cluster identification setting result which enables the energy of the object relationship network graph to be minimum.
Firstly, using a probability scheme to randomly exchange the cluster identifiers of partial vertexes in the object relationship network graph to obtain the object relationship network graph after updating the cluster identifiers. And exchanging the cluster identification of the vertex in the object relationship network graph, wherein the object corresponding to the vertex in the cluster to which the cluster identification belongs can be exchanged. For example, for vertex 201 and vertex 204 in FIG. 2; after exchanging the cluster identifier of vertex 201 with the cluster identifier of vertex 204, the cluster identifier of vertex 201 becomes 2 and the cluster identifier of vertex 204 becomes 1, meaning that user a corresponding to vertex 201 is divided into cluster 2 and user D corresponding to vertex 204 is divided into cluster 1. By performing a cluster identity exchange, the object composition in each cluster can be adjusted.
In some embodiments of the present application, said randomly exchanging the cluster identifications of the part of the vertices based on the determined probabilities includes: the following operations are performed for each two clusters: determining exchange cluster identifier candidate vertexes corresponding to the two clusters and the number of the exchange cluster identifier candidate vertexes according to the cluster identifier distribution condition of the neighbor vertex of each vertex in the object relationship network graph; determining the vertex exchange probability corresponding to the two clusters according to the number of candidate vertexes corresponding to the exchange cluster identifier; randomly selecting a specified number of the exchange cluster identifier candidate vertexes from the exchange cluster identifier candidate vertexes corresponding to the two clusters respectively to perform cluster identifier exchange; wherein the specified number is determined according to the vertex exchange probability.
Because the objects are divided into different clusters by adopting a random division mode while the object relation network is constructed, a strong network relation may exist between the objects in every two clusters, and a weak network relation may exist between different objects in the same cluster, and the objects included in each cluster need to be adjusted through further network relation judgment, so that the network relation between the objects in different clusters is weakened, and the network relation between the objects in the same cluster is strengthened. In order to balance the number of objects in each cluster, when adjusting the objects included in each cluster, the adjustment is performed in an object exchange manner.
Next, it is first required to determine objects that may need to be swapped in every two clusters, that is, determine vertices that may need to be swapped for every two clusters (that is, swap cluster identification candidate vertices).
In some embodiments of the present application, taking the two clusters as a first cluster and a second cluster as an example, the determining the candidate vertices of the switching cluster identifier corresponding to the two clusters includes: for a first cluster in the two clusters, determining a cluster identifier of the first cluster and a cluster identifier of a neighbor vertex in the object relationship network graph, wherein the cluster identifier of the neighbor vertex comprises a vertex of a cluster identifier of a second cluster in the two clusters, as a switching cluster identifier candidate vertex corresponding to the first cluster; and for the second cluster, determining the cluster identifier of the second cluster and the cluster identifier of the neighbor vertex including the vertex of the cluster identifier of the first cluster in the object relationship network graph as a switching cluster identifier candidate vertex corresponding to the second cluster.
Taking the two clusters as cluster 1 and cluster 2 respectively for example, for cluster 1, traversing the vertices in the object relationship network, identifying the cluster as number 1, and determining that the vertex 201 is a candidate vertex of an exchange cluster identifier corresponding to cluster 1, where the cluster identifier of the neighbor vertex includes the vertex of number 2 (e.g., vertex 201 in fig. 2, the cluster identifier of the neighbor vertex is number 1, and the cluster identifier of the neighbor vertex 204 is number 2). All switch cluster identification candidate vertices for cluster 1 are determined according to this method. Similarly, for cluster 2, traversing the vertex in the object relationship network, identifying the cluster as number 2, and determining the cluster identification of the neighbor vertex including the vertex with number 1 (such as vertex 204 in fig. 2, the cluster identification of which is number 2, and the cluster identification of neighbor vertex 201 is number 1) as a candidate vertex of the exchange cluster identification corresponding to cluster 2. All switching cluster identification candidate vertices of the corresponding cluster 2 are determined according to the method.
In some embodiments of the present application, the number and sum of the candidate vertices of the switching cluster identifier corresponding to the first cluster in the object-relational network graph may be determinedThe number of candidate vertices of the switch cluster corresponding to the second cluster is identified. According to the method, the number of the switching cluster identification candidate vertexes corresponding to the first cluster and the number of the switching cluster identification candidate vertexes corresponding to the second cluster in every two clusters can be determined. If the example is given by taking every two clusters as the cluster i and the cluster j respectively, the number of the candidate vertices of the switching cluster identifier in the cluster i can be identified as mijI.e. m in cluster iijEach vertex may need to exchange cluster identification with a vertex in cluster j; identifying the number of the switching cluster identification candidate vertexes in the cluster j as mjiI.e. m in cluster jjiIndividual vertices may need to exchange cluster identities with vertices in cluster i.
Then, it can be represented by the formula xij＝min(mij,mji) And determining the minimum vertex number of the exchangeable cluster identifiers in the two clusters, and taking the minimum vertex number of the exchangeable cluster identifiers in the two clusters as the vertex exchange probability corresponding to the two clusters.
And finally, for one cluster, such as the cluster i, determining the specified number of the exchange cluster identifier candidate vertexes for carrying out cluster identifier exchange according to the vertex exchange probability, then randomly selecting the specified number of the exchange cluster identifier candidate vertexes from the exchange cluster identifier candidate vertexes with the cluster identifier i, and randomly selecting the specified number of the exchange cluster identifier candidate vertexes from the exchange cluster identifier candidate vertexes with the cluster j to exchange cluster identifiers, so as to obtain the object relationship network graph after one exchange cluster identifier operation (namely one coarse clustering) is carried out. Corresponding to the random selection of x from objects in cluster i that can be divided into clusters jijDividing each object into a cluster j, and simultaneously randomly selecting x from the objects which can be divided into the cluster i in the cluster jijAnd dividing the objects into clusters i, and adjusting the objects in the two clusters.
For every two clusters, the cluster identifier exchange of the vertex is respectively executed, and the objects included in the M clusters can be adjusted. For example, the object relationship network graph obtained after the vertex in the object relationship network graph described in fig. 2 performs the cluster identification exchange may be as shown in fig. 3.
In some embodiments of the present application, the number of clusters to be adjusted may be set, for example, cluster object adjustment is performed for M/2 clusters of M clusters.
Through preliminary adjustment, the probability that the objects in the same cluster have the preset network relationship is increased, and the probability that the objects in different clusters have the preset network relationship is reduced.
Next, fine clustering is performed on the coarse clustering results. In the fine clustering process, the minimum energy of the object relationship network graph after the cluster identification is exchanged is taken as a target, and the cluster identifications of at least part of vertexes in the object relationship network graph are exchanged through a clustering method of simulated annealing.
In some embodiments of the present application, a JABEJA (an open source clustering method) clustering method may be adopted to perform fine clustering, and further adjust the object in each cluster obtained in the previous step, so as to partition out clusters with a clean network relationship as much as possible.
In some embodiments of the present application, exchanging the cluster identifiers of some or all of the vertices in the object relationship network graph by a clustering method of simulated annealing, and determining a cluster identifier setting result that minimizes the energy of the object relationship network graph includes: identifying each pre-exchange cluster identification vertex and a pre-exchange target vertex of the pre-exchange cluster identification vertex according to the cluster identification distribution condition of the neighbor vertex of each vertex in the object relationship network graph; executing pre-exchange cluster identification on each pre-exchange cluster identification vertex and each pre-exchange target vertex of the pre-exchange cluster identification vertex respectively through a clustering method of simulated annealing, and determining a vertex needing to exchange cluster identification and a corresponding exchange target vertex, wherein the vertex needing to exchange cluster identification and the corresponding exchange target vertex are respectively: after executing the pre-exchange cluster identification, enabling the energy of the object relationship network graph to be reduced, and enabling the pre-exchange cluster identification vertex and the pre-exchange target vertex to be identified with the pre-exchange cluster identification; and executing the switching cluster identification on the determined vertex needing the switching cluster identification and the corresponding switching target vertex. At this time, the energy in the obtained object relationship network graph is the minimum.
Firstly, according to the cluster identification distribution condition of the neighbor vertex of each vertex in the object relationship network graph obtained by the rough clustering, identifying each pre-exchange cluster identification vertex and the pre-exchange target vertex of the pre-exchange cluster identification vertex. In some embodiments of the present application, if a neighboring vertex of a vertex includes a point with a different cluster identifier from the neighboring vertex, the vertex is identified as a pre-exchanged cluster identifier vertex (that is, a user with a weaker relationship exists around a certain user, and the user is considered to be divided into other clusters). Taking the object relationship network diagram shown in fig. 3 as an example, where the cluster identifier of vertex 201 is number 2, the cluster identifiers of its neighboring vertices 202, 203, 206, and 204 are number 1, and the cluster identifier of its neighboring vertex 207 is number 2, then vertex 201 is determined to be a pre-swap cluster identifier vertex, and vertices 202, 203, 206, and 204 are determined to be pre-swap target vertices of vertex 201. All pre-swap cluster identifier vertices corresponding to each cluster identifier and the pre-swap target vertex of each pre-swap cluster identifier vertex can be determined according to the method.
And then, carrying out cluster identifier pre-exchange operation on each pre-exchange cluster identifier vertex by a clustering method of simulated annealing, determining whether to execute cluster identifier exchange on the vertex according to the relation network energy change brought by the pre-exchange operation, and finally determining the vertex needing to exchange the cluster identifier and an exchange target vertex corresponding to the vertex. And exchanging the cluster identifier of the vertex, namely adjusting the cluster where the object corresponding to the vertex is located, and further clustering the object by exchanging the cluster identifier of the vertex.
In some embodiments of the present application, the determining a vertex needing a cluster exchange identifier and a corresponding exchange target vertex by performing pre-exchange cluster identifiers on each pre-exchange cluster identifier vertex and each pre-exchange target vertex of the pre-exchange cluster identifier vertex through a clustering method of simulated annealing includes: respectively executing pre-exchange cluster identification on the pre-exchange cluster identification vertex and a pre-exchange target vertex of the pre-exchange cluster identification vertex to obtain a second object relationship network graph corresponding to each pre-exchange cluster identification; calculating energy change brought by the corresponding pre-exchange cluster identifier according to a second object relationship network graph corresponding to each pre-exchange cluster identifier; responding to the energy change brought by the pre-exchange cluster identification as the maximum energy drop, and determining the pre-exchange cluster identification vertex and the pre-exchange target vertex corresponding to the corresponding pre-exchange cluster identification as the vertex needing to exchange the cluster identification and the corresponding exchange target vertex; or, in response to the energy change caused by the pre-switching cluster identifier being an energy increase, giving up executing switching cluster identifier on the pre-switching cluster identifier vertex corresponding to the corresponding pre-switching cluster identifier in the object relationship network graph.
As an example shown in fig. 3, the object relationship network graph obtained after rough clustering may determine that pre-exchange target vertices corresponding to a pre-exchange cluster identification vertex 201 are vertices 202, 203, 206, and 204, and pre-exchange target vertices corresponding to the pre-exchange cluster identification vertex 204 are vertices 201 and 205, and then perform pre-exchange cluster identification on the pre-exchange cluster identification vertex 201 and its pre-exchange target vertices 202, 203, 206, and 204, the pre-exchange cluster identification vertex 204 and its pre-exchange target vertices 201 and 205, and perform pre-exchange cluster identification once, so as to obtain a corresponding second object relationship network graph, where the second object relationship network graph is different from the object relationship network graph in that cluster identifications of partial vertices in the graph are changed. Taking the object relationship network diagram shown in fig. 3 as an example, after the pre-switch cluster identification vertex 201 and the pre-switch target vertices 202, 203, 206, and 204 thereof perform cluster identification pre-switch, second object relationship network diagrams N1, N2, N3, and N4 are obtained, and after the pre-switch cluster identification vertex 204 and the pre-switch target vertices 201 and 205 thereof perform cluster identification pre-switch, second object relationship network diagrams N5 and N6 are obtained.
And then, calculating the energy change brought by the corresponding pre-exchange cluster identifier according to a second object relationship network graph obtained by the pre-exchange cluster identifier each time. For example, the energies of the second object relationship network maps N1, N2, N3 and N4 are respectively determined by counting the number of edges connecting vertices having different cluster identifications in the second object relationship network maps N1, N2, N3 and N4, respectively.
Then, the energies of the second object relationship network graphs N1, N2, N3 and N4 are compared with the energies of the object relationship network graphs, respectively, and if the energies of the second object relationship network graphs N1, N2, N3 and N4 are increased from the energies of the object relationship network graphs, the pre-exchange cluster identification vertex 201 and the pre-exchange target vertices 202, 203, 206 and 204 corresponding to the pre-exchange cluster identification of this time are abandoned.
If the energy of the second object relationship network graphs N5 and N6 is determined to be smaller than the energy of the object relationship network graphs by counting the number of edges connecting the vertexes with different cluster identifications in the second object relationship network graphs N5 and N6 respectively, the pre-exchange cluster identification vertex 204 and the pre-exchange target vertexes 201 and 205 corresponding to the pre-exchange cluster identification of this time are determined to be used as the vertex needing to exchange the cluster identification and the corresponding exchange target vertex.
In other embodiments of the present application, if a certain pre-swap cluster identifier vertex corresponds to multiple pre-swap target vertices, multiple times of cluster identifier swap are performed on the certain pre-swap cluster identifier vertex, and a second object relationship network graph corresponding to each cluster identifier swap is obtained, a pre-swap target vertex for which the cluster identifier swap corresponding to the second object relationship network graph with the minimum energy is directed is selected as a swap target vertex of the pre-swap cluster identifier vertex. For example, if the energy of the second object relationship network map N5 is less than the energy of the second object relationship network map N6, the pre-swap cluster identifier vertex 204 and the pre-swap target vertex 201 corresponding to the pre-swap cluster identifier of this time are determined as the vertex requiring the swap cluster identifier and the corresponding swap target vertex.
And finally, executing the exchange cluster identification on the determined vertex (such as the vertex 204) needing the exchange cluster identification and the corresponding exchange target vertex (such as the vertex 201) to obtain the finely clustered object relationship network graph. For example, after performing cluster identification exchange on the vertices, the object relationship network diagram shown in fig. 3 may obtain the object relationship network diagram shown in fig. 4.
Wherein the energy of the object relational network graph is in direct proportion to the number of edges connecting vertices with different cluster identifications in the object relational network graph. For example, the energy of the object relationship network graph may be represented by the number of edges connecting vertices with different cluster identifications in the object relationship network graph. According to the cluster identifier exchange scheme, each cluster identifier exchange is conditioned on the energy reduction of the object relationship network graph, so that the energy of the object relationship network graph obtained after multiple cluster identifier exchanges can reach a smaller value.
And step 130, dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects.
In some embodiments of the present application, a cluster to which an object corresponding to each vertex belongs is determined according to a cluster identifier of each vertex in the object relationship network graph.
In the object relationship network graph obtained after the fine clustering, the number of vertexes and the edge connection relationship between the vertexes are the same as those of the object relationship network graph before the coarse clustering, and the difference is that the cluster identifications of partial vertexes are exchanged. The cluster identifier of each vertex is used to indicate the cluster to which the object corresponding to the vertex belongs, so that the objects corresponding to the vertices of the same cluster identifier are divided into the same cluster. For example, the object corresponding to vertex 202, 203, 204 with the number 1 identification would be divided into cluster 1, while the object corresponding to vertex 205 with the number 2 identification would be divided into cluster 2.
At this point, the fine clustering of the selected objects is completed, and M object clusters are obtained.
And 140, randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity.
In the obtained M clusters, the network relationship between the objects in different clusters is weakened, and the objects with the stronger network relationship are divided into the same cluster, so that each cluster can be further used as an independent individual to perform strategy execution effect evaluation or other applications influenced by the network relationship.
Taking the A/B test for executing the recommended strategy as an example, the obtained M cluster objects are randomly divided into two groups, each group comprises M/2 cluster users, and the evaluation of the execution effect of different strategies is carried out through the data of the two groups of users obtained by executing different strategies.
And 150, executing different target strategies for each group of objects respectively, and determining the execution effect of each target strategy according to the object data obtained by executing the different target strategies.
In some embodiments of the present application, different target policies are respectively executed on each group of objects, so as to determine an execution effect of each target policy according to object data obtained by executing the different target policies, where the preset network relationship is associated with the target policies, and the method includes: respectively executing different target strategies for each group of objects, and determining the object data of each object after the corresponding target strategies are executed; according to the object data, carrying out linear fitting on the influence weights of the strategy factors, the intra-group network relation factors and the extra-group network relation factors to obtain the respective influence weights of the strategy factors, the intra-group network relation factors and the extra-group network relation factors; and determining the influence of the strategy factors, the intra-group network relation factors and the extra-group network relation factors on the execution effect of each target strategy according to the influence weight obtained by linear fitting. Wherein the values of the intra-group network relationship factors represent: the proportion of the objects in the same group with the current object in the objects with the preset network relationship with the current object; the values of the out-of-group network relationship factors represent: and comparing the proportion of the objects in different groups with the current object in the objects with the preset network relationship with the current object.
Wherein the preset network relationship is associated with the target policy.
For example, by evaluating the influence of the target marketing strategy 1 and the target marketing strategy 2 on the order number, the M clusters of users obtained by clustering in the foregoing steps may be randomly divided into two groups, which are denoted as group 1 and group 2 in this embodiment. Then, applying a target marketing strategy 1 to the users in the group 1, and collecting order quantity data of each user in the group 1; target marketing strategy 2 is applied to the users in group 2 and order volume data is gathered for each user in group 2. And finally, according to the order quantity data of each user, performing linear fitting on the weights of the 3 factors, namely the strategy factor, the intra-group network relation factor and the extra-group network relation factor, which influence the order quantity data, and determining the influence of the strategy factor on the order quantity data according to the fitting result.
For example, a linear fit is performed according to the formula y α, strategy value, intra-group network relation factor value, network + γ, and hill, wherein the observed quantity includes the object data y, strategy value, intra-group network relation factor value, network, and extra-group network relation factor value, α, β, and γ as required parameter values.
The object data y, that is, order volume data, is determined according to the actual order volume of each user obtained by executing the targeted marketing strategies 1 and 2 for the two groups of users, respectively.
The tangential factor value "1" is determined according to the currently used target strategy, and if the target marketing strategy 1 corresponds to 1 and the target marketing strategy 2 corresponds to 0, the value "1" is determined for all users who are subject to the target marketing strategy 1, and the value "0" is determined for all users who are subject to the target marketing strategy 2.
And the intra-group network relation factor value network is determined according to the object relation network and the specific object group. For example, for user a, which is classified in group 1, and is executed with target marketing strategy 1, if users B and C corresponding to users B and 203 corresponding to neighbor vertex 202 of vertex 201 in the object relationship network are both classified in group 1, and user D corresponding to neighbor vertex 204 of vertex 201 is classified in group 2, the intra-group network relationship factor value network of user a is 2/3.
Similarly, the out-of-group network relationship factor value spill is determined according to the object relationship network and the specific object group. For example, for user a, which is classified in group 1, the targeted marketing strategy 1 is executed, if users B and 203 corresponding to the neighbor vertex 202 of the user corresponding to the vertex 201 in the object relationship network are both classified in group 1, and user D corresponding to the neighbor vertex 204 of the vertex 201 is classified in group 2, the value of the out-of-group network relationship factor spill of user a is 1/3.
According to the method described above, a formula containing 3 parameters can be determined based on the order data size of each user, e.g. for user A
Figure BDA0002268690020000141
Where 80 is the order quantity for order user a. all user data in groups 1 and 2, including only parameters α, β, and γ, are fitted linearly, respectively, to obtain two sets of α, β, and γ values.
The values of the two groups α, β and γ indicate that different target strategies are used to obtain different object data, and therefore, finally, the value of parameter α is taken as the weight of the target strategy's influence on the object data.
In other embodiments of the present application, taking an AB test application scenario as an example, the step of executing different target policies by each group of objects, and determining an execution effect of each target policy according to object data obtained by executing the different target policies includes: selecting different groups of objects to respectively form an experiment group and a control group, executing a first target strategy on the objects in the experiment group, and determining first object data after the first target strategy is executed; executing a second target strategy on the objects in the comparison group, and executing second object data after the second target strategy is executed; according to the first object data, carrying out linear fitting on the influence weights of a strategy factor, an intra-group network relation factor and an extra-group network relation factor, and determining respective first influence weights of the strategy factor, the intra-group network relation factor and the extra-group network relation factor; according to the second object data, carrying out linear fitting on the influence weights of the strategy factors, the intra-group network relation factors and the out-of-group network relation factors, and determining respective second influence weights of the strategy factors, the intra-group network relation factors and the out-of-group network relation factors; determining the influence of the strategy factors, the intra-group network relation factors and the extra-group network relation factors on the execution effect of the first target strategy and the second target strategy according to the first influence weight and the second influence weight of the strategy factors, the intra-group network relation factors and the extra-group network relation factors.
For example, the M clusters of users clustered in the foregoing steps may be randomly divided into two groups, which are denoted as an experimental group and a control group in this embodiment. Then, applying a first target strategy (such as estimating the order quantity of the user based on the user figure) to the users in the experimental group, and collecting order quantity data of each user in the experimental group; a second objective policy is applied to the users in the control group (e.g., estimating the number of orders for the users based on the user representation, geographic location), and order volume data is gathered for each user in the control group. And finally, according to the order quantity data of each user, performing linear fitting on the weights of the 3 factors, namely the strategy factor, the intra-group network relation factor and the extra-group network relation factor, which influence the order quantity data, and determining the influence of the strategy factor on the order quantity data according to the fitting result.
The specific method for performing the current sum based on the data of the control group users and the experimental group users is described in the related description of the linear fitting, and will not be described herein again.
And the object data y, namely order quantity data, is determined according to the actual order quantity of each user obtained by respectively executing the first target strategy and the second target strategy on the two groups of users.
The tangential factor value "1" is determined according to the currently used target policy, and assuming that the first target policy corresponds to "1" and the second target policy corresponds to "0", the value "1" is determined for all users who are executing the first target policy, and the value "0" is determined for all users who are executing the second target policy 2.
And the intra-group network relation factor value network is determined according to the object relation network and the specific object group. For example, for the user a, which is classified in the experimental group, the first objective policy is executed, and if the users B and C corresponding to the users B and 203 corresponding to the neighbor vertex 202 of the vertex 201 in the object relationship network are classified in the experimental group, and the user D corresponding to the neighbor vertex 204 of the vertex 201 is classified in the control group, the intra-group network relationship factor value network of the user a is 2/3.
Similarly, the out-of-group network relationship factor value spill is determined according to the object relationship network and the specific object group. For example, for the user a, which is classified in the experimental group, the first objective policy is executed, and if the users B and C corresponding to the users B and 203 corresponding to the neighbor vertex 202 of the vertex 201 in the object relationship network are classified in the experimental group, and the user D corresponding to the neighbor vertex 204 of the vertex 201 is classified in the control group, the value of the out-of-group network relationship factor spill of the user a is 1/3.
According to the method described above, a formula containing 3 parameters can be determined based on the order data size of each user, e.g. for user A
Figure BDA0002268690020000161
The values of β 0, β 1 and gamma are linearly fitted according to expressions which are obtained from all user data in a control group and only comprise parameters α, β and gamma, so that a group of values of α, β and gamma can be obtained, and the values of α, β and gamma are linearly fitted according to expressions which are obtained from all user data in an experimental group and only comprise parameters α, β and gamma, so that a group of values of α, β and gamma can also be obtained.
Since the users of the experimental combination control group have consistency, the values of the two groups α, β and γ represent that different target strategies are adopted to obtain different object data, and finally, the value of the parameter α is taken as the influence weight of the target strategies on the object data.
The method for determining the strategy execution effect, disclosed by the embodiment of the application, comprises the steps of constructing an object relation network graph according to a preset network relation between objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship; determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition by exchanging the cluster identifiers of part or all vertexes of the object relationship network graph; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications; dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects; randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity; respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies; the preset network relation is associated with the target strategy, and accuracy of determining strategy execution effect is improved.
According to the method and the device, the network relation graph is constructed for the objects with the network relation or the propagation effect, the objects are uniformly grouped, and the switching strategy is adopted in the subsequent clustering process, so that the consistent quantity of the objects in each cluster is ensured, and the condition that the influence weight of each cluster is different due to different quantities of the objects can be avoided. On the other hand, the clustering method based on the combination of the random method and the probability method carries out coarse clustering, realizes fast object grouping, can improve the processing efficiency of object data and save operation resources. Furthermore, fine clustering is performed based on the coarse clustering result, and the stability of object grouping is improved on the premise of fast grouping.
The method for determining the strategy execution effect disclosed in the embodiment of the application is suitable for an effect measurement scheme in a network transmission scene, and can reduce the influence of the network relation on the strategy by grouping objects with untight contact and respectively applying the objects with weak or no network relation to different strategies.
Example two
As shown in fig. 5, a device of a method for determining a policy execution effect disclosed in an embodiment of the present application includes:
an object relationship network graph constructing module 510, configured to construct an object relationship network graph according to a preset network relationship between objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship;
a cluster identifier exchanging module 520, configured to determine a cluster identifier setting result that minimizes the energy of the object relationship network graph under a preset condition by exchanging cluster identifiers of some or all vertices in the object relationship network graph; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications;
a first object grouping module 530, configured to divide the objects corresponding to the vertices with the same cluster identifier into the same cluster, so as to obtain M cluster objects;
a second object grouping module 540, configured to randomly divide the M cluster objects into at least two groups of objects with a cluster as a grouping granularity;
a policy execution effect determining module 550, configured to respectively execute different target policies on each group of objects, so as to determine an execution effect of each target policy according to object data obtained by executing the different target policies;
wherein the preset network relationship is associated with the target policy, and M is a natural number greater than 2.
In some embodiments of the present application, as shown in fig. 6, the cluster identification exchanging module 520 further includes:
a coarse clustering submodule 5201, configured to perform coarse clustering for a preset number of times, where the coarse clustering includes: based on the determined probability, carrying out random exchange on the cluster identifications of the partial vertexes;
a fine clustering submodule 5202, configured to perform fine clustering on the coarse clustering result, where the fine clustering includes: and exchanging the cluster identifications of part or all of the vertexes in the object relationship network graph by a clustering method of simulated annealing, and determining a cluster identification setting result which enables the energy of the object relationship network graph to be minimum.
In some embodiments of the present application, the coarse clustering submodule 5201 is further configured to:
the following operations are performed for each two clusters:
determining exchange cluster identifier candidate vertexes corresponding to the two clusters and the number of the exchange cluster identifier candidate vertexes according to the cluster identifier distribution condition of the neighbor vertex of each vertex in the object relationship network graph;
determining the vertex exchange probability corresponding to the two clusters according to the number of candidate vertexes corresponding to the exchange cluster identifier;
randomly selecting a specified number of the exchange cluster identifier candidate vertexes from the exchange cluster identifier candidate vertexes corresponding to the two clusters respectively to perform cluster identifier exchange; wherein the specified number is determined according to the vertex exchange probability.
In some embodiments of the present application, the determining candidate vertices of switching cluster identifiers corresponding to the two clusters includes:
for a first cluster in the two clusters, determining a cluster identifier of the first cluster and a cluster identifier of a neighbor vertex in the object relationship network graph, wherein the cluster identifier of the neighbor vertex comprises a vertex of a cluster identifier of a second cluster in the two clusters, as a switching cluster identifier candidate vertex corresponding to the first cluster;
and for the second cluster, determining the cluster identifier of the second cluster and the cluster identifier of the neighbor vertex including the vertex of the cluster identifier of the first cluster in the object relationship network graph as a switching cluster identifier candidate vertex corresponding to the second cluster.
In some embodiments of the present application, the fine clustering submodule 5202 is further configured to:
identifying each pre-exchange cluster identification vertex and a pre-exchange target vertex of the pre-exchange cluster identification vertex according to the cluster identification distribution condition of the neighbor vertex of each vertex in the object relationship network graph;
respectively executing pre-exchange cluster identification on each pre-exchange cluster identification vertex and each pre-exchange target vertex of the pre-exchange cluster identification vertex by a clustering method of simulated annealing, and determining a vertex needing to exchange cluster identification and a corresponding exchange target vertex; wherein, the vertex needing to exchange the cluster identifier and the corresponding exchange target vertex are respectively: after executing the pre-exchange cluster identification, enabling the energy of the object relationship network graph to be reduced, and enabling the pre-exchange cluster identification vertex and the pre-exchange target vertex to be identified with the pre-exchange cluster identification;
and executing the switching cluster identification on the determined vertex needing the switching cluster identification and the corresponding switching target vertex.
In some embodiments of the present application, the policy enforcement effect determining module 550 is further configured to:
respectively executing different target strategies for each group of objects, and determining the object data of each object after the corresponding target strategies are executed;
according to the object data, carrying out linear fitting on the influence weights of the strategy factors, the intra-group network relation factors and the extra-group network relation factors to obtain the respective influence weights of the strategy factors, the intra-group network relation factors and the extra-group network relation factors;
and determining the influence of the strategy factors, the intra-group network relation factors and the extra-group network relation factors on the execution effect of each target strategy according to the influence weight obtained by linear fitting.
In some embodiments of the present application, the policy enforcement effect determining module 550 is further configured to:
selecting different groups of objects to respectively form an experiment group and a control group, executing a first target strategy on the objects in the experiment group, and determining first object data after the first target strategy is executed; executing a second target strategy on the objects in the comparison group, and executing second object data after the second target strategy is executed;
according to the first object data, carrying out linear fitting on the influence weights of a strategy factor, an intra-group network relation factor and an extra-group network relation factor, and determining respective first influence weights of the strategy factor, the intra-group network relation factor and the extra-group network relation factor; according to the second object data, carrying out linear fitting on the influence weights of the strategy factors, the intra-group network relation factors and the out-of-group network relation factors, and determining respective second influence weights of the strategy factors, the intra-group network relation factors and the out-of-group network relation factors;
determining the influence of the strategy factors, the intra-group network relation factors and the extra-group network relation factors on the execution effect of the first target strategy and the second target strategy according to the first influence weight and the second influence weight of the strategy factors, the intra-group network relation factors and the extra-group network relation factors.
The device for determining a policy enforcement effect disclosed in the embodiment of the present application is used to implement each step of the method for determining a policy enforcement effect described in the first embodiment of the present application, and specific implementation manners of each module of the device refer to the corresponding step, which is not described herein again.
The method and the device for determining the strategy execution effect, disclosed by the embodiment of the application, construct an object relationship network graph according to the preset network relationship among objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship; determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition by exchanging the cluster identifiers of part or all vertexes of the object relationship network graph; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications; dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects; randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity; respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies; the preset network relation is associated with the target strategy, and accuracy of determining strategy execution effect is improved.
According to the method and the device, the network relation graph is constructed for the objects with the network relation or the propagation effect, the objects are uniformly grouped, and the switching strategy is adopted in the subsequent clustering process, so that the consistent quantity of the objects in each cluster is ensured, and the condition that the influence weight of each cluster is different due to different quantities of the objects can be avoided. On the other hand, the clustering method based on the combination of the random method and the probability method carries out coarse clustering, realizes fast object grouping, can improve the processing efficiency of object data and save operation resources. Furthermore, fine clustering is performed based on the coarse clustering result, and the stability of object grouping is improved on the premise of fast grouping.
The device for determining the strategy execution effect disclosed in the embodiment of the application is suitable for an effect measurement scheme in a network transmission scene, and can reduce the influence of the network relation on the strategy by grouping objects with untight contact and respectively applying the objects with weak or no network relation to different strategies.
Correspondingly, the application also discloses an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor implements the policy execution effect determination method according to the first embodiment of the application when executing the computer program. The electronic device can be a PC, a mobile terminal, a personal digital assistant, a tablet computer and the like.
The present application also discloses a computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of the policy enforcement effect determination method according to the first embodiment of the present application.
The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.
The policy enforcement effect determining method and device provided by the present application are introduced in detail above, and a specific example is applied in the text to explain the principle and the implementation of the present application, and the description of the above embodiment is only used to help understanding the method and the core idea of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Claims (11)

1. A method for determining the effect of policy enforcement, comprising:
constructing an object relationship network graph according to a preset network relationship between objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship;
determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition by exchanging the cluster identifiers of part or all vertexes of the object relationship network graph; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications;
dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects;
randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity;
respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies;
wherein the preset network relationship is associated with the target policy, and M is a natural number greater than 2.
2. The method according to claim 1, wherein the step of determining a cluster identifier setting result that minimizes the energy of the object relationship network graph under a preset condition by exchanging cluster identifiers of some or all vertices of the object relationship network graph comprises:
performing a preset number of coarse clustering, the coarse clustering comprising: based on the determined probability, carrying out random exchange on the cluster identifications of the partial vertexes;
performing fine clustering on the coarse clustering result, wherein the fine clustering comprises the following steps: and exchanging the cluster identifications of part or all of the vertexes in the object relationship network graph by a clustering method of simulated annealing, and determining a cluster identification setting result which enables the energy of the object relationship network graph to be minimum.
3. The method of claim 2, wherein the step of randomly swapping the cluster identifications of the vertices based on the determined probabilities comprises:
the following operations are performed for each two clusters:
determining exchange cluster identifier candidate vertexes corresponding to the two clusters and the number of the exchange cluster identifier candidate vertexes according to the cluster identifier distribution condition of the neighbor vertex of each vertex in the object relationship network graph;
determining the vertex exchange probability corresponding to the two clusters according to the number of candidate vertexes corresponding to the exchange cluster identifier;
randomly selecting a specified number of the exchange cluster identifier candidate vertexes from the exchange cluster identifier candidate vertexes corresponding to the two clusters respectively to perform cluster identifier exchange; wherein the specified number is determined according to the vertex exchange probability.
4. The method of claim 3, wherein the step of determining the swap cluster identification candidate vertices for the two clusters comprises:
for a first cluster in the two clusters, determining a cluster identifier of the first cluster and a cluster identifier of a neighbor vertex in the object relationship network graph, wherein the cluster identifier of the neighbor vertex comprises a vertex of a cluster identifier of a second cluster in the two clusters, as a switching cluster identifier candidate vertex corresponding to the first cluster;
and for the second cluster, determining the cluster identifier of the second cluster and the cluster identifier of the neighbor vertex including the vertex of the cluster identifier of the first cluster in the object relationship network graph as a switching cluster identifier candidate vertex corresponding to the second cluster.
5. The method according to claim 2, wherein the step of exchanging the cluster identifiers of some or all of the vertices in the object relationship network graph by a clustering method of simulated annealing to determine the cluster identifier setting result that minimizes the energy of the object relationship network graph comprises:
identifying each pre-exchange cluster identification vertex and a pre-exchange target vertex of the pre-exchange cluster identification vertex according to the cluster identification distribution condition of the neighbor vertex of each vertex in the object relationship network graph;
respectively executing pre-exchange cluster identification on each pre-exchange cluster identification vertex and each pre-exchange target vertex of the pre-exchange cluster identification vertex by a clustering method of simulated annealing, and determining a vertex needing to exchange cluster identification and a corresponding exchange target vertex; wherein, the vertex needing to exchange the cluster identifier and the corresponding exchange target vertex are respectively: after executing the pre-exchange cluster identification, enabling the energy of the object relationship network graph to be reduced, and enabling the pre-exchange cluster identification vertex and the pre-exchange target vertex to be identified with the pre-exchange cluster identification;
and executing the switching cluster identification on the determined vertex needing the switching cluster identification and the corresponding switching target vertex.
6. The method according to claim 1, wherein the step of executing different target policies for each group of objects respectively to determine the execution effect of each target policy according to the object data obtained by executing the different target policies comprises:
respectively executing different target strategies for each group of objects, and determining the object data of each object after the corresponding target strategies are executed;
according to the object data, carrying out linear fitting on the influence weights of the strategy factors, the intra-group network relation factors and the extra-group network relation factors to obtain the respective influence weights of the strategy factors, the intra-group network relation factors and the extra-group network relation factors;
and determining the influence of the strategy factors, the intra-group network relation factors and the extra-group network relation factors on the execution effect of each target strategy according to the influence weight obtained by linear fitting.
7. The method according to claim 1, wherein the step of executing different target policies for each group of objects respectively to determine the execution effect of each target policy according to the object data obtained by executing the different target policies comprises:
selecting different groups of objects to respectively form an experiment group and a control group, executing a first target strategy on the objects in the experiment group, and determining first object data after the first target strategy is executed; executing a second target strategy on the objects in the comparison group, and executing second object data after the second target strategy is executed;
according to the first object data, carrying out linear fitting on the influence weights of a strategy factor, an intra-group network relation factor and an extra-group network relation factor, and determining respective first influence weights of the strategy factor, the intra-group network relation factor and the extra-group network relation factor; according to the second object data, carrying out linear fitting on the influence weights of the strategy factors, the intra-group network relation factors and the out-of-group network relation factors, and determining respective second influence weights of the strategy factors, the intra-group network relation factors and the out-of-group network relation factors;
determining the influence of the strategy factors, the intra-group network relation factors and the extra-group network relation factors on the execution effect of the first target strategy and the second target strategy according to the first influence weight and the second influence weight of the strategy factors, the intra-group network relation factors and the extra-group network relation factors.
8. A method and device for determining the effect of policy execution are characterized by comprising the following steps:
the object relationship network graph building module is used for building an object relationship network graph according to the preset network relationship among the objects; the top points in the object relation network graph correspond to the objects one by one, and M cluster identifiers are preset and uniformly arranged on the top points; the edges in the object relationship network graph are connected with the vertexes corresponding to the objects with the preset network relationship;
the cluster identifier exchange module is used for exchanging the cluster identifiers of part or all of the vertexes of the object relationship network graph and determining a cluster identifier setting result which enables the energy of the object relationship network graph to be minimum under a preset condition; wherein the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifications;
the first object grouping module is used for dividing the objects corresponding to the vertexes provided with the same cluster identifier into the same cluster to obtain M cluster objects;
the second object grouping module is used for randomly dividing the M cluster objects into at least two groups of objects by taking the clusters as grouping granularity;
the strategy execution effect determining module is used for respectively executing different target strategies for each group of objects so as to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies;
wherein the preset network relationship is associated with the target policy, and M is a natural number greater than 2.
9. The apparatus of claim 8, wherein the cluster identity switching module further comprises:
the coarse clustering submodule is used for executing coarse clustering of preset times, and the coarse clustering comprises the following steps: based on the determined probability, carrying out random exchange on the cluster identifications of the partial vertexes;
and the fine clustering submodule is used for performing fine clustering on the coarse clustering result, and the fine clustering comprises the following steps: and exchanging the cluster identifications of part or all of the vertexes in the object relationship network graph by a clustering method of simulated annealing, and determining a cluster identification setting result which enables the energy of the object relationship network graph to be minimum.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the policy enforcement effect determination method according to any one of claims 1 to 7 when executing the computer program.
11. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the policy enforcement effect determination method according to any one of claims 1 to 7.
CN201911097186.6A 2019-11-11 2019-11-11 Policy execution effect determination method and device, electronic equipment and storage medium Pending CN111061624A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911097186.6A CN111061624A (en) 2019-11-11 2019-11-11 Policy execution effect determination method and device, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911097186.6A CN111061624A (en) 2019-11-11 2019-11-11 Policy execution effect determination method and device, electronic equipment and storage medium

Publications (1)

Publication Number Publication Date
CN111061624A true CN111061624A (en) 2020-04-24

Family

ID=70298540

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911097186.6A Pending CN111061624A (en) 2019-11-11 2019-11-11 Policy execution effect determination method and device, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN111061624A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113792089A (en) * 2021-09-16 2021-12-14 平安银行股份有限公司 Illegal behavior detection method, device, equipment and medium based on artificial intelligence
CN116149284A (en) * 2023-04-23 2023-05-23 广东麦可瑞化工科技有限公司 Papermaking defoamer production control method and system
WO2023191717A1 (en) * 2022-04-01 2023-10-05 Grabtaxi Holdings Pte. Ltd. Method and system for adaptively dividing graph network into subnetworks
CN117056239A (en) * 2023-10-11 2023-11-14 腾讯科技（深圳）有限公司 Method, device, equipment and storage medium for determining test function using characteristics

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113792089A (en) * 2021-09-16 2021-12-14 平安银行股份有限公司 Illegal behavior detection method, device, equipment and medium based on artificial intelligence
CN113792089B (en) * 2021-09-16 2024-03-22 平安银行股份有限公司 Illegal behavior detection method, device, equipment and medium based on artificial intelligence
WO2023191717A1 (en) * 2022-04-01 2023-10-05 Grabtaxi Holdings Pte. Ltd. Method and system for adaptively dividing graph network into subnetworks
CN116149284A (en) * 2023-04-23 2023-05-23 广东麦可瑞化工科技有限公司 Papermaking defoamer production control method and system
CN116149284B (en) * 2023-04-23 2023-08-04 广东麦可瑞化工科技有限公司 Papermaking defoamer production control method and system
CN117056239A (en) * 2023-10-11 2023-11-14 腾讯科技（深圳）有限公司 Method, device, equipment and storage medium for determining test function using characteristics
CN117056239B (en) * 2023-10-11 2024-01-30 腾讯科技（深圳）有限公司 Method, device, equipment and storage medium for determining test function using characteristics

Similar Documents

Publication Publication Date Title
CN111061624A (en) Policy execution effect determination method and device, electronic equipment and storage medium
WO2019205325A1 (en) Method for determining risk level of user, terminal device, and computer-readable storage medium
TW201835789A (en) Method and device for constructing scoring model and evaluating user credit
CN108197285A (en) A kind of data recommendation method and device
CN106709625A (en) Electricity market demand response planning evaluation method
CN110322295B (en) Relationship strength determination method and system, server and computer readable medium
CN106326297B (en) Application program recommendation method and device
CN114168761B (en) Multimedia data pushing method and device, electronic equipment and storage medium
CN108985954A (en) A kind of method and relevant device of incidence relation that establishing each mark
CN110362563A (en) The processing method and processing device of tables of data, storage medium, electronic device
CN111797320A (en) Data processing method, device, equipment and storage medium
CN117172633B (en) Manufacturing service subgraph simulation method and system for industrial Internet platform
CN109800477B (en) Lithium ion battery three-factor inconsistency prediction method and device
CN111797319A (en) Recommendation method, device, equipment and storage medium
CN103473437B (en) Cloud service platform selection method based on fuzzy ideal target analysis
CN106611021B (en) Data processing method and equipment
CN107609570A (en) Micro- video popularity Forecasting Methodology based on attributive classification and various visual angles Fusion Features
CN112598132A (en) Model training method and device, storage medium and electronic device
CN113194107B (en) Internet-based regional characteristic addressing method and device
CN115982634A (en) Application program classification method and device, electronic equipment and computer program product
CN113657525A (en) KMeans-based cross-feature federated clustering method and related equipment
CN112085328A (en) Risk assessment method, system, electronic device and storage medium
CN106301880A (en) One determines that cyberrelationship degree of stability, Internet service recommend method and apparatus
CN110569475A (en) Evaluation method, device, equipment and storage medium for netizen influence
CN112288867A (en) Smart building three-dimensional model rendering method and smart building system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination