Policy execution effect determination method and apparatus, electronic device, and storage medium

Abstract

The embodiment of the application discloses a strategy execution effect determination method, belongs to the computer technical field, and helps to improve the accuracy of strategy execution effect determination. The method comprises the following steps: constructing an object relationship network graph according to a preset network relationship between objects; presetting M cluster identifiers on the vertices; exchanging the cluster identifiers of part or all vertices in the object relationship network graph to determine a cluster identifier setting result that minimizes the energy of the object relationship network graph under a preset condition; the energy of the object relationship network graph is determined according to the number of edges connecting vertices with different cluster identifiers; the objects corresponding to the vertices with the same cluster identifier are divided into the same cluster; the M cluster objects are randomly divided into at least two groups of objects in the cluster grouping granularity; different target strategies are respectively executed on each group of objects to determine the execution effect of each target strategy according to object data obtained by executing the different target strategies.

Description

Technical Field

[0001] The embodiments of this application relate to the field of computer technology, and in particular to a method, apparatus, electronic device, and computer-readable storage medium for determining the effect of strategy execution. Background Technology

[0002] Traditional strategy (such as model, algorithm, etc.) performance evaluation typically employs A / B testing. Taking raw data as the object data as an example, the basic idea is to divide the objects into two independent groups, A and B, then apply different strategies to the objects in groups A and B, and finally evaluate the effect of each strategy. Alternatively, experimental and control group data can be constructed separately, and the effectiveness of different strategies can be evaluated based on the performance of the experimental and control group data under different strategies. These methods are suitable for scenarios where different groups of data are not correlated, allowing for random grouping of the data. However, for data with network relationships, if A / B groups are still obtained through random grouping, the correlation between different groups of data will lead to inaccurate evaluation results.

[0003] It is evident that the existing methods for grouping data used to determine the effectiveness of policy execution, and the methods for determining the effectiveness of policy execution based on the grouped data obtained from these grouping methods, still need improvement. Summary of the Invention

[0004] The embodiments of this application provide a method for determining the effect of strategy execution, which improves the grouping method of data applicable to different strategies and can improve the accuracy of determining the effect of strategy execution.

[0005] To address the aforementioned problems, in a first aspect, embodiments of this application provide a method for determining the effect of strategy execution, comprising:

[0006] An object relationship network graph is constructed based on the preset network relationships between objects; wherein, the vertices in the object relationship network graph correspond one-to-one with the objects, and M preset cluster identifiers are evenly distributed on the vertices; the edges in the object relationship network graph connect the vertices corresponding to the objects that have the preset network relationships;

[0007] By exchanging the cluster identifiers of some or all vertices in the object relationship network graph, the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions is determined; wherein, the energy of the object relationship network graph is determined based on the number of edges connecting vertices with different cluster identifiers;

[0008] Objects corresponding to vertices with the same cluster identifier are grouped into the same cluster to obtain M-cluster objects;

[0009] Using clusters as the grouping granularity, the M cluster objects are randomly divided into at least two groups of objects;

[0010] Different target strategies are executed for each group of objects, and the execution effect of each target strategy is determined based on the object data obtained from executing the different target strategies.

[0011] The preset network relationship is associated with the target strategy.

[0012] Secondly, embodiments of this application provide an apparatus for determining the effect of strategy execution, comprising:

[0013] An object relationship network graph construction module is used to construct an object relationship network graph based on a preset network relationship between objects; wherein, the vertices in the object relationship network graph correspond one-to-one with the objects, and M preset cluster identifiers are evenly distributed on the vertices; the edges in the object relationship network graph connect the vertices corresponding to the objects that have the preset network relationship;

[0014] The cluster identifier exchange module is used to determine the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions by exchanging the cluster identifiers of some or all vertices in the object relationship network graph; wherein, the energy of the object relationship network graph is determined based on the number of edges connecting vertices with different cluster identifiers;

[0015] The first object grouping module is used to group objects corresponding to the vertices that have the same cluster identifier into the same cluster to obtain M cluster objects;

[0016] The second object grouping module is used to randomly divide the M cluster objects into at least two groups of objects with clusters as the grouping granularity.

[0017] The strategy execution effect determination module is used to execute different target strategies on each group of objects, so as to determine the execution effect of each target strategy based on the object data obtained by executing the different target strategies;

[0018] Wherein, the preset network relationship is associated with the target strategy, and M is a natural number greater than 2.

[0019] Thirdly, embodiments of this application also disclose an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the strategy execution effect determination method described in the embodiments of this application.

[0020] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, represents the steps of the strategy execution effect determination method disclosed in embodiments of this application.

[0021] The method for determining the strategy execution effect disclosed in the embodiments of this application constructs an object relationship network graph based on a preset network relationship between objects. In this graph, each vertex corresponds to an object, and M preset cluster identifiers are evenly distributed on the vertices. Edges in the object relationship network graph connect the vertices corresponding to objects with the preset network relationship. By exchanging the cluster identifiers of some or all vertices in the object relationship network graph, the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions is determined. The energy of the object relationship network graph is determined based on the number of edges connecting vertices with different cluster identifiers. Objects corresponding to vertices with the same cluster identifier are grouped into the same cluster, resulting in M ​​cluster objects. Using clusters as the grouping granularity, the M cluster objects are randomly divided into at least two groups of objects. Different target strategies are executed on each group of objects, and the execution effect of each target strategy is determined based on the object data obtained from executing the different target strategies. The preset network relationship is associated with the target strategy, which helps improve the accuracy of determining the strategy execution effect. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the strategy execution effect determination method in Embodiment 1 of this application;

[0024] Figure 2 This is a schematic diagram of a partial object relationship network in Embodiment 1 of this application;

[0025] Figure 3 This is a schematic diagram of a partial object relationship network in Embodiment 1 of this application;

[0026] Figure 4 This is a schematic diagram of a partial object relationship network in Embodiment 1 of this application;

[0027] Figure 5 This is one of the structural schematic diagrams of the strategy execution effect determination device in Embodiment 2 of this application;

[0028] Figure 6 This is the second schematic diagram of the strategy execution effect determination device in Embodiment 2 of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] Example 1

[0031] The embodiments of this application disclose a method for determining the effect of strategy execution, such as... Figure 1 As shown, the method includes steps 110 to 150.

[0032] Step 110: Construct an object relationship network diagram based on the preset network relationships between objects.

[0033] In this context, each vertex in the object relationship network graph corresponds one-to-one with an object, and M preset cluster identifiers are evenly distributed on the vertices; the edges in the object relationship network graph connect the vertices corresponding to objects that have the preset network relationship.

[0034] In the specific implementation of this application, it is first necessary to construct an object relationship network graph. Then, with the goal of minimizing the preset network relationship between objects in different categories, the objects are clustered based on the preset network relationship to obtain M cluster objects.

[0035] Where M is a natural number greater than 2, and the number of objects in each cluster is the same or similar;

[0036] The object described in this embodiment is the target of the strategy to be evaluated. The object type is determined according to the executed strategy and can be a merchant or an individual user. For example, when the strategy is to push information to users or market takeout products, the object is an individual user; when the strategy is to rank merchants, the object is a merchant.

[0037] The preset network relationships described in this application embodiment are used to represent the data associations between different objects during the execution of the target strategy. The preset network relationships are associated with the target strategy. For example, when the target strategy to be evaluated is a strategy for users to share products, and the effect of the target strategy is reflected in the click-through rate of users on the shared products, then the preset network relationship can include the relationship between product sharing and accepting sharing; when the target strategy to be evaluated is a strategy for friends to help each other, and the effect of the target strategy is reflected in the click-through rate of friends on the products they help with, then the preset network relationship can include friend relationships.

[0038] In its specific implementation, this application first requires constructing an object relationship network graph based on the selected objects. This graph represents whether there are pre-defined network relationships between different pairs and which objects have pre-defined network relationships. Then, balanced graph clustering is performed based on the constructed object relationship network graph to cluster the selected objects into multiple clusters. This ensures that the number of objects in each cluster is the same or similar, and that users within each cluster have relatively close pre-defined network relationships, while the pre-defined network relationships between users in different clusters are weakened as much as possible.

[0039] In a specific implementation of this application, each object corresponds to a vertex, and the vertices corresponding to objects with preset network relationships are connected by edges to construct an object relationship network graph.

[0040] Taking users as an example, assuming the target strategy is a share-click strategy, the preset network relationship is: sharing and accepting sharing. Using selected users as vertices of the object relationship network graph, for two users with a sharing and accepting sharing relationship (for example, user A shares a food delivery order, and user B clicks on the food delivery order shared by user A, then a sharing and accepting sharing relationship exists between user A and user B), they are connected by an edge. The object relationship network graph is constructed using this method.

[0041] When constructing the object relationship network graph, a cluster identifier is randomly assigned to each vertex, so that the cluster identifiers of the M clusters are evenly distributed among the vertices in the object relationship network graph.

[0042] In some embodiments of this application, the selected users can first be randomly and evenly divided into M clusters, such that the number of users in each cluster is the same or similar (for example, 100,000 users can be randomly divided into 20 clusters, with each cluster containing 5,000 users). Then, each cluster is represented by a unique cluster identifier, for example, using numbers 1 to M to identify the M clusters. Next, for each vertex in the object relation network graph, the cluster identifier of the cluster to which the user corresponding to that vertex belongs is set as the cluster identifier of that vertex. For example, if user A is in the cluster identified by number 1, then the cluster identifier of the vertex vertex A corresponding to user A is set to the number 1.

[0043] The cluster identifier is used to indicate that the object corresponding to the vertex is assigned to the cluster to which the cluster identifier belongs. As mentioned above, the cluster identifier of vertexA is the number 1, indicating that user A corresponding to vertexA is assigned to cluster 1 identified by the number 1.

[0044] After setting a cluster identifier for each vertex in the object relationship network graph, each vertex will include user information (such as user identifier) ​​and information about the user's cluster (such as cluster identifier). To facilitate the reader's understanding of this application, Figure 2The diagram illustrates the object relationship network graph after cluster identifiers are set. In actual implementation, as mentioned above, if 100,000 users are randomly divided into 20 clusters, with each cluster containing 5,000 users, then the object relationship network graph will contain 100,000 vertices. Each set of 5,000 vertices will have the same cluster identifier, and the 100,000 vertices will correspond to 20 cluster identifiers.

[0045] like Figure 2 As shown: Vertex 201 corresponds to user A, and the cluster identifier for vertex 201 is set to 1; Vertex 202 corresponds to user B, and the cluster identifier for vertex 202 is set to 1; Vertex 203 corresponds to user C, and the cluster identifier for vertex 203 is set to 1; Vertex 204 corresponds to user D, and the cluster identifier for vertex 204 is set to 2; Vertex 205 corresponds to user E, and the cluster identifier for vertex 205 is set to 2; Vertex 206 corresponds to user F, and the cluster identifier for vertex 206 is set to 2; Vertex 207 corresponds to user G, and the cluster identifier for vertex 207 is set to 1.

[0046] Step 120: By exchanging the cluster identifiers of some or all vertices in the object relationship network graph, determine the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions.

[0047] The energy of the object relationship network graph is determined by the number of edges connecting vertices with different cluster identifiers. Since vertices with different cluster identifiers correspond to objects in different clusters, and the edges connecting these vertices represent the pre-defined network relationships between the objects, the number of edges connecting vertices with different cluster identifiers represents the number of objects in different clusters that have these pre-defined network relationships. In other words, the energy of the object relationship network graph represents the number of objects in different clusters that have these pre-defined network relationships. Therefore, the lower the energy of the object relationship network graph, the more objects with pre-defined network relationships are grouped into the same cluster, resulting in a more reasonable cluster division.

[0048] In some embodiments of this application, the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions is determined by exchanging the cluster identifiers of some or all vertices in the object relationship network graph. This includes: performing a preset number of coarse clustering operations, wherein the coarse clustering includes: randomly exchanging the cluster identifiers of some vertices based on a determined probability; and performing fine clustering on the coarse clustering result, wherein the fine clustering includes: exchanging the cluster identifiers of some or all vertices in the object relationship network graph using a simulated annealing clustering method to determine the cluster identifier setting result that minimizes the energy of the object relationship network graph.

[0049] First, using a probabilistic scheme, the cluster identifiers of some vertices in the object relationship network graph are randomly swapped to obtain an updated object relationship network graph. Swapping the cluster identifiers of vertices in the object relationship network graph can swap the objects within the cluster to which the cluster identifier belongs that correspond to the vertex. For example, for... Figure 2 Consider vertices 201 and 204. After swapping the cluster identifiers of vertex 201 and 204, the cluster identifier of vertex 201 becomes 2, and the cluster identifier of vertex 204 becomes 1. This means that user A, corresponding to vertex 201, is assigned to cluster 2, while user D, corresponding to vertex 204, is assigned to cluster 1. By swapping cluster identifiers, the object composition within each cluster can be adjusted.

[0050] In some embodiments of this application, the step of randomly exchanging the cluster identifiers of some vertices based on a determined probability includes: performing the following operations for each pair of clusters: determining candidate vertices for exchanging cluster identifiers and the number of such candidate vertices based on the distribution of cluster identifiers of the neighboring vertices of each vertex in the object relation network graph; determining the vertex exchange probability corresponding to the two clusters based on the number of such candidate vertices; and randomly selecting a specified number of such candidate vertices from the candidate vertices for exchanging cluster identifiers of the two clusters for cluster identifier exchange; wherein the specified number is determined based on the vertex exchange probability.

[0051] Because objects are randomly assigned to different clusters during the construction of the object relationship network, there may be strong network relationships between objects in any two clusters, and weak network relationships between different objects within the same cluster. Further network relationship assessment is needed to adjust the objects included in each cluster, weakening the network relationships between objects in different clusters while strengthening the network relationships between objects within the same cluster. To balance the number of objects in each cluster, object swapping is used when adjusting the objects included in each cluster.

[0052] Next, we need to determine the objects that may need to be swapped between each pair of clusters, that is, to determine the vertices that may need to swap cluster identifiers between each pair of clusters (i.e., candidate vertices for swapping cluster identifiers).

[0053] In some embodiments of this application, taking the two clusters as a first cluster and a second cluster as examples, the step of determining the candidate vertices for the exchange cluster identifiers corresponding to the two clusters includes: for the first cluster, determining the vertices in the object relation network graph that have the cluster identifier of the first cluster and whose neighboring vertices have the cluster identifier of the second cluster as candidate vertices for the exchange cluster identifiers corresponding to the first cluster; for the second cluster, determining the vertices in the object relation network graph that have the cluster identifier of the second cluster and whose neighboring vertices have the cluster identifier of the first cluster as candidate vertices for the exchange cluster identifiers corresponding to the second cluster.

[0054] Taking the two clusters as cluster 1 and cluster 2 as examples, for cluster 1, traverse the vertices in the object relation network, identify the cluster as number 1, and the cluster identifiers of the neighboring vertices include vertices with number 2 (such as...). Figure 2 Vertex 201 in the network has a cluster identifier of number 1, and its neighboring vertex 204 has a cluster identifier of number 2. Vertex 201 is determined to be a candidate vertex for swapping cluster identifiers for the corresponding cluster 1. All candidate vertices for swapping cluster identifiers for the corresponding cluster 1 are determined in this way. Similarly, for cluster 2, vertices in the object relation network are traversed, and those vertices with a cluster identifier of number 2 and whose neighboring vertices include the cluster identifier of number 1 (such as...) are identified. Figure 2 Vertex 204 in the cluster (whose cluster identifier is the number 2, and whose neighboring vertex 201 has a cluster identifier of the number 1) is identified as a candidate vertex for swapping cluster identifiers of the corresponding cluster 2. All candidate vertices for swapping cluster identifiers of the corresponding cluster 2 are determined in this way.

[0055] In some embodiments of this application, the number of candidate vertices for exchange cluster identification corresponding to the first cluster and the number of candidate vertices for exchange cluster identification corresponding to the second cluster in the object relationship network graph can be determined. Following this method, the number of candidate vertices for exchange cluster identification corresponding to the first cluster and the number of candidate vertices for exchange cluster identification corresponding to the second cluster in every two clusters can be determined. Taking cluster i and cluster j as examples, the number of candidate vertices for exchange cluster identification in cluster i can be denoted as m. ij That is, there is m in cluster i ij A vertex may need to exchange its cluster identifier with a vertex in cluster j; the number of candidate vertices in cluster j for exchanging cluster identifiers is marked as m. ji That is, there is m in cluster j. ji A vertex may need to exchange its cluster identifier with a vertex in cluster i.

[0056] Then, it can be done through formula x ij =min(m ij ,m jiDetermine the minimum number of vertices that can exchange cluster identifiers in the two clusters, and use the minimum number of vertices that can exchange cluster identifiers in the two clusters as the vertex exchange probability corresponding to the two clusters.

[0057] Finally, for one of the clusters, such as cluster i, a specified number of candidate vertices for cluster identifier swapping are determined based on the vertex swapping probability. Then, the specified number of candidate vertices for cluster identifier swapping are randomly selected from the candidate vertices for cluster identifier swapping of cluster i, and the specified number of candidate vertices for cluster identifier swapping are randomly selected from the candidate vertices for cluster identifier swapping of cluster j to swap cluster identifiers, resulting in an object relation network graph after performing one cluster identifier swapping operation (i.e., one coarse clustering). This is equivalent to randomly selecting x from objects in cluster i that can be partitioned into cluster j. ij Each object is assigned to cluster j. Simultaneously, x is randomly selected from the objects in cluster j that can be assigned to cluster i. ij Each object is assigned to cluster i, and the objects included in the two clusters are adjusted accordingly.

[0058] By performing the aforementioned cluster identifier swapping at each pair of clusters, the objects included in the M clusters can be adjusted. For example, Figure 2 The object relation network graph obtained after performing cluster identifier swapping on the vertices in the aforementioned object relation network graph can be as follows: Figure 3 As shown.

[0059] In some embodiments of this application, the number of clusters of the adjustment object can be set, for example, performing cluster object adjustment for M / 2 out of M clusters.

[0060] After initial adjustments, the probability of objects in the same cluster having a pre-defined network relationship increases, while the probability of objects in different clusters having a pre-defined network relationship decreases.

[0061] Next, fine clustering is performed on the coarse clustering results. In the fine clustering process, with the goal of minimizing the energy of the object relationship network graph after exchanging cluster labels, the cluster labels of at least some vertices in the object relationship network graph are exchanged using a simulated annealing clustering method.

[0062] In some embodiments of this application, the JABEJA (an open-source clustering method) clustering method can be used for fine clustering to further adjust the objects in each cluster obtained in the previous step, so as to divide the clusters with pure network relationships as much as possible.

[0063] In some embodiments of this application, cluster identifiers of some or all vertices in the object relation network graph are swapped using a simulated annealing clustering method to determine the cluster identifier setting result that minimizes the energy of the object relation network graph. This includes: identifying each vertex with a pre-swapped cluster identifier and its pre-swapped target vertex based on the cluster identifier distribution of each vertex's neighboring vertices in the object relation network graph; performing pre-swapped cluster identifiers on each vertex with a pre-swapped cluster identifier and each pre-swapped target vertex using the simulated annealing clustering method to determine the vertices whose cluster identifiers need to be swapped and their corresponding target vertices, wherein the vertices whose cluster identifiers need to be swapped and their corresponding target vertices are respectively: the vertex with the pre-swapped cluster identifier that reduces the energy of the object relation network graph after performing pre-swapped cluster identifiers and the target vertex whose cluster identifier is swapped with it; and performing cluster identifier swaps on the determined vertices whose cluster identifiers need to be swapped and their corresponding target vertices. At this point, the energy in the resulting object relation network graph is minimized.

[0064] First, based on the distribution of cluster identifiers of the neighboring vertices of each vertex in the object relationship network graph obtained by the coarse clustering, each pre-exchange cluster identifier vertex and its pre-exchange target vertex are identified. In some embodiments of this application, if a vertex's neighboring vertices include points with different cluster identifiers, then that vertex is identified as a pre-exchange cluster identifier vertex (i.e., if a user has users with weaker relationships around them, then that user is considered to need to be assigned to another cluster). Figure 3 Taking the object relationship network diagram shown as an example, vertex 201 has a cluster identifier of number 2, its neighboring vertices 202, 203, 206, and 204 have cluster identifiers of number 1, and its neighboring vertex 207 has a cluster identifier of number 2. Therefore, 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. Following this method, all pre-swap cluster identifier vertices corresponding to each cluster identifier, as well as the pre-swap target vertices for each pre-swap cluster identifier vertex, can be determined.

[0065] Subsequently, using simulated annealing clustering, a cluster identifier pre-swapping operation is performed on each vertex with pre-swapped cluster identifiers. The change in relational network energy resulting from the pre-swapping operation determines whether to perform a cluster identifier swap on that vertex. This ultimately identifies the vertices whose cluster identifiers need to be swapped and their corresponding target vertices. Swapping a vertex's cluster identifier is equivalent to adjusting the cluster of the object corresponding to that vertex. Further clustering of the objects is then performed through this process.

[0066] In some embodiments of this application, a simulated annealing clustering method is used to perform pre-exchange cluster identification on each pre-exchange cluster identifier vertex and each pre-exchange target vertex of the pre-exchange cluster identifier vertex to determine the vertex that needs to exchange cluster identifiers and the corresponding exchange target vertex. This includes: performing pre-exchange cluster identification on the pre-exchange cluster identifier vertex and the pre-exchange target vertex of the pre-exchange cluster identifier vertex respectively to obtain a second object relation network graph corresponding to each pre-exchange cluster identification; calculating the energy change brought about by the corresponding pre-exchange cluster identification based on the second object relation network graph corresponding to each pre-exchange cluster identification; in response to the energy change brought about by the pre-exchange cluster identification being the largest energy decrease, determining the pre-exchange cluster identifier vertex and the pre-exchange target vertex corresponding to the corresponding pre-exchange cluster identification as the vertex that needs to exchange cluster identifiers and the corresponding exchange target vertex; or, in response to the energy change brought about by the pre-exchange cluster identification being the energy increase, abandoning the execution of exchange cluster identification on the pre-exchange cluster identifier vertex corresponding to the corresponding pre-exchange cluster identification in the object relation network graph.

[0067] The object relationship network graph obtained after coarse clustering is as follows: Figure 3 As shown in the example, the pre-swapping target vertices corresponding to the pre-swapping cluster identifier vertex 201 can be determined to be vertices 202, 203, 206, and 204, and the pre-swapping target vertices corresponding to the pre-swapping cluster identifier vertex 204 can be determined to be vertices 201 and 205. Next, pre-swapping cluster identifiers are performed on the pre-swapping cluster identifier vertex 201 and its pre-swapping target vertices 202, 203, 206, and 204, and on the pre-swapping cluster identifier vertex 204 and its pre-swapping target vertices 201 and 205, respectively. Performing one pre-swapping cluster identifier will result in a corresponding second object relation network graph. The difference between the second object relation network graph and the first object relation network graph is that the cluster identifiers of some vertices in the second graph have changed. Figure 3 Taking the object relationship network graph shown as an example, after the pre-exchange cluster identifier vertex 201 and its pre-exchange target vertices 202, 203, 206 and 204 perform cluster identifier pre-exchange, the second object relationship network graphs N1, N2, N3 and N4 will be obtained. After the pre-exchange cluster identifier vertex 204 and its pre-exchange target vertices 201 and 205 perform cluster identifier pre-exchange, the second object relationship network graphs N5 and N6 will be obtained.

[0068] Next, based on the second object relation network graph obtained from each pre-exchange of cluster identifiers, the energy change brought about by the corresponding pre-exchange of cluster identifiers is calculated. For example, by counting the number of edges connecting vertices with different cluster identifiers in the second object relation network graphs N1, N2, N3, and N4 respectively, the energies of the second object relation network graphs N1, N2, N3, and N4 are determined respectively.

[0069] Then, the energies of the second object relation network graphs N1, N2, N3, and N4 are compared with the energy of the object relation network graph. If the energies of the second object relation network graphs N1, N2, N3, and N4 are higher than the energy of the object relation network graph, then the exchange cluster identification is abandoned for the pre-exchange cluster identifier vertex 201 and the pre-exchange target vertices 202, 203, 206, and 204 corresponding to the current pre-exchange cluster identifier.

[0070] If, by counting the number of edges connecting vertices with different cluster identifiers in the second object relation network graphs N5 and N6 respectively, it is determined that the energy of the second object relation network graphs N5 and N6 is less than the energy of the object relation network graph, then the pre-exchange cluster identifier vertex 204 and the pre-exchange target vertices 201 and 205 corresponding to the current pre-exchange cluster identifier are determined as the vertex whose cluster identifier needs to be exchanged and the corresponding exchange target vertex.

[0071] In other embodiments of this application, if a pre-exchange cluster identifier vertex corresponds to multiple pre-exchange target vertices, performing multiple cluster identifier exchanges on the pre-exchange cluster identifier vertex will result in a second object relation network graph corresponding to each cluster identifier exchange. Then, the pre-exchange target vertex targeted by the cluster identifier exchange corresponding to the second object relation network graph with the lowest energy is selected as the exchange target vertex of the pre-exchange cluster identifier vertex. For example, if the energy of the second object relation network graph N5 is less than the energy of the second object relation network graph N6, then the pre-exchange cluster identifier vertex 204 and the pre-exchange target vertex 201 corresponding to the current pre-exchange cluster identifier are determined as the vertex whose cluster identifier needs to be exchanged and the corresponding exchange target vertex.

[0072] Finally, the cluster identification is swapped for the determined vertices that need to have their cluster identifiers swapped (e.g., vertex 204) and the corresponding target vertex (e.g., vertex 201), resulting in a finely clustered object relation network graph. For example, Figure 3 The object relationship network graph shown can be obtained after performing cluster identifier swapping on the vertices. Figure 4 The diagram shows the object relationship network.

[0073] The energy of the object relation network graph is proportional to the number of edges connecting vertices with different cluster identifiers. For example, the energy of the object relation network graph can be represented by the number of edges connecting vertices with different cluster identifiers. As can be seen from the aforementioned cluster identifier exchange scheme, each cluster identifier exchange is conditional on a decrease in the energy of the object relation network graph. Therefore, after multiple cluster identifier exchanges, the energy of the resulting object relation network graph can reach a relatively small value.

[0074] Step 130: Assign objects to vertices that have the same cluster identifier to the same cluster to obtain M-cluster objects.

[0075] In some embodiments of this application, the cluster to which the object corresponding to each vertex belongs is determined based on the cluster identifier of each vertex in the object relationship network graph.

[0076] In the object relationship network graph obtained after fine clustering, the number of vertices and the edge connections between vertices are the same as those in the object relationship network graph before coarse clustering. The difference is that the cluster identifiers of some vertices have been swapped. Each vertex's cluster identifier indicates the cluster to which the object corresponding to that vertex belongs. Therefore, objects corresponding to vertices with the same cluster identifier will be assigned to the same cluster. For example, the objects corresponding to vertices 202, 203, and 204 with the identifier 1 will be assigned to cluster 1, while the object corresponding to vertex 205 with the identifier 2 will be assigned to cluster 2.

[0077] At this point, the fine clustering of the selected objects has been completed, resulting in M ​​object clusters.

[0078] Step 140: Using clusters as the grouping granularity, randomly divide the M cluster objects into at least two groups of objects.

[0079] In the resulting M clusters, the network relationships between objects in different clusters are weakened, and objects with strong network relationships are grouped into the same cluster. Therefore, each cluster can be further treated as an independent individual to evaluate the effectiveness of strategy execution affected by network relationships, or for other applications.

[0080] Taking the A / B test of the recommendation strategy as an example, the M cluster objects can be randomly divided into two groups, each group including M / 2 clusters of users. The data of the two groups of users obtained by executing different strategies are used to evaluate the effect of different strategies.

[0081] Step 150: Execute different target strategies for each group of objects to determine the execution effect of each target strategy based on the object data obtained from executing the different target strategies.

[0082] In some embodiments of this application, different target strategies are executed for each group of objects to determine the execution effect of each target strategy based on the object data obtained from executing the different target strategies. The preset network relationship is associated with the target strategy, including: executing different target strategies for each group of objects to determine the object data of each object after executing the corresponding target strategy; performing linear fitting on the influence weights of strategy factors, intra-group network relationship factors, and extra-group network relationship factors based on the object data to obtain the respective influence weights of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors; and determining the influence of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors on the execution effect of each target strategy based on the influence weights obtained from the linear fitting. The value of the intra-group network relationship factor represents the proportion of objects in the same group as the current object among those with the preset network relationship; the value of the extra-group network relationship factor represents the proportion of objects in different groups among those with the preset network relationship.

[0083] The preset network relationship is associated with the target strategy.

[0084] Taking the evaluation of the impact of target marketing strategy 1 and target marketing strategy 2 on order volume as an example, the M clusters of users obtained from the previous steps can be randomly divided into two groups, referred to as group 1 and group 2 in this embodiment. Then, target marketing strategy 1 is applied to the users in group 1, and order volume data for each user in group 1 is collected; target marketing strategy 2 is applied to the users in group 2, and order volume data for each user in group 2 is collected. Finally, based on the order volume data of each user, the weights of the three factors affecting order volume data—strategy factors, intra-group network relationship factors, and inter-group network relationship factors—are linearly fitted, and the impact of strategy factors on order volume data is determined based on the fitting results.

[0085] For example, linear fitting is performed according to the formula y = α*treatment + β*network + γ*spill, where the observations include: object data y, strategy factor value treatment, intra-group network relationship factor value network, and out-of-group network relationship factor value spill, and α, β, and γ are the required parameter values. Linear fitting using the above formula seeks the optimal estimates of parameters α, β, and γ using the object data y, strategy factor value treatment, intra-group network relationship factor value network, and out-of-group network relationship factor value spill. Specifically, in this embodiment, the above observations are determined in the following way.

[0086] The object data y, i.e. the order volume data, is determined based on the actual order volume of each user obtained by implementing target marketing strategies 1 and 2 for the two groups of users mentioned above.

[0087] The treatment value is determined based on the target strategy currently in use. For example, if target marketing strategy 1 corresponds to treatment = 1 and target marketing strategy 2 corresponds to treatment = 0, then for all users who are subject to target marketing strategy 1, the treatment = 1, and for all users who are subject to target marketing strategy 2, the treatment = 0.

[0088] The intra-group network relationship factor value `network` is determined based on the object relationship network and the specific object grouping. For example, for user A, who is in group 1 and is subject to target marketing strategy 1, if user B (corresponding to vertex 202) and user C (corresponding to vertex 203) in the object relationship network corresponding to vertex 201 are both in group 1, while user D (corresponding to vertex 204) is in group 2, then user A's intra-group network relationship factor value `network` = 2 / 3.

[0089] Similarly, the spill value of the out-of-group network relationship factor is determined based on the object relationship network and the specific object grouping. For example, for user A, who is in group 1 and is subject to target marketing strategy 1, if user B, whose neighboring vertex 202 corresponds to vertex 202, and user C, whose neighboring vertex 203 corresponds to vertex 203, are both in group 1, while user D, whose neighboring vertex 204 corresponds to vertex 201, is in group 2, then user A's spill value of the out-of-group network relationship factor is 1 / 3.

[0090] Following the aforementioned method, a formula containing three parameters can be determined based on the amount of order data for each user. For example, for user A, the formula can be determined as follows: 80 represents the number of orders placed by user A. All user data in groups 1 and 2 are used to obtain linear fits of parameters α, β, and γ, resulting in two sets of values ​​for α, β, and γ.

[0091] The two sets of α, β, and γ values ​​represent different object data obtained by using different target strategies. Therefore, the final value of parameter α is taken as the weight of the target strategy's influence on the object data. The mean or sum of the object data of all objects in each group can be used as the basis for evaluating the effectiveness of the target strategy.

[0092] In some other embodiments of this application, taking an A / B testing application scenario as an example, the step of determining the execution effect of each target strategy based on the object data obtained from executing the different target strategies by each group of objects includes: selecting different groups of objects to form an experimental group and a control group, and executing a first target strategy on the objects in the experimental group to determine the first object data after executing the first target strategy; and executing a second target strategy on the objects in the control group to determine the second object data after executing the second target strategy; and performing linear fitting on the influence weights of strategy factors, intra-group network relationship factors, and extra-group network relationship factors based on the first object data to determine the... The strategy factors, the intra-group network relationship factors, and the inter-group network relationship factors each have a first influence weight; and, based on the second object data, the influence weights of the strategy factors, the intra-group network relationship factors, and the inter-group network relationship factors are linearly fitted to determine the second influence weights of the strategy factors, the intra-group network relationship factors, and the inter-group network relationship factors; based on the first influence weights and the second influence weights of the strategy factors, the intra-group network relationship factors, and the inter-group network relationship factors, the influence of the strategy factors, the intra-group network relationship factors, and the inter-group network relationship factors on the execution effect of the first target strategy and the second target strategy is determined.

[0093] For example, the M clusters of users obtained from the aforementioned steps can be randomly divided into two groups, referred to in this embodiment as the experimental group and the control group. Then, a first objective strategy (e.g., estimating the number of user orders based on user profiles) is applied to the users in the experimental group, and order volume data for each user in the experimental group is collected. A second objective strategy (e.g., estimating the number of user orders based on user profiles and geographical location) is applied to the users in the control group, and order volume data for each user in the control group is collected. Finally, based on the order volume data for each user, a linear fit is performed on the weights of the three factors influencing order volume data: strategy factors, intra-group network relationship factors, and inter-group network relationship factors. The impact of strategy factors on order volume data is determined based on the fitting results.

[0094] For the specific methods of performing current fitting based on data from the control group and experimental group users, please refer to the relevant description of linear fitting mentioned above, which will not be repeated here.

[0095] The object data y, i.e. the order volume data, is determined by the actual order volume of each user obtained by implementing the first target strategy and the second target strategy on the two groups of users mentioned above.

[0096] The treatment value is determined based on the current target strategy. Assuming the first target strategy corresponds to treatment=1 and the second target strategy corresponds to treatment=0, then for all users who are subject to the first target strategy, treatment=1, and for all users who are subject to the second target strategy, treatment=0.

[0097] The intra-group network relationship factor value `network` is determined based on the object relationship network and the specific object grouping. For example, for user A, who is assigned to the experimental group and is subject to the first target strategy, if user B (corresponding to vertex 202) and user C (corresponding to vertex 203) in the object relationship network corresponding to vertex 201 are both assigned to the experimental group, while user D (corresponding to vertex 204) is assigned to the control group, then user A's intra-group network relationship factor value `network` = 2 / 3.

[0098] Similarly, the spill value of the out-of-group network relationship factor is determined based on the object relationship network and the specific object grouping. For example, for user A, who is in the experimental group and is subject to the first target strategy, if user B, whose neighboring vertex 202 corresponds to vertex 202, and user C, whose neighboring vertex 203 corresponds to vertex 203, are both in the experimental group, while user D, whose neighboring vertex 204 corresponds to vertex 201, is in the control group, then user A's spill value of the out-of-group network relationship factor is 1 / 3.

[0099] Following the aforementioned method, a formula containing three parameters can be determined based on the amount of order data for each user. For example, for user A, the formula can be determined as follows: Here, 60 represents the order quantity of user A. Based on the expression containing only parameters α, β, and γ obtained from all user data in the control group, a set of values ​​for α, β, and γ can be obtained by linearly fitting the values ​​of α, β, and γ. Similarly, based on the expression containing only parameters α, β, and γ obtained from all user data in the experimental group, a set of values ​​for α, β, and γ can also be obtained by linearly fitting the values ​​of α, β, and γ.

[0100] Since the users in the experimental and control groups are consistent, the values ​​of α, β, and γ in the two groups represent the different object data obtained by using different target strategies. Finally, the value of parameter α is taken as the weight of the impact of the target strategy on the object data. The mean or sum of the object data of all objects in each group can be used as the basis for evaluating the effectiveness of the target strategy.

[0101] The method for determining the strategy execution effect disclosed in this application constructs an object relationship network graph based on a preset network relationship between objects. Each vertex in the object relationship network graph corresponds one-to-one with an object, and M preset cluster identifiers are evenly distributed on each vertex. Edges in the object relationship network graph connect the vertices corresponding to objects with the preset network relationship. By exchanging the cluster identifiers of some or all vertices in the object relationship network graph, the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions is determined. The energy of the object relationship network graph is determined based on the number of edges connecting vertices with different cluster identifiers. Objects corresponding to vertices with the same cluster identifier are grouped into the same cluster, resulting in M ​​cluster objects. Using clusters as the grouping granularity, the M cluster objects are randomly divided into at least two groups of objects. Different target strategies are executed on each group of objects, and the execution effect of each target strategy is determined based on the object data obtained from executing the different target strategies. The preset network relationship is associated with the target strategy, which helps improve the accuracy of determining the strategy execution effect.

[0102] This application constructs a network graph for objects exhibiting network relationships or propagation effects. By uniformly grouping objects and employing an exchange strategy in the subsequent clustering process, it ensures a consistent number of objects in each cluster, avoiding situations where varying object numbers lead to different influence weights for each cluster. Furthermore, a coarse clustering method combining randomness and probability is used for rapid object grouping, improving data processing efficiency and saving computational resources. Moreover, fine clustering based on the coarse clustering results enhances the stability of object grouping while maintaining rapid grouping.

[0103] The method for determining the effect of strategy execution disclosed in this application is applicable to the effect measurement scheme in network propagation scenarios. By grouping objects with weak connections and applying different strategies to objects with weak or no network relationships, the influence of network relationships on the strategy can be reduced.

[0104] Example 2

[0105] This application discloses a method and apparatus for determining the effect of strategy execution, such as... Figure 5 As shown, the device includes:

[0106] The object relationship network graph construction module 510 is used to construct an object relationship network graph based on a preset network relationship between objects; wherein, the vertices in the object relationship network graph correspond one-to-one with the objects, and M preset cluster identifiers are evenly distributed on the vertices; the edges in the object relationship network graph connect the vertices corresponding to the objects that have the preset network relationship;

[0107] The cluster identifier exchange module 520 is used to determine the cluster identifier setting result that minimizes the energy of the object relationship network graph under preset conditions by exchanging the cluster identifiers of some or all vertices in the object relationship network graph; wherein, the energy of the object relationship network graph is determined based on the number of edges connecting vertices with different cluster identifiers;

[0108] The first object grouping module 530 is used to group objects corresponding to the vertices that have the same cluster identifier into the same cluster to obtain M cluster objects;

[0109] The second object grouping module 540 is used to randomly divide the M cluster objects into at least two groups of objects with clusters as the grouping granularity.

[0110] The strategy execution effect determination module 550 is used to execute different target strategies on each group of objects to determine the execution effect of each target strategy based on the object data obtained from executing the different target strategies.

[0111] Wherein, the preset network relationship is associated with the target strategy, and M is a natural number greater than 2.

[0112] In some embodiments of this application, such as Figure 6 As shown, the cluster identifier exchange module 520 further includes:

[0113] The coarse clustering submodule 5201 is used to perform coarse clustering a preset number of times. The coarse clustering includes: randomly swapping the cluster identifiers of some of the vertices based on a determined probability.

[0114] The fine clustering submodule 5202 is used to perform fine clustering on the coarse clustering results. The fine clustering includes: exchanging the cluster labels of some or all vertices in the object relationship network graph using a simulated annealing clustering method, and determining the cluster label setting result that minimizes the energy of the object relationship network graph.

[0115] In some embodiments of this application, the coarse clustering submodule 5201 is further configured to:

[0116] Perform the following operations for each pair of clusters:

[0117] Based on the distribution of cluster identifiers of the neighboring vertices of each vertex in the object relation network graph, determine the candidate vertices for exchanging cluster identifiers corresponding to the two clusters and the number of such candidate vertices;

[0118] The vertex swapping probability between the two clusters is determined based on the number of candidate vertices corresponding to the swap cluster identifiers.

[0119] From the candidate vertices for exchanging cluster identifiers corresponding to the two clusters, a specified number of candidate vertices for exchanging cluster identifiers are randomly selected for cluster identifier exchange; wherein, the specified number is determined according to the vertex exchange probability.

[0120] In some embodiments of this application, determining the candidate vertex for the swapped cluster identifier corresponding to the two clusters includes:

[0121] For the first of the two clusters, the vertex in the object relation network graph that has the cluster identifier of the first cluster and whose neighboring vertex cluster identifier includes the cluster identifier of the second of the two clusters is determined as the candidate vertex for the exchange cluster identifier of the first cluster.

[0122] For the second cluster, vertices in the object relation network graph that have the cluster identifier of the second cluster and whose neighboring vertices have the cluster identifier of the first cluster are determined as candidate vertices for the exchange cluster identifier of the corresponding second cluster.

[0123] In some embodiments of this application, the fine clustering submodule 5202 is further used for:

[0124] Based on the distribution of cluster identifiers of the neighboring vertices of each vertex in the object relation network graph, identify each pre-exchange cluster identifier vertex and the pre-exchange target vertex of the pre-exchange cluster identifier vertex;

[0125] By using a simulated annealing clustering method, pre-exchange cluster identification is performed on each pre-exchange cluster identifier vertex and each pre-exchange target vertex of the pre-exchange cluster identifier vertex to determine the vertex that needs to exchange cluster identifiers and the corresponding exchange target vertex; wherein, the vertex that needs to exchange cluster identifiers and the corresponding exchange target vertex are respectively: the pre-exchange cluster identifier vertex that causes the energy of the object relation network graph to decrease after performing pre-exchange cluster identification and the pre-exchange target vertex that exchanges cluster identifiers with it;

[0126] Perform cluster swapping on the identified vertices that require cluster swapping and their corresponding target vertices.

[0127] In some embodiments of this application, the strategy execution effect determination module 550 is further configured to:

[0128] Different target strategies are executed for each group of objects, and the object data of each object is determined after the execution of the corresponding target strategy.

[0129] Based on the object data, the influence weights of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors are linearly fitted to obtain the respective influence weights of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors.

[0130] Based on the influence weights obtained through linear fitting, the influence of the strategy factors, the intra-group network relationship factors, and the inter-group network relationship factors on the execution effect of each target strategy is determined.

[0131] In some embodiments of this application, the strategy execution effect determination module 550 is further configured to:

[0132] Different groups of objects are selected to form experimental and control groups, and a first target strategy is executed on the objects in the experimental group to determine the first object data after executing the first target strategy; and a second target strategy is executed on the objects in the control group to determine the second object data after executing the second target strategy.

[0133] Based on the first object data, a linear fit is performed on the influence weights of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors to determine the first influence weights of each of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors; and based on the second object data, a linear fit is performed on the influence weights of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors to determine the second influence weights of each of the strategy factors, intra-group network relationship factors, and extra-group network relationship factors.

[0134] Based on the first influence weight and the second influence weight of the strategy factor, the intra-group network relationship factor, and the extra-group network relationship factor, determine the influence of the strategy factor, the intra-group network relationship factor, and the extra-group network relationship factor on the execution effect of the first target strategy and the second target strategy.

[0135] The strategy execution effect determination method and apparatus disclosed in this application is used to implement the steps of the strategy execution effect determination method described in Embodiment 1 of this application. The specific implementation of each module of the apparatus is described in the corresponding steps, and will not be repeated here.

[0136] The strategy execution effect determination method and apparatus disclosed in this application constructs an object relationship network graph based on a preset network relationship between objects. In this graph, each vertex corresponds to an object, and M preset cluster identifiers are evenly distributed on the vertices. Edges in the graph connect the vertices corresponding to objects with the preset network relationship. By exchanging the cluster identifiers of some or all vertices in the graph, the cluster identifier setting result that minimizes the energy of the graph under preset conditions is determined. The energy of the graph is determined based on the number of edges connecting vertices with different cluster identifiers. Objects corresponding to vertices with the same cluster identifier are grouped into the same cluster, resulting in M ​​cluster objects. Using clusters as the grouping granularity, the M cluster objects are randomly divided into at least two groups. Different target strategies are executed on each group of objects, and the execution effect of each target strategy is determined based on the object data obtained from executing the different target strategies. The preset network relationship is associated with the target strategy, which helps improve the accuracy of strategy execution effect determination.

[0137] This application constructs a network graph for objects exhibiting network relationships or propagation effects. By uniformly grouping objects and employing an exchange strategy in the subsequent clustering process, it ensures a consistent number of objects in each cluster, avoiding situations where varying object numbers lead to different influence weights for each cluster. Furthermore, a coarse clustering method combining randomness and probability is used for rapid object grouping, improving data processing efficiency and saving computational resources. Moreover, fine clustering based on the coarse clustering results enhances the stability of object grouping while maintaining rapid grouping.

[0138] The strategy execution effect determination device disclosed in this application is applicable to the effect measurement scheme in network propagation scenarios. By grouping objects with weak connections and applying different strategies to objects with weak or no network relationships, the influence of network relationships on the strategy can be reduced.

[0139] Accordingly, this application also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the strategy execution effect determination method as described in Embodiment 1 of this application. The electronic device can be a PC, mobile terminal, personal digital assistant, tablet computer, etc.

[0140] This application also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the strategy execution effect determination method as described in Embodiment 1 of this application.

[0141] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus embodiments, since they are fundamentally similar to the method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0142] The above provides a detailed description of a method and apparatus for determining the execution effect of a strategy provided by this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and its core ideas. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0143] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

Claims

1. A method for determining the effect of a sharing click-through strategy, characterized in that, The method comprises the steps of: constructing an object relationship network graph according to preset network relationships between objects; wherein the vertices in the object relationship network graph correspond to the objects one by one, and preset M cluster identifiers are uniformly arranged on the vertices; the edges in the object relationship network graph connect the vertices corresponding to the objects having the preset network relationships; determining a cluster identifier arrangement result that minimizes the energy of the object relationship network graph under a preset condition by exchanging the cluster identifiers of part or all of the 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 the vertices having different cluster identifiers; dividing the objects corresponding to the vertices having 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 in terms of the cluster as the grouping granularity; respectively executing different target strategies on each group of objects to determine the execution effect of each target strategy according to the object data obtained by executing the different target strategies; wherein the preset network relationship is associated with the target strategy, and M is a natural number greater than 2; the objects are users, and the preset network relationship is the relationship of sharing and accepting sharing.

2. The method of claim 1, wherein, The step of determining the cluster identifier arrangement result that minimizes the energy of the object relationship network graph under a preset condition by exchanging the cluster identifiers of part or all of the vertices in the object relationship network graph comprises: performing a preset number of coarse clustering, the coarse clustering comprising: based on a determined probability, randomly exchanging the cluster identifiers of part of the vertices; performing fine clustering on the coarse clustering result, the fine clustering comprising: exchanging the cluster identifiers of part or all of the vertices in the object relationship network graph by a simulated annealing clustering method to determine the cluster identifier arrangement result that minimizes the energy of the object relationship network graph, the object relationship network graph being obtained after coarse clustering; The step of randomly exchanging the cluster identifiers of part of the vertices based on a determined probability comprises: For each of the two clusters, the following operations are performed: determining the exchange cluster identifier candidate vertices corresponding to the two clusters and the number of the exchange cluster identifier candidate vertices according to the cluster identifier distribution of the neighbor vertices of each vertex in the object relationship network graph; determining the vertex exchange probability corresponding to the two clusters according to the number of the exchange cluster identifier candidate vertices; randomly selecting a specified number of exchange cluster identifier candidate vertices from the exchange cluster identifier candidate vertices corresponding to the two clusters for cluster identifier exchange; wherein the specified number is determined according to the vertex exchange probability; The vertex exchange probability comprises: determining the minimum number of vertices that can exchange cluster identifiers in the two clusters by the formula xij = min(mij, mji), and taking the minimum number of vertices that can exchange cluster identifiers in the two clusters as the vertex exchange probability corresponding to the two clusters, wherein mij represents the number of exchange cluster identifier candidate vertices in cluster i, and mji represents the number of exchange cluster identifier candidate vertices in cluster j.

3. The method of claim 2, wherein, The step of determining the exchange cluster identifier candidate vertices corresponding to the two clusters comprises: For the first cluster of the two clusters, a vertex in the object relationship network graph provided with the cluster identifier of the first cluster and having a neighbor vertex provided with the cluster identifier of the second cluster of the two clusters is determined as a candidate vertex of the exchange cluster identifier corresponding to the first cluster; For the second cluster, a vertex in the object relationship network graph provided with the cluster identifier of the second cluster and having a neighbor vertex provided with the cluster identifier of the first cluster is determined as a candidate vertex of the exchange cluster identifier corresponding to the second cluster.

4. The method of claim 2, wherein, The step of determining the cluster identifier setting result that minimizes the energy of the object relationship network graph by exchanging the cluster identifiers of part or all vertices in the object relationship network graph through the simulated annealing clustering method comprises: According to the cluster identifier distribution of the neighbor vertices of each vertex in the object relationship network graph, each pre-exchange cluster identifier vertex and the pre-exchange target vertex of the pre-exchange cluster identifier vertex are identified; The pre-exchange cluster identifier of each pre-exchange cluster identifier vertex and each pre-exchange target vertex of the pre-exchange cluster identifier vertex is executed through the simulated annealing clustering method to determine the vertex that needs to exchange the cluster identifier and the corresponding exchange target vertex; wherein the vertex that needs to exchange the cluster identifier and the corresponding exchange target vertex are respectively the pre-exchange cluster identifier vertex and the pre-exchange target vertex whose cluster identifiers are exchanged after the execution of the pre-exchange cluster identifier, which makes the energy of the object relationship network graph decrease; The exchange cluster identifier is executed on the determined vertex that needs to exchange the cluster identifier and the corresponding exchange target vertex.

5. The method of claim 1, wherein, The step of executing different target strategies on each group of objects respectively to determine the execution effect of each target strategy according to the object data obtained by executing the different target strategies comprises: Different target strategies are executed on each group of objects respectively to determine the object data of each object after the execution of the corresponding target strategy; The influence weights of the strategy factor, the intra-group network relationship factor and the extra-group network relationship factor are linearly fitted according to the object data to obtain the influence weights of the strategy factor, the intra-group network relationship factor and the extra-group network relationship factor respectively; The influence of the strategy factor, the intra-group network relationship factor and the extra-group network relationship factor on the execution effect of each target strategy is determined according to the influence weights obtained by linear fitting.

6. The method of claim 1, wherein, The step of executing different target strategies on each group of objects respectively to determine the execution effect of each target strategy according to the object data obtained by executing the different target strategies comprises: Different groups of objects are selected to form an experimental group and a control group respectively, a first target strategy is executed on the objects in the experimental group to determine first object data after the execution of the first target strategy, and a second target strategy is executed on the objects in the control group to determine second object data after the execution of the second target strategy; According to the first object data, the influence weights of the policy factor, the intra-group network relationship factor and the extra-group network relationship factor are linearly fitted to determine first influence weights of the policy factor, the intra-group network relationship factor and the extra-group network relationship factor respectively; and according to the second object data, the influence weights of the policy factor, the intra-group network relationship factor and the extra-group network relationship factor are linearly fitted to determine second influence weights of the policy factor, the intra-group network relationship factor and the extra-group network relationship factor respectively; According to the first influence weights and the second influence weights of the policy factor, the intra-group network relationship factor and the extra-group network relationship factor respectively, the influence of the policy factor, the intra-group network relationship factor and the extra-group network relationship factor on the execution effect of the first target policy and the second target policy is determined.

7. A sharing click strategy performance effect determination apparatus characterized by comprising: Comprise: An object relationship network graph construction module configured to construct an object relationship network graph according to a preset network relationship between objects; Wherein, vertices in the object relationship network graph correspond to the objects one by one, and preset M cluster identifiers are uniformly arranged on the vertices; edges in the object relationship network graph connect the vertices corresponding to the objects having the preset network relationship; A cluster identifier exchange module configured to exchange cluster identifiers of part or all of the vertices in the object relationship network graph to determine a cluster identifier arrangement result that minimizes the energy of the object relationship network graph under a preset condition; wherein, the energy of the object relationship network graph is determined according to the number of edges connecting vertices having different cluster identifiers; A first object grouping module configured to divide the objects corresponding to the vertices arranged with the same cluster identifier into the same cluster to obtain M cluster objects; A second object grouping module configured to divide the M cluster objects into at least two groups of objects randomly with a cluster as a grouping granularity; A sharing click policy execution effect determination module configured to execute different target policies for each group of objects respectively to determine the 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; The objects are users, and the preset network relationship is a sharing and accepting sharing relationship.

8. The apparatus of claim 7, wherein, The cluster identifier exchange module further comprises: A coarse clustering submodule configured to perform a preset number of coarse clustering, the coarse clustering comprising: exchanging the cluster identifiers of part of the vertices randomly based on a determined probability; A fine clustering submodule configured to perform fine clustering on the coarse clustering result, the fine clustering comprising: exchanging the cluster identifiers of part or all of the vertices in the object relationship network graph by a simulated annealing clustering method to determine a cluster identifier arrangement result that minimizes the energy of the object relationship network graph; The step of exchanging the cluster identifiers of part of the vertices randomly comprises: For each two clusters, the following operations are performed respectively: According to the cluster identifier distribution of the neighbor vertices of each vertex in the object relationship network graph, exchange cluster identifier candidate vertices corresponding to the two clusters and the number of the exchange cluster identifier candidate vertices are determined; determining a vertex exchange probability corresponding to the two clusters according to a number of the exchange cluster identification candidate vertices corresponding to the two clusters; randomly selecting a specified number of the exchange cluster identification candidate vertices from the exchange cluster identification candidate vertices corresponding to the two clusters respectively for cluster identification exchange; wherein the specified number is determined according to the vertex exchange probability; the vertex exchange probability comprises: determining a minimum vertex number of exchangeable cluster identification in the two clusters through a formula xij = min(mij, mji), taking the minimum vertex number of exchangeable cluster identification in the two clusters as the vertex exchange probability corresponding to the two clusters, wherein mij is a number identifier of the exchange cluster identification candidate vertex in cluster i, and mji represents a number identifier of the exchange cluster identification candidate vertex in cluster j.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, the processor implements the sharing click strategy execution effect determination method of any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, the program is executed by the processor to implement the steps of the sharing click strategy execution effect determination method of any one of claims 1 to 6.

Images