Activity plan and object matching methods, hypergraph creation methods and related equipment

Abstract

This application discloses an activity-object matching method, a hypergraph creation method, and related equipment. The application establishes an activity-object hypergraph, including activity plan nodes and activity object nodes. Activity plan nodes and activity object nodes with matching relationships are connected by edges, modeling low-order matching relationships between activity plans and activity objects. For each node in the hypergraph, high-order attribute features can be extracted, and nodes with the same high-order attribute feature values ​​are connected by the same hyperedge, modeling high-order attribute features between nodes. Based on the created hypergraph, both low-order matching relationship features and high-order attribute features between activity plans and activity objects can be captured, obtaining high-quality hypergraph feature representations of activity plans and activity objects. Furthermore, the matching degree between activity plans and activity objects is calculated, and the matching relationship between activity plans and activity objects is determined according to the matching degree, improving the accuracy of matching relationship prediction.

Description

Technical Field

[0001] This application relates to the field of data matching technology, and more specifically, to an activity scheme and object matching method, a hypergraph creation method, and related equipment. Background Technology

[0002] With the development of modern society, more and more institutions and organizations are beginning to focus on the planning and implementation of activities. Different activity plans may target different audiences; for example, activity policies issued by government departments may be aimed at specific types of enterprises.

[0003] To facilitate participation in matched activity plans, it's necessary to match published activity plans with multiple potential participants. This allows for recommending activity plans to matched participants or vice versa. With the development of graph neural networks, increasingly more solutions rely on graph structures to represent pairwise matching relationships between activity plans and participants. Graph convolution techniques are used to learn representation vectors for each node in the graph, and finally, matching relationships are predicted based on these vectors. However, existing graph structures can only model low-order matching relationships between activity plan and participant nodes. For example, in government policy announcements, existing graph structures can only model the matching relationship (also known as the policy implementation relationship) between policy nodes and enterprise nodes. This results in inaccurate node feature representations learned from these ordinary graph structures, affecting the accuracy of subsequent predictions of matching relationships between activity plans and participants. Summary of the Invention

[0004] In view of the above problems, this application proposes to provide an activity scheme and object matching method, a hypergraph creation method, and related equipment, so as to simultaneously model the low-order matching relationship and high-order correlation features between activity schemes and activity objects through the hypergraph, improve the accuracy of node feature representation, and thus improve the accuracy of matching relationship prediction. The specific solution is as follows:

[0005] Firstly, a method for matching activity plans with objects is provided, including:

[0006] Based on the configured activity-object hypergraph, the hypergraph feature representations of activity schemes and activity objects are obtained. The activity-object hypergraph includes activity scheme nodes and activity object nodes. Activity scheme nodes and activity object nodes that are known to have a matching relationship are connected by edges. Activity scheme nodes and / or activity object nodes with the same higher-order attribute feature values ​​are connected by the same hyperedge.

[0007] Based on the hypergraph feature representations of the activity plan and the activity object respectively, the matching degree between the activity plan and the activity object is calculated;

[0008] Based on the degree of matching, the matching relationship between the activity plan and the activity object is determined.

[0009] Preferably, the process of obtaining the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph includes:

[0010] In the activity-object hypergraph, for each node, the neighboring node information of the node is aggregated to obtain the low-order features of each node;

[0011] In the activity-object hypergraph, for each hyperedge, the information of all nodes connected by the hyperedge is aggregated to obtain the feature representation of each hyperedge;

[0012] Based on the dependency relationship between nodes and hyperedges, for each node, the feature representations of each hyperedge associated with the node are aggregated to obtain the higher-order features of each node.

[0013] The low-order and high-order features of each node are combined to form the hypergraph feature representation of each node.

[0014] Preferably, there are multiple activity-object hypergraphs, and different activity-object hypergraphs correspond to different time periods. The corresponding activity-object hypergraph is created by collecting activity plan information, activity object information and activity plan implementation information within the corresponding time period. The activity plan implementation information represents the pairing information consisting of activity objects and the activity plans they have participated in.

[0015] The activity-object hypergraphs corresponding to each time period constitute the activity-object spatiotemporal hypergraph.

[0016] Preferably, the process of obtaining the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph includes:

[0017] In the activity-object hypergraph corresponding to each time period, determine the hypergraph feature representation of each node;

[0018] For each node, the hypergraph feature representations of the node in each time period are sequentially input into a pre-trained recurrent neural network according to the timeline, and the last hidden state of the recurrent neural network is taken as the spatiotemporal hypergraph feature representation of the node.

[0019] Preferably, based on the hypergraph feature representations of the activity plan and the activity object respectively, the degree of matching between the activity plan and the activity object is calculated, including:

[0020] Perform a dot product operation on the hypergraph feature representation of the activity scheme and the hypergraph feature representation of the activity object to obtain the matching value between the activity scheme and the activity object;

[0021] Then, based on the degree of matching, the matching relationship between the activity plan and the activity object is determined, including:

[0022] For any target activity plan, sort the activities in descending order according to the matching value between the target activity plan and each activity object, and return the top N1 activities with the highest ranking as the activity objects matched by the target activity plan.

[0023] or,

[0024] For any target activity object, sort them in descending order according to the matching value between the target activity object and each activity plan, and return the top N2 activity plans with the highest ranking as the activity plans matched by the target activity object.

[0025] Preferably, the activity plan includes policy plans, and the target of the activity includes enterprises;

[0026] The higher-order attribute features include industry characteristics and / or regional characteristics.

[0027] Secondly, a method for creating a hypergraph is provided, including:

[0028] Obtain information on each published activity plan, information on each activity target, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents pairing information consisting of activity targets and the activity plans they have participated in.

[0029] Based on the activity plan implementation information, an activity-object pairing relationship graph is established. The activity-object pairing relationship graph contains each activity plan node and each activity object node, and the activity plan nodes and activity object nodes that have a pairing relationship are connected by edges.

[0030] Extract the attribute value of each preset higher-order attribute feature from each of the activity plan information and each of the activity object information;

[0031] On the activity-object pairing graph, a hyperedge is constructed for each attribute value of each preset higher-order attribute feature, and the activity scheme nodes and / or activity objects with the attribute values ​​are connected through the hyperedge to obtain the activity-object hypergraph.

[0032] Preferably, before establishing the activity-object pairing graph, the method further includes:

[0033] The acquired activity plan information, activity object information, and activity plan implementation information are divided into several different time periods. Each time period contains the activity plan information implemented within that time period, the existing activity object information, and the activity plan implementation information.

[0034] The process of establishing an activity-object pairing relationship diagram based on the activity plan implementation information includes:

[0035] Based on the activity plan information, activity target information, and activity plan implementation information for each time period, establish an activity-target pairing relationship diagram for each time period;

[0036] Then, the process of constructing a hyperedge for each attribute value of each preset higher-order attribute feature on the activity-object pairing graph, and connecting the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph includes:

[0037] In the activity-object pairing graph of each time period, a hyperedge is constructed for each attribute value of each preset higher-order attribute feature, and the activity scheme nodes and / or activity objects with the attribute values ​​are connected through the hyperedge to obtain the activity-object hypergraph of each time period.

[0038] Connecting nodes of the same activity object and nodes of the same activity scheme in the activity-object hypergraph of adjacent time periods yields the activity-object spatiotemporal hypergraph.

[0039] Preferably, the activity plan includes policy plans, and the target of the activity includes enterprises;

[0040] The higher-order attribute features include industry characteristics and / or regional characteristics.

[0041] Thirdly, an activity scheme and object matching device is provided, including:

[0042] The feature representation acquisition unit is used to acquire the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph. The activity-object hypergraph includes activity scheme nodes and activity object nodes. Activity scheme nodes and activity object nodes that are known to have a matching relationship are connected by edges. Activity scheme nodes and / or activity object nodes with the same higher-order attribute feature values ​​are connected by the same hyperedge.

[0043] The matching degree calculation unit is used to calculate the matching degree between the activity plan and the activity object based on the hypergraph feature representations of the activity plan and the activity object respectively;

[0044] The matching relationship determination unit is used to determine the matching relationship between the activity plan and the activity object according to the matching degree.

[0045] Fourthly, a hypergraph creation apparatus is provided, comprising:

[0046] The information acquisition unit is used to acquire information on each published activity plan, information on each activity target, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents pairing information consisting of activity targets and the activity plans they have participated in.

[0047] The relationship graph establishment unit is used to establish an activity-object pairing relationship graph based on the activity plan implementation information. The activity-object pairing relationship graph includes each activity plan node and each activity object node, and the activity plan nodes and activity object nodes that have a pairing relationship are connected by edges.

[0048] The attribute value extraction unit is used to extract the attribute value of each preset high-order attribute feature from each of the activity scheme information and each of the activity object information.

[0049] The relationship graph editing unit is used to construct a hyperedge for each attribute value of each preset higher-order attribute feature on the activity-object pairing relationship graph, and connect the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph.

[0050] Fifthly, an electronic device is provided, comprising: a memory and a processor;

[0051] The memory is used to store programs;

[0052] The processor is configured to execute the program to implement the steps of the activity scheme and object matching method as described above, or to implement the steps of the hypergraph creation method as described above.

[0053] In a sixth aspect, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the steps of the activity scheme and object matching method as described above, or implements the steps of the hypergraph creation method as described above.

[0054] Using the above technical solution, this application pre-establishes an activity-object hypergraph. Compared with ordinary graph structures, a hypergraph can connect multiple nodes simultaneously through hyperedges. Based on this, this application places activity scheme nodes and activity object nodes in the hypergraph, and connects activity scheme nodes and activity object nodes with known matching relationships through edges based on prior data. That is, it can model low-order matching relationships between activity schemes and activity objects. Furthermore, for each node in the hypergraph (including activity scheme nodes and activity object nodes), feature values ​​of high-order attribute features can be extracted, and nodes with the same high-order attribute feature values ​​are connected through the same hyperedge, thereby modeling the high-order attribute features between nodes. Through the aforementioned activity-object hypergraph, hypergraph convolution techniques can be used to learn the hypergraph feature representations of both the activity scheme and the activity object. These hypergraph feature representations integrate low-order matching relationship features and high-order attribute features between the activity scheme and the activity object, resulting in high-quality feature representations of the activity scheme and the activity object. Furthermore, the degree of matching between the activity scheme and the activity object can be calculated based on the obtained hypergraph feature representations, and the matching relationship between the activity scheme and the activity object can be determined according to the degree of matching, thereby improving the accuracy of matching relationship prediction. Attached Figure Description

[0055] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0056] Figure 1 This example illustrates a flowchart of a hypergraph creation method.

[0057] Figure 2 This example illustrates an activity-object pairing relationship.

[0058] Figure 3 An example of an activity-object hypergraph;

[0059] Figure 4 This example illustrates another method for creating a hypergraph.

[0060] Figure 5 An example of an activity-object spatiotemporal hypergraph;

[0061] Figure 6 This example illustrates a flowchart of an activity plan and object matching method.

[0062] Figure 7 An example is shown in the schematic diagram illustrating the process of determining the hypergraph feature representation of a node;

[0063] Figure 8An example is shown in the schematic diagram illustrating the process of determining the spatiotemporal hypergraph feature representation of a node;

[0064] Figure 9 This example illustrates an overall framework for creating a spatiotemporal hypergraph and performing policy and firm matching tasks based on it.

[0065] Figure 10 An example is a schematic diagram of a hypergraph creation device;

[0066] Figure 11 Example: A schematic diagram of an activity scheme and object matching device;

[0067] Figure 12 A schematic diagram of the structure of an electronic device is shown. Detailed Implementation

[0068] 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, and 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.

[0069] This application provides a scheme for creating an activity-object hypergraph, and for learning hypergraph feature representations of activity schemes and activity objects based on the created activity-object hypergraph. Based on this, downstream general matching tasks can be performed using the obtained hypergraph feature representations of activity schemes and activity objects. General matching tasks include, but are not limited to: matching tasks: outputting matching entities (such as matched activity schemes or activity objects) within a certain distance range in the hypergraph feature representation space; recommendation tasks: calculating a scalar recommendation value using the hypergraph feature representation and outputting a list of the top N recommendation values; search tasks: vectorizing the search input using an NLP model, then calculating the distance between the input and the hypergraph feature representation of the matching entity, and using matching entities within a certain distance range as search results, and other tasks.

[0070] This application solution can be applied to a variety of real-world scenarios, where the roles of the activity plan and the participants differ.

[0071] For example, in a government department setting, policy proposals issued by government departments can serve as activity plans, with businesses as the target audience. The matching relationship between the activity plan and the target audience represents the policy implementation relationship between the policy plan and the business. In the past, the implementation of government policies involved a time-consuming and cumbersome process for businesses, from receiving policy notifications, understanding the policy content, applying for policy support, to finally receiving actual support. To change the traditional way businesses seek policies, the ideal approach is for policies to proactively reach out to businesses, meaning government departments should actively recommend suitable policy plans to businesses. However, in practice, this relies on government staff collecting data offline and manually screening businesses, which is not only time-consuming and labor-intensive but also risky. To provide more effective and precise policy support to businesses, and to enable government departments to serve businesses in a more intelligent way, a precise policy implementation method is urgently needed. The solution provided in this application can effectively solve the above problems, helping government departments to implement precise policies and improve office efficiency.

[0072] For example, in a commercial setting, the manufacturer's marketing strategy can be used as an activity plan, targeting specific consumer groups such as students or businesses. Previously, manufacturers needed to widely disseminate their marketing strategies through multiple channels, such as advertisements, flyers, and SMS, which was costly and had a low hit rate with the target audience. Furthermore, for users, selecting a suitable marketing strategy was time-consuming and laborious. The solution provided in this application allows for a more precise match between marketing strategies and target audiences, bringing significant convenience to both manufacturers and participants.

[0073] The method and process provided in this application will be described below in conjunction with the aforementioned application scenarios.

[0074] The method provided in this application can be divided into an activity-object hypergraph creation phase and a subject matching phase based on the created activity-object hypergraph. These two phases can be deployed on the same device or on different devices. For example, the activity-object hypergraph creation phase can be deployed in the cloud or on a server, while the subsequent subject matching phase can be deployed on smart terminals such as mobile phones, tablets, and computers.

[0075] For ease of understanding, this application describes the creation phase of the activity-object hypergraph and the subject matching phase based on the activity-object hypergraph.

[0076] I. Activity - Creation Phase of Object Hypergraph

[0077] Combination Figure 1This application provides a method for creating a hypergraph, which may include the following steps:

[0078] Step S100: Obtain information on each published activity plan, information on each activity target, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents the pairing information consisting of the activity target and the activity plan it has participated in.

[0079] Specifically, in different application scenarios, information on all published activity plans, all participants in the activity plans, and potential participants can be collected. Furthermore, activity plan implementation information can be obtained, namely, the pairing information of participants in the activity plans and the activity plans they have participated in.

[0080] Taking government affairs scenarios as an example, it can collect information on all policy proposals issued by government departments, as well as information on all enterprises that may apply for policy proposals. At the same time, it can also obtain policy implementation information; that is, for enterprises that have already applied for policy proposals, the relationship between the applied policy proposals and the enterprises' policy implementation can be obtained as prior data.

[0081] It should be noted that all the information obtained above can be obtained through public and legal channels.

[0082] Step S110: Establish an activity-object pairing relationship diagram based on the activity plan implementation information.

[0083] Specifically, the activity plans and activity objects obtained in the previous step can be placed as independent nodes in the activity-object pairing graph. At the same time, the activity plan implementation information discloses the pairing information of the activity objects and the activity plans they have participated in. Therefore, activity plan nodes and activity object nodes that have been determined to have a pairing relationship can be connected by edges, that is, a low-order matching relationship feature between activity plan nodes and activity object nodes can be established.

[0084] like Figure 2 This example illustrates an activity-object pairing graph. Figure 2 The example uses policy proposals as activity plans and enterprises as the activity targets. Nodes with policy implementation relationships (i.e., enterprises have submitted policy proposals) are connected by policy implementation relationship edges.

[0085] Step S120: Extract the attribute values ​​of each preset higher-order attribute feature from each of the activity scheme information and each of the activity object information.

[0086] It should be noted that step S120 can be executed in any order after step S100 and before step S130. Figure 1 Only one optional execution order is shown.

[0087] In matching problems, the matching relationship between matching entities is the most basic relationship, also known as low-order correlation, which is a pairwise relationship. Attributes, on the other hand, are a common abstract feature; matching entities with the same attributes usually exhibit similarity, i.e., high-order correlation, which is a grouped relationship. Traditional techniques use ordinary graph structures for modeling. Since edges in ordinary graph structures can only connect two nodes, they can only model one-to-one low-order matching relationships and cannot model such grouped high-order correlation relationships. In reality, multiple matching entities may possess the same attribute features, and these high-order correlation features are crucial for accurately predicting the matching relationships between entities. Taking the matching between policy proposals and enterprise targets as an example, ordinary graph structures can only model the pairwise policy implementation relationship between policy proposals and enterprise targets, but cannot model the high-order correlation features of policy proposals and enterprise targets, such as regional and industry-specific features. However, the design and implementation of policy activities have obvious regional and industry-specific characteristics. Government agencies typically formulate personalized policies to promote development for specific industries in a particular region; therefore, such high-order correlation features are crucial for precise policy implementation.

[0088] Therefore, in this embodiment, based on modeling the low-order matching relationship between activity plans and activity objects, it is desirable to further model high-order attribute features. Thus, this step first extracts the attribute values ​​of each preset high-order attribute feature from the activity plan information and activity object information.

[0089] Taking policy proposals and target enterprises as examples, we can extract the attribute values ​​of higher-order characteristics for each policy proposal, such as the applicable geographical location and applicable industry type. We can also extract the location of the target enterprise and the industry type it belongs to.

[0090] Of course, the preset high-level attribute features can include not only industry features and regional features, but also other types of features, which can be selected according to the application scenario of the solution.

[0091] Step S130: On the activity-object pairing relationship graph, construct a hyperedge for each attribute value of each preset higher-order attribute feature, and connect the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph.

[0092] Specifically, a hypergraph is a generalized graph structure consisting of a set of nodes and hyperedges. Unlike ordinary graph structures where an edge can only connect two nodes, a hyperedge can connect any number of nodes. Compared to ordinary graph structures that can only model pairwise matching relationships, hypergraphs have significant advantages in modeling higher-order attribute features.

[0093] In this step, an attribute-based hypergraph construction method can be used to construct hyperedges on the activity-object pairing graph. Each attribute value of each higher-order attribute feature can construct a hyperedge, and then the nodes with that attribute value (including activity scheme nodes and activity object nodes) are connected through the hyperedge to obtain the activity-object hypergraph.

[0094] exist Figure 2 The example activity-object pairing graph uses the method in step S130 to construct hyperedges, resulting in the activity-object hypergraph as shown below. Figure 3 As shown, nodes with the same higher-order attribute eigenvalues ​​are connected by hyperedges. Figure 3 Each node within a hyperedge can be considered to be connected by the hyperedge, and different hyperedges represent different higher-order attribute feature values.

[0095] Furthermore, the weights of the hyperedges can be assigned based on the heat kernel method.

[0096] The hypergraph creation method provided in this application places activity scheme nodes and activity object nodes in the hypergraph, and connects activity scheme nodes and activity object nodes with known matching relationships based on prior data through edges. This model can model low-order matching relationships between activity schemes and activity objects. Furthermore, for each node in the hypergraph (including activity scheme nodes and activity object nodes), feature values ​​of high-order attribute features can be extracted. Nodes with the same high-order attribute feature values ​​are then connected through the same hyperedge, thereby modeling the high-order attribute features between nodes. Through the above-described activity-object hypergraph, hypergraph feature representations of activity schemes and activity objects can be learned using hypergraph convolution techniques. Compared to traditional methods, the feature representations obtained are more accurate, further improving the accuracy of matching relationship prediction when applied to downstream matching tasks.

[0097] In some embodiments of this application, it is further considered that the matching subjects in the matching task may have obvious temporal characteristics. For example, the creation and existence of matching subjects have temporal features. On the one hand, the creation of matching subjects may be inherited (such as product upgrades and iterations, annual policies, etc.), and on the other hand, the existence time of matching subjects may span a long period of time, existing continuously in multiple time periods. Taking the matching between policy schemes and enterprise objects as an example, the implementation of policy schemes has a fixed time and similar policy schemes have continuity. At the same time, the growth of enterprises also has temporal characteristics. Enterprises may match different policies at different development stages. Therefore, the temporal nature of the matching of activity schemes and activity objects can be further considered.

[0098] Therefore, this embodiment further provides another method for creating hypergraphs, which can realize the creation of activity-object spatiotemporal hypergraphs. Combined with... Figure 4 The specific steps include:

[0099] Step S200: Obtain information on each published activity plan, information on each activity target, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents the pairing information consisting of the activity target and the activity plan it has participated in.

[0100] Step S210: Divide the activity plan information, the activity object information, and the activity plan implementation information into several different time periods.

[0101] Each time period includes information on the activity plan implemented within that time period, information on the existing activity participants, and information on the implementation of the activity plan.

[0102] The length of the time period can be selected based on the amount of information obtained from the above steps and other strategies, such as dividing the time period by day, month, or year.

[0103] Step S220: Based on the activity plan information, activity object information, and activity plan implementation information for each time period, establish an activity-object pairing relationship diagram for each time period.

[0104] The process of establishing the activity-object pairing relationship diagram for each time period can be described with reference to the aforementioned embodiments. The only difference is that the information referenced in this step is the information within the corresponding time period, while the information referenced in the aforementioned steps is the full amount of information.

[0105] Step S230: Extract the attribute values ​​of each preset higher-order attribute feature from each of the activity scheme information and each of the activity object information.

[0106] Step S240: On the activity-object pairing relationship graph of each time period, construct a hyperedge for each attribute value of each preset higher-order attribute feature, and connect the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph of each time period.

[0107] Step S250: Connect the nodes of the same activity object and the nodes of the same activity scheme in the activity-object hypergraph of adjacent time periods to obtain the activity-object spatiotemporal hypergraph.

[0108] A spatiotemporal hypergraph is a combination of a spatiotemporal graph and a hypergraph, that is, establishing temporal connections between activity-object hypergraphs at different time periods to obtain an activity-object spatiotemporal hypergraph. In this application, the activity-object spatiotemporal hypergraph can be considered as a type of activity-object hypergraph.

[0109] Specifically, according to the chronological order of time periods, nodes of the same activity object and nodes of the same activity scheme are connected in the activity-object hypergraph of two adjacent time periods. Optionally, depending on the needs of the actual scenario, two activity schemes that have continuity can be identified as the same activity scheme node. For example, for policy schemes, two policy schemes that are released successively in time may have continuity, and these two policy schemes can be identified as the same policy scheme.

[0110] Reference Figure 5 This example illustrates an activity-object spatiotemporal hypergraph. It includes activity-object hypergraphs spanning n time periods from T1 to Tn, which are formed by temporally connecting nodes with the same activity scheme and nodes with the same activity object through temporal relationship edges.

[0111] The hypergraph creation method provided in this application constructs a spatiotemporal hypergraph based on the information obtained by time decomposition, further considering temporal information to model the low-order matching relationship features, high-order attribute features, and temporal features of activity plans and activity objects. When applied to policy plan and enterprise object matching scenarios, it can effectively model the temporal, regional, and industry-specific characteristics existing in policy-enterprise policy implementation matching tasks.

[0112] II. The stage of subject matching based on the activity-object hypergraph

[0113] Based on the creation of the activity-object hypergraph (including the activity-object spatiotemporal hypergraph) described in the above embodiments, the task of predicting downstream matching relationships is performed based on the created activity-object hypergraph.

[0114] Reference Figure 6 This example demonstrates a method for matching activity plans with objects, which includes the following steps:

[0115] Step S300: Based on the configured activity-object hypergraph, obtain the hypergraph feature representations of the activity scheme and the activity object respectively.

[0116] The configured activity-object hypergraph can adopt the aforementioned method. Figure 1 The activity-object hypergraph obtained by the hypergraph creation method in the corresponding embodiment. The activity-object hypergraph includes activity scheme nodes and activity object nodes. Activity scheme nodes and activity object nodes that are known to have a matching relationship are connected by edges. Activity scheme nodes and / or activity object nodes with the same higher-order attribute feature values ​​are connected by the same hyperedge.

[0117] In this step, hypergraph convolution operations can be used to learn the hypergraph feature representations of each node from the activity-object hypergraph, that is, to obtain the hypergraph feature representations of the activity scheme and the activity object respectively. The hypergraph feature representation includes both low-order matching relationship features and high-order attribute features.

[0118] Step S310: Calculate the matching degree between the activity plan and the activity object based on their respective hypergraph feature representations.

[0119] Specifically, the hypergraph feature representation of the activity scheme and the hypergraph feature representation of the activity object can be subjected to a dot product operation to obtain the matching value between the activity scheme and the activity object.

[0120] Step S320: Determine the matching relationship between the activity plan and the activity object according to the matching degree.

[0121] Depending on the specific application requirements, this step can either match the target activity plan with the corresponding activity object, or match the target activity object with the corresponding activity plan. For example:

[0122] For any target activity plan, sort the activities in descending order according to the matching value between the target activity plan and each activity object, and return the top N1 activities with the highest ranking as the activity objects matched by the target activity plan. N1 can be set according to user needs.

[0123] For any target activity object, sort the activity schemes in descending order according to the matching value between the target activity object and each activity scheme, and return the top N2 activity schemes with the highest ranking as the activity schemes matched with the target activity object. N2 can be set according to user needs.

[0124] Taking the matching process between policy proposals and target enterprises as an example, the matching value between the target policy proposal and all enterprises can be calculated, sorted from largest to smallest according to the matching value, and the top N1 enterprises with the highest ranking are returned as the policy matching recommendation results.

[0125] The activity scheme and object matching method provided in this application embodiment is based on a pre-established activity-object hypergraph. Through hypergraph convolution technology, the hypergraph feature representations of the activity scheme and the activity object can be learned. The hypergraph feature representation integrates low-order matching relationship features and high-order attribute features between the activity scheme and the activity object, thereby obtaining high-quality feature representations of the activity scheme and the activity object. Furthermore, the matching degree between the activity scheme and the activity object can be calculated based on the obtained hypergraph feature representation, and the matching relationship between the activity scheme and the activity object can be determined according to the matching degree, thereby improving the accuracy of matching relationship prediction.

[0126] When applied to the matching process between policy proposals and enterprise objects, the hypergraph feature representation of policy proposals and enterprise objects obtained by the proposed solution based on the activity-object hypergraph integrates low-order matching relationship features and high-order attribute features (such as regional features and industry features) of policies and enterprises, thereby obtaining high-quality policy and enterprise feature representations and achieving better matching and recommendation results.

[0127] In some embodiments of this application, the process of obtaining the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph in step S300 is described.

[0128] Combination Figure 7 As shown:

[0129] In the activity-object hypergraph, a standard graph convolution operation is performed on the matching relationships between activity schemes and activity objects, aggregating the neighbor node information of nodes to obtain feature representations of activity scheme nodes and activity object nodes containing low-order matching relationship features. That is, for each node in the hypergraph, the neighbor node information of that node is aggregated to obtain the low-order features of each node.

[0130] In the activity-object hypergraph, for each hyperedge, a hypergraph convolution operation is performed to aggregate the information of all nodes connected to the hyperedge to obtain the feature representation of each hyperedge. Then, based on the membership relationship between nodes and hyperedges, for each node, the feature representations of each hyperedge associated with the node are aggregated to obtain the higher-order features of each node.

[0131] Finally, the low-order and high-order features of each node are combined to form the hypergraph feature representation of each node.

[0132] Furthermore, when the activity-object hypergraph uses the aforementioned... Figure 4 When the activity-object spatiotemporal hypergraph obtained by the hypergraph creation method in the corresponding embodiment is an activity-object spatiotemporal hypergraph, the activity-object spatiotemporal hypergraph includes multiple activity-object hypergraphs. Different activity-object hypergraphs correspond to different time periods. The corresponding activity-object hypergraph is created by collecting activity plan information, activity object information and activity plan implementation information within the corresponding time period. The activity-object hypergraphs corresponding to each time period are connected in chronological order to form an activity-object spatiotemporal hypergraph.

[0133] Based on this, the process of obtaining the hypergraph feature representations of activity schemes and activity objects, using this activity-object spatiotemporal hypergraph, can be combined with... Figure 8 As shown, it can specifically include:

[0134] In the activity-object hypergraph corresponding to each time period, determine the hypergraph feature representation of each node. The specific process can be found in the previous introduction and will not be repeated here. For the activity-object hypergraph across n time periods T1-Tn, determine the hypergraph feature representation of each node.

[0135] Furthermore, for each node, the hypergraph feature representations of the node in each time period are sequentially input into the pre-trained recurrent neural network according to the timeline, and the last hidden state of the recurrent neural network is taken as the spatiotemporal hypergraph feature representation of the node.

[0136] In the training phase, the recurrent neural network can use the hidden state of the last layer of the recurrent neural network as the spatiotemporal hypergraph feature representation of the node, and then perform inference calculation for the downstream matching task based on the spatiotemporal hypergraph feature representation of the node. The pre-labeled matching results are used as sample labels, and the value of the loss function between the inference calculation result and the sample label is calculated. The parameters of the recurrent neural network are updated according to the value of the loss function to obtain the trained recurrent neural network.

[0137] In this embodiment, by sequentially inputting the hypergraph feature representations of a node at different time periods into a recurrent neural network, temporal features can be learned. The last hidden state is taken as the spatiotemporal hypergraph feature representation of the node, which integrates low-order matching relationship features, high-order attribute features, and time dependency features between the activity scheme and the activity object, thus obtaining a high-quality feature representation of the activity scheme and the activity object. Furthermore, the degree of matching between the activity scheme and the activity object can be calculated based on the obtained hypergraph feature representation, and the matching relationship between the activity scheme and the activity object can be determined according to the degree of matching, thereby improving the accuracy of the matching relationship prediction.

[0138] When applied to the matching process between policy proposals and enterprise objects, the hypergraph feature representation of policy proposals and enterprise objects obtained by the proposed scheme based on the activity-object spatiotemporal hypergraph integrates low-order matching relationship features, high-order attribute features (such as regional features, industry features, etc.) and time-dependent features of policies and enterprises, thereby obtaining high-quality policy and enterprise feature representations and achieving better matching and recommendation results.

[0139] Further reference Figure 9 Taking the matching process between policy proposals and enterprise objects as an example, it illustrates an overall framework diagram for creating a spatiotemporal hypergraph and performing policy and enterprise matching tasks based on it.

[0140] First, obtain and process information on the published activity plans, the target audience for each activity, and the implementation information of the activity plans, such as extracting feature values ​​of preset high-order attribute features.

[0141] Furthermore, an activity-object spatiotemporal hypergraph is created based on the acquired information and processing results. Then, the hypergraph feature representations of the same node at different time periods are calculated based on the activity-object spatiotemporal hypergraph, and these representations are sequentially input into a recurrent neural network. The last hidden state of the recurrent neural network is taken as the spatiotemporal hypergraph feature representation of the node.

[0142] The spatiotemporal hypergraph feature representations of policies and enterprises are subjected to dot product operation to calculate matching values. The matching values ​​are then sorted according to their magnitude, and the top N values ​​with the highest values ​​are selected as the matching recommendation results.

[0143] The following describes the hypergraph creation apparatus provided in the embodiments of this application. The hypergraph creation apparatus described below can be referred to in correspondence with the hypergraph creation method described above.

[0144] See Figure 10 , Figure 10 This is a schematic diagram of a supergraph creation device disclosed in an embodiment of this application.

[0145] like Figure 10 As shown, the device may include:

[0146] Information acquisition unit 11 is used to acquire information on each published activity plan, information on each activity object, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents pairing information consisting of activity objects and the activity plans they have participated in.

[0147] The relationship graph establishment unit 12 is used to establish an activity-object pairing relationship graph based on the activity plan implementation information. The activity-object pairing relationship graph includes each activity plan node and each activity object node, and the activity plan nodes and activity object nodes that have a pairing relationship are connected by edges.

[0148] The attribute value extraction unit 13 is used to extract the attribute value of each preset high-order attribute feature from each of the activity scheme information and each of the activity object information.

[0149] The relationship graph editing unit 14 is used to construct a hyperedge for each attribute value of each preset higher-order attribute feature on the activity-object pairing relationship graph, and connect the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph.

[0150] Optionally, the above-mentioned hypergraph creation apparatus may further include:

[0151] The data partitioning unit is used to divide the acquired activity plan information, activity object information, and activity plan implementation information into several different time periods before the relationship graph building unit processes them. Each time period contains activity plan information implemented within that time period, existing activity object information, and activity plan implementation information. Based on this, the relationship graph building unit is specifically used to: build an activity-object pairing relationship graph for each time period based on the activity plan information, activity object information, and activity plan implementation information for each time period. Further, the relationship graph editing unit is specifically used to: construct a hyperedge for each attribute value of each preset higher-order attribute feature on the activity-object pairing relationship graph for each time period, and connect activity plan nodes and / or activity objects with the attribute values ​​through the hyperedges to obtain an activity-object hypergraph for each time period; connect the same activity object nodes and the same activity plan nodes in the activity-object hypergraphs of adjacent time periods to obtain an activity-object spatiotemporal hypergraph.

[0152] Optionally, the above-mentioned activity plan includes policy plans, and the target of the activity includes enterprises; the higher-level attribute characteristics include industry characteristics and / or regional characteristics.

[0153] Furthermore, the activity scheme and object matching device provided in the embodiments of this application will be described. The activity scheme and object matching device described below can be referred to in correspondence with the activity scheme and object matching method described above.

[0154] See Figure 11 , Figure 11 This is a schematic diagram of an activity scheme and object matching device disclosed in an embodiment of this application.

[0155] like Figure 11 As shown, the device may include:

[0156] The feature representation acquisition unit 21 is used to acquire the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph. The activity-object hypergraph includes activity scheme nodes and activity object nodes. Activity scheme nodes and activity object nodes that are known to have a matching relationship are connected by an edge. Activity scheme nodes and / or activity object nodes with the same higher-order attribute feature values ​​are connected by the same hyperedge.

[0157] The matching degree calculation unit 22 is used to calculate the matching degree between the activity plan and the activity object based on the hypergraph feature representations of the activity plan and the activity object respectively;

[0158] The matching relationship determination unit 23 is used to determine the matching relationship between the activity plan and the activity object according to the matching degree.

[0159] Optionally, the process by which the feature representation acquisition unit acquires the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph includes:

[0160] In the activity-object hypergraph, for each node, the neighboring node information of the node is aggregated to obtain the low-order features of each node;

[0161] In the activity-object hypergraph, for each hyperedge, the information of all nodes connected by the hyperedge is aggregated to obtain the feature representation of each hyperedge;

[0162] Based on the dependency relationship between nodes and hyperedges, for each node, the feature representations of each hyperedge associated with the node are aggregated to obtain the higher-order features of each node.

[0163] The low-order and high-order features of each node are combined to form the hypergraph feature representation of each node.

[0164] Optionally, the above-configured activity-object hypergraph can be multiple, with different activity-object hypergraphs corresponding to different time periods. The corresponding activity-object hypergraph is created from activity plan information, activity object information, and activity plan implementation information collected within the corresponding time period. The activity plan implementation information represents pairing information consisting of activity objects and the activity plans they have participated in. The activity-object hypergraphs corresponding to each time period form an activity-object spatiotemporal hypergraph. Based on this, the process by which the feature representation acquisition unit acquires the hypergraph feature representations of activity plans and activity objects based on the configured activity-object hypergraphs includes:

[0165] In the activity-object hypergraph corresponding to each time period, determine the hypergraph feature representation of each node;

[0166] For each node, the hypergraph feature representations of the node in each time period are sequentially input into a pre-trained recurrent neural network according to the timeline, and the last hidden state of the recurrent neural network is taken as the spatiotemporal hypergraph feature representation of the node.

[0167] Optionally, the matching degree calculation unit above calculates the matching degree between the activity scheme and the activity object based on their respective hypergraph feature representations, including:

[0168] The matching value between the activity plan and the activity object is obtained by performing a dot product operation on the hypergraph feature representation of the activity plan and the hypergraph feature representation of the activity object. Based on this, the matching relationship determination unit determines the matching relationship between the activity plan and the activity object according to the matching degree, including:

[0169] For any target activity plan, sort the activities in descending order according to the matching value between the target activity plan and each activity object, and return the top N1 activities with the highest ranking as the activity objects matched by the target activity plan.

[0170] or,

[0171] For any target activity object, sort them in descending order according to the matching value between the target activity object and each activity plan, and return the top N2 activity plans with the highest ranking as the activity plans matched by the target activity object.

[0172] The hypergraph creation device, activity scheme, and object matching device provided in this application can be applied to the same or different electronic devices. Taking their application to the same electronic device as an example... Figure 12 The hardware structure block diagram of the electronic device is shown. The hardware structure of the electronic device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4.

[0173] In this embodiment, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4.

[0174] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0175] Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0176] The memory stores a program that the processor can call. The program is used to: execute the steps of the hypergraph creation method, or execute the steps of the activity scheme and object matching method.

[0177] This application embodiment also provides a storage medium that can store a program suitable for execution by a processor, the program being used to: execute various steps of a hypergraph creation method, or execute various steps of an activity scheme and object matching method.

[0178] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0179] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0180] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An activity program and subject matching method, characterized by, include: Based on the configured activity-object hypergraph, the hypergraph feature representations of activity schemes and activity objects are obtained. The activity-object hypergraph includes activity scheme nodes and activity object nodes. Activity scheme nodes and activity object nodes that are known to have a matching relationship are connected by edges. Activity scheme nodes and / or activity object nodes with the same higher-order attribute feature values ​​are connected by the same hyperedge. Based on the hypergraph feature representations of the activity plan and the activity object respectively, the matching degree between the activity plan and the activity object is calculated; Based on the degree of matching, the matching relationship between the activity plan and the activity object is determined; The process of obtaining hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph includes: In the activity-object hypergraph, for each node, the neighboring node information of the node is aggregated to obtain the low-order features of each node; In the activity-object hypergraph, for each hyperedge, the information of all nodes connected by the hyperedge is aggregated to obtain the feature representation of each hyperedge; Based on the dependency relationship between nodes and hyperedges, for each node, the feature representations of each hyperedge associated with the node are aggregated to obtain the higher-order features of each node. The low-order and high-order features of each node are combined to form the hypergraph feature representation of each node.

2. The method of claim 1, wherein, There are multiple activity-object hypergraphs, and different activity-object hypergraphs correspond to different time periods. The corresponding activity-object hypergraph is created by the activity plan information, activity object information and activity plan implementation information collected within the corresponding time period. The activity plan implementation information represents the pairing information consisting of activity objects and the activity plans they have participated in. The activity-object hypergraphs corresponding to each time period constitute the activity-object spatiotemporal hypergraph.

3. The method of claim 2, wherein, The process of obtaining hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph includes: In the activity-object hypergraph corresponding to each time period, determine the hypergraph feature representation of each node; For each node, the hypergraph feature representations of the node in each time period are sequentially input into a pre-trained recurrent neural network according to the timeline, and the last hidden state of the recurrent neural network is taken as the spatiotemporal hypergraph feature representation of the node.

4. The method of claim 1, wherein, Based on the hypergraph feature representations of the activity plan and the activity object respectively, the matching degree between the activity plan and the activity object is calculated, including: Perform a dot product operation on the hypergraph feature representation of the activity scheme and the hypergraph feature representation of the activity object to obtain the matching value between the activity scheme and the activity object; Then, based on the degree of matching, the matching relationship between the activity plan and the activity object is determined, including: For any target activity plan, sort the activities in descending order according to the matching value between the target activity plan and each activity object, and return the top N1 activities with the highest ranking as the activity objects matched by the target activity plan. or, For any target activity object, sort them in descending order according to the matching value between the target activity object and each activity plan, and return the top N2 activity plans with the highest ranking as the activity plans matched by the target activity object.

5. The method according to any one of claims 1 to 4, characterized in that, The activity plan includes policy plans, and the target audience includes enterprises. The higher-order attribute features include industry characteristics and / or regional characteristics.

6. A method for creating a hypergraph, characterized in that, Includes: Obtain information on each published activity plan, information on each activity target, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents pairing information consisting of activity targets and the activity plans they have participated in. Based on the activity plan implementation information, an activity-object pairing relationship graph is established. The activity-object pairing relationship graph contains each activity plan node and each activity object node, and the activity plan nodes and activity object nodes that have a pairing relationship are connected by edges. Extract the attribute value of each preset higher-order attribute feature from each of the activity plan information and each of the activity object information; On the activity-object pairing graph, a hyperedge is constructed for each attribute value of each preset higher-order attribute feature, and the activity scheme nodes and / or activity objects with the attribute values ​​are connected through the hyperedge to obtain the activity-object hypergraph.

7. The method according to claim 6, characterized in that, Before creating the activity-object pairing graph, the following is also included: The acquired activity plan information, activity object information, and activity plan implementation information are divided into several different time periods. Each time period contains the activity plan information implemented within that time period, the existing activity object information, and the activity plan implementation information. The process of establishing an activity-object pairing relationship diagram based on the activity plan implementation information includes: Based on the activity plan information, activity target information, and activity plan implementation information for each time period, establish an activity-target pairing relationship diagram for each time period; Then, the process of constructing a hyperedge for each attribute value of each preset higher-order attribute feature on the activity-object pairing graph, and connecting the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph includes: In the activity-object pairing graph of each time period, a hyperedge is constructed for each attribute value of each preset higher-order attribute feature, and the activity scheme nodes and / or activity objects with the attribute values ​​are connected through the hyperedge to obtain the activity-object hypergraph of each time period. Connecting nodes of the same activity object and nodes of the same activity scheme in the activity-object hypergraph of adjacent time periods yields the activity-object spatiotemporal hypergraph.

8. The method according to claim 6, characterized in that, The activity plan includes policy plans, and the target audience includes enterprises. The higher-order attribute features include industry characteristics and / or regional characteristics.

9. An activity plan and object matching device, characterized in that, include: The feature representation acquisition unit is used to acquire the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph. The activity-object hypergraph includes activity scheme nodes and activity object nodes. Activity scheme nodes and activity object nodes with known matching relationships are connected by edges. Activity scheme nodes and / or activity object nodes with the same higher-order attribute feature values ​​are connected by the same hyperedge. The process of acquiring the hypergraph feature representations of activity schemes and activity objects based on the configured activity-object hypergraph includes: In the activity-object hypergraph, for each node, the neighboring node information of the node is aggregated to obtain the low-order features of each node; In the activity-object hypergraph, for each hyperedge, the information of all nodes connected by the hyperedge is aggregated to obtain the feature representation of each hyperedge; Based on the dependency relationship between nodes and hyperedges, for each node, the feature representations of each hyperedge associated with the node are aggregated to obtain the higher-order features of each node. The low-order and high-order features of each node are combined to form the hypergraph feature representation of each node; The matching degree calculation unit is used to calculate the matching degree between the activity plan and the activity object based on the hypergraph feature representations of the activity plan and the activity object respectively; The matching relationship determination unit is used to determine the matching relationship between the activity plan and the activity object according to the matching degree.

10. A hypergraph creation apparatus, characterized in that, include: The information acquisition unit is used to acquire information on each published activity plan, information on each activity target, and information on the implementation of the activity plan. The information on the implementation of the activity plan represents pairing information consisting of activity targets and the activity plans they have participated in. The relationship graph establishment unit is used to establish an activity-object pairing relationship graph based on the activity plan implementation information. The activity-object pairing relationship graph includes each activity plan node and each activity object node, and the activity plan nodes and activity object nodes that have a pairing relationship are connected by edges. The attribute value extraction unit is used to extract the attribute value of each preset high-order attribute feature from each of the activity scheme information and each of the activity object information. The relationship graph editing unit is used to construct a hyperedge for each attribute value of each preset higher-order attribute feature on the activity-object pairing relationship graph, and connect the activity scheme nodes and / or activity objects with the attribute values ​​through the hyperedge to obtain the activity-object hypergraph.

11. An electronic device, characterized in that, include: Memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the steps of the activity scheme and object matching method as described in any one of claims 1 to 5, or to implement the steps of the hypergraph creation method as described in any one of claims 6 to 8.

12. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the activity scheme and object matching method as described in any one of claims 1 to 5, or implements the steps of the hypergraph creation method as described in any one of claims 6 to 8.

Images