Object matching model training method and object matching method

Abstract

The application discloses a kind of object matching model training method and object matching method, belong to internet technical field.Method includes: obtaining reference interaction network;Based on reference interaction network, from multiple providing objects, determine relevant providing object, the possibility that sample receiving object and relevant providing object interact is greater than first threshold value;Through first network model, based on the basic information of sample receiving object, the basic information of sample providing object and the basic information of relevant providing object, determine the predicted matching result between sample receiving object and sample providing object;Based on predicted matching result and labeled matching result, first network model is trained, and object matching model is obtained.The basic information of relevant providing object contains the information that each providing object common with sample receiving object interacts, and the accuracy of predicted matching result determined based on the common information is higher, so the accuracy of object matching model is higher.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of Internet technology, and in particular to a training method of an object matching model and an object matching method. BACKGROUND

[0002] In the technical field of Internet technology, object matching technology is widely used in recommendation, search and other scenarios. For example, in a search scenario, a providing object that matches a search word of a receiving object can be determined, and information of the providing object is displayed as a search result to the receiving object.

[0003] In related technology, object matching is performed by an object matching model. The training method of the object matching model includes: determining, by a neural network model, a predicted matching result between a sample receiving object and a sample providing object based on basic information of the sample receiving object and basic information of the sample providing object. An annotated matching result between the sample receiving object and the sample providing object is obtained by annotation. The neural network model is trained based on the predicted matching result and the annotated matching result to obtain the object matching model.

[0004] The object matching model trained by the above technology has low accuracy, which affects the accuracy of the matching result. SUMMARY

[0005] The present application provides a training method of an object matching model and an object matching method, which can be used to solve the problems in related technology. The technical solution includes the following contents.

[0006] In one aspect, a training method of an object matching model is provided, which includes:

[0007] Obtaining a reference interaction network, the reference interaction network including a plurality of nodes, any node being a node of a receiving object or a node of a providing object, an edge between any receiving object node and any providing object node representing the number of interactions between the any receiving object and the any providing object, and a plurality of receiving objects including a sample receiving object;

[0008] Determining, based on the reference interaction network, a related providing object from the plurality of providing objects, the sample receiving object having a probability greater than a first threshold value of interacting with the related providing object;

[0009] Determining, by a first network model, a predicted matching result between the sample receiving object and a sample providing object based on basic information of the sample receiving object, basic information of the sample providing object and basic information of the related providing object;

[0010] Based on the predicted matching results and the labeled matching results, the first network model is trained to obtain an object matching model. The labeled matching results are the matching results between the sample receiving object and the sample providing object obtained through labeling. The object matching model is used to determine the matching results between the target receiving object and the target providing object.

[0011] On the other hand, an object matching method is provided, the method comprising:

[0012] Obtain a reference interaction network, which includes multiple nodes, any node being either a receiving object node or a providing object node, and the edge between any receiving object node and any providing object node represents the number of interactions between the receiving object and the providing object, and the multiple receiving objects include a target receiving object;

[0013] Based on the reference interaction network, a reference providing object is determined from the plurality of providing objects, and the probability that the target receiving object interacts with the reference providing object is greater than a first threshold.

[0014] Using an object matching model, based on the basic information of the target receiving object, the basic information of the target providing object, and the basic information of the reference providing object, the matching result between the target receiving object and the target providing object is determined. The object matching model is trained according to any of the object matching model training methods described above.

[0015] On the other hand, a training apparatus for an object matching model is provided, the apparatus comprising:

[0016] The acquisition module is used to acquire a reference interaction network, which includes multiple nodes, any node being either a receiving object node or a providing object node, and the edge between any receiving object node and any providing object node represents the number of interactions between the receiving object and the providing object, and the multiple receiving objects include sample receiving objects.

[0017] The determining module is used to determine a relevant providing object from the plurality of providing objects based on the reference interaction network, wherein the probability that the sample receiving object interacts with the relevant providing object is greater than a first threshold.

[0018] The determining module is further configured to determine the prediction matching result between the sample receiving object and the sample providing object based on the basic information of the sample receiving object, the basic information of the sample providing object, and the basic information of the related providing object through the first network model;

[0019] The training module is used to train the first network model based on the predicted matching result and the labeled matching result to obtain an object matching model. The labeled matching result is the matching result between the sample receiving object and the sample providing object obtained through labeling. The object matching model is used to determine the matching result between the target receiving object and the target providing object.

[0020] On the other hand, an object matching apparatus is provided, the apparatus comprising:

[0021] The acquisition module is used to acquire a reference interaction network, which includes multiple nodes, any node being either a receiving object node or a providing object node, and the edge between any receiving object node and any providing object node represents the number of interactions between the receiving object and the providing object, and the multiple receiving objects include a target receiving object.

[0022] The determining module is configured to determine a reference providing object from the plurality of providing objects based on the reference interaction network, wherein the probability that the target receiving object interacts with the reference providing object is greater than a first threshold.

[0023] The determining module is further configured to determine the matching result between the target receiving object and the target providing object based on the basic information of the target receiving object, the basic information of the target providing object, and the basic information of the reference providing object through an object matching model, wherein the object matching model is trained according to any of the above-described object matching model training methods.

[0024] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the training method or object matching method of any of the object matching models described above.

[0025] On the other hand, a computer-readable storage medium is also provided, wherein at least one computer program is stored in the computer-readable storage medium, the at least one computer program being loaded and executed by a processor to enable an electronic device to implement the training method or object matching method of any of the object matching models described above.

[0026] On the other hand, a computer program or computer program product is also provided, wherein at least one computer program is stored in the computer program or computer program product, and the at least one computer program is loaded and executed by a processor to enable an electronic device to implement the training method or object matching method of any of the above-mentioned object matching models.

[0027] The technical solution provided in this application brings at least the following beneficial effects:

[0028] In the technical solution provided in this application, a reference interaction network records the interaction information between the receiving object and the providing object. Based on the reference interaction network, relevant providing objects with a high probability of interacting with the sample receiving object are identified from multiple providing objects. This ensures that the basic information of the relevant providing objects includes information common to all providing objects interacting with the sample receiving object. The predicted matching results between the sample receiving object and the sample providing object determined based on this common information, the basic information of the sample receiving object, and the basic information of the sample providing object have high accuracy. Therefore, the object matching model trained based on the predicted matching results and the labeled matching results has high accuracy, thereby improving the accuracy of the object matching results. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0030] Figure 1 This is a schematic diagram of a training method for an object matching model or an implementation environment for an object matching method provided in an embodiment of this application;

[0031] Figure 2 This is a flowchart of a training method for an object matching model provided in an embodiment of this application;

[0032] Figure 3 This is a schematic diagram of a reference interactive network provided in an embodiment of this application;

[0033] Figure 4 This is a schematic diagram of another reference interactive network provided in an embodiment of this application;

[0034] Figure 5 This is a schematic diagram illustrating the determination of a third splicing feature provided in an embodiment of this application;

[0035] Figure 6 This is a schematic diagram illustrating the determination of a predicted matching result provided in an embodiment of this application;

[0036] Figure 7 This is a flowchart of an object matching method provided in an embodiment of this application;

[0037] Figure 8 This is a schematic diagram of the structure of a training device for an object matching model provided in an embodiment of this application;

[0038] Figure 9 This is a schematic diagram of the structure of an object matching device provided in an embodiment of this application;

[0039] Figure 10 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application;

[0040] Figure 11 This is a schematic diagram of the structure of a server provided in an embodiment of this application. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0042] Figure 1 This is a schematic diagram of a training method for an object matching model or an implementation environment for an object matching method provided in an embodiment of this application, such as... Figure 1 As shown, the implementation environment includes a terminal device 101 and a server 102. The training method or object matching method of the object matching model in this embodiment can be executed by the terminal device 101, by the server 102, or jointly by both the terminal device 101 and the server 102.

[0043] Terminal device 101 can be a smartphone, game console, desktop computer, tablet computer, laptop computer, smart TV, smart in-vehicle device, smart voice interaction device, smart home appliance, etc. Server 102 can be a single server, a server cluster consisting of multiple servers, or any of the following: cloud computing platform and virtualization center. This application embodiment does not limit this. Server 102 can communicate with terminal device 101 via a wired network or wireless network. Server 102 can have functions such as data processing, data storage, and data transmission and reception. This application embodiment does not limit this. The number of terminal devices 101 and servers 102 is not limited and can be one or more.

[0044] Object matching technology is often used in scenarios such as recommendation and search. For example, in a search scenario, it can identify the providers that match the search terms of the recipient, and then display the information of the providers as search results to the recipient.

[0045] Generally, a neural network model can be trained to obtain an object matching model, which can then be used for object matching. However, when training an object matching model, related techniques typically rely on basic information about the receiving and providing objects to determine a predicted matching result between them. Because the accuracy of this predicted matching result is low, the object matching model trained based on this result also lacks accuracy, resulting in low accuracy when using this model for object matching.

[0046] This application provides a training method for an object matching model, which can be applied in the above-described implementation environment and can improve the accuracy of the object matching model. Figure 2 The flowchart shown in this embodiment of the present application illustrates a training method for an object matching model. For ease of description, the terminal device 101 or server 102 executing the training method of the object matching model in this embodiment is referred to as an electronic device. This method can be executed by an electronic device. Figure 2 As shown, the method includes the following steps.

[0047] Step 201: Obtain the reference interaction network. The reference interaction network includes multiple nodes and multiple edges.

[0048] Each node can be either a receiving node or a providing node. In other words, the reference interaction network includes multiple receiving nodes and multiple providing nodes, with the multiple receiving objects including sample receiving objects.

[0049] The `receiving object` node represents the receiving object, which refers to the object that receives resources, including multimedia resources, food resources, clothing resources, etc. Optionally, the receiving object can passively receive resources recommended by the application, or it can actively retrieve resources matching search keywords by entering those keywords into the application. The `providing object` node represents the providing object, which refers to the object that provides resources. For example, a providing object could be an object that publishes multimedia resources, an object that displays clothing resources, etc.

[0050] Any edge represents the interaction between the receiving object and the providing object, with the two nodes at either end being the receiving object and the providing object, respectively. Interactions between the receiving object and the providing object include, but are not limited to, the receiving object clicking on the providing object, the receiving object selecting a resource provided by the providing object, etc.

[0051] In one possible implementation, step 201 includes steps 2011 to 2014.

[0052] Step 2011: Obtain basic information of multiple receiving objects, basic information of multiple providing objects, and the number of interactions between each receiving object and each providing object.

[0053] This application does not limit the method of obtaining the basic information of the receiving object. For example, the electronic device can retrieve the basic information of the receiving object from the network, or the electronic device can obtain the basic information input by the receiving object. The basic information of the receiving object includes basic information such as the receiving object's age, gender, and date of birth. Optionally, the basic information of the receiving object also includes the receiving object's identifier, which is information that can identify the receiving object. The receiving object's identifier includes at least one character, which can be numbers, letters, special symbols, etc.

[0054] This application does not limit the method of obtaining the basic information of the providing object. For example, an electronic device can retrieve the basic information of the providing object from a network, or the electronic device can receive basic information input by the providing object. The basic information of the providing object includes information about the resources provided by the providing object, including but not limited to the name, category, quantity, and price of the resources. The quantity of resources can be the remaining quantity or the quantity of resources already provided. Optionally, the basic information of the providing object also includes an identifier of the providing object. The identifier of the providing object is information that can identify the providing object, and the identifier of the providing object also includes at least one character, which can be numbers, letters, special symbols, etc.

[0055] This application does not limit the method for obtaining the number of interactions between the receiving object and the providing object. For example, an electronic device can capture the number of interactions between the receiving object and the providing object from the network; alternatively, a server can record interaction information when the receiving object and the providing object interact. This interaction information includes, but is not limited to, the identifier of the receiving object, the identifier of the providing object, the interaction time, the interaction method, the search terms entered by the receiving object, and environmental information during the interaction. The interaction method may be the receiving object clicking on the providing object, the receiving object selecting resources provided by the providing object, etc. The environmental information includes, but is not limited to, time information, weather information, and address information.

[0056] Step 2012: Determine the nodes of the multiple receiving objects included in the reference interaction network based on the basic information of the multiple receiving objects.

[0057] In this embodiment, the basic information of any receiving object can be directly determined as a node of a receiving object included in the reference interaction network. Alternatively, feature extraction can be performed on the basic information of any receiving object to obtain the basic features of the receiving object, and these basic features can be used as a node of a receiving object included in the reference interaction network.

[0058] This application does not limit the method of feature extraction. For example, principal component analysis can be performed on the basic information of the receiving object to extract key information, which can then be used as the basic features of the receiving object. Alternatively, feature extraction can be performed on the basic information of the receiving object based on word frequency statistics to obtain the basic features of the receiving object.

[0059] Optionally, the basic information of the receiving object includes the receiving object's identifier and other information. The other information includes information other than the receiving object's identifier, such as the receiving object's gender and age. The receiving object's identifier can be directly encoded to obtain its identifier features. The other information of the receiving object can be binned according to the information range corresponding to each bin, resulting in a reference bin for each receiving object. Encoding the reference bins yields the receiving object's bin features. For example, if the first bin corresponds to an age range of 0 to 20, the second bin to 20 to 35, and the third bin to 35 to 100, then based on these age ranges, the receiving object's age of 40 belongs to the third bin. In other words, the reference bin for the receiving object is the third bin. Encoding the third bin yields the receiving object's bin features. Concatenating the receiving object's identifier features and bin features yields the receiving object's basic features.

[0060] Step 2013: Determine the nodes of the multiple providing objects included in the reference interaction network based on the basic information of the multiple providing objects.

[0061] In this embodiment, the basic information of any providing object can be directly determined as a node of a providing object included in the reference interaction network. Alternatively, features can be extracted from the basic information of any providing object to obtain its basic features, and these basic features can be used as a node of a providing object included in the reference interaction network. The method of feature extraction from the basic information of the providing object is similar to the method of feature extraction from the basic information of the receiving object, and will not be described again here.

[0062] Optionally, the basic information provided by the object includes the object's identifier, the identifier of the resource provided by the object, and other information about the object. The other information about the object includes information other than the object's identifier and the identifier of the resource provided. For example, if the resources provided by the object include primary resources, secondary resources, and tertiary resources, then the identifiers of the resources provided by the object include the identifiers of the primary resources, secondary resources, and tertiary resources, and the other information about the object includes the names and quantities of the primary resources, secondary resources, and tertiary resources, etc.

[0063] The identifier of the providing object can be directly encoded to obtain the identifier feature of the providing object. Similarly, the identifier of the resource provided by the providing object can be directly encoded to obtain the identifier feature of the resource provided by the providing object. Other information of the providing object can be bucketed according to the information range corresponding to each bucket, resulting in a reference bucket for the providing object. Encoding the reference bucket for the providing object yields the bucket feature of the providing object. Finally, the identifier feature of the providing object, the identifier feature of the resource provided by the providing object, and the bucket feature of the providing object are concatenated to obtain the basic feature of the providing object.

[0064] Step 2014: For any receiving object and any providing object, determine the edge between the nodes of any receiving object and any providing object based on the number of interactions between them.

[0065] Optionally, the number of interactions between any receiving object and any providing object can be used as the edge between the nodes of the receiving object and the providing object. In this case, the edge between the nodes of the receiving object and the providing object can directly reflect the number of interactions between them.

[0066] Alternatively, based on the mapping relationship between frequency and weight, the number of interactions between any receiving object and any providing object can be mapped to a weight, and this weight can be used as an edge between the nodes of the receiving object and the providing object. The more times the receiving object and the providing object interact, the greater the weight. In this case, the edge between the nodes of the receiving object and the providing object represents the weight between them. Since this weight reflects the number of interactions between the receiving object and the providing object, the edge between the nodes of the receiving object and the providing object can indirectly reflect the number of interactions between them.

[0067] Alternatively, based on the frequency range corresponding to each bucket, the number of interactions between any receiving object and any providing object is divided into buckets to obtain target buckets. The weight corresponding to the target bucket is determined according to the mapping relationship between buckets and weights, and this weight is used as the edge between the node of the receiving object and the node of the providing object. Each bucket corresponds to the first and second counts, with the first count being less than the second count. The range between the first and second counts is the frequency range corresponding to that bucket. Optionally, the average value corresponding to that bucket is determined based on the first and second counts, and the weight of the bucket is proportional to the average value. In this case, the edge between the node of the receiving object and the node of the providing object represents the weight between them, and this weight reflects the number of interactions between them. Therefore, the edge between the node of the receiving object and the node of the providing object can indirectly reflect the number of interactions between them.

[0068] In this embodiment of the application, obtaining the nodes of each receiving object included in the reference interaction network, the nodes of each providing object included in the reference interaction network, and the edges between the nodes of the receiving object and the nodes of the providing object are equivalent to obtaining the reference interaction network.

[0069] Please see Figure 3 , Figure 3 This is a schematic diagram of a reference interaction network provided in an embodiment of this application. The reference interaction network includes receiving object nodes U1 to U6, providing object nodes P1 to P5, and edges between the receiving object nodes and the providing object nodes. The edges between the receiving object nodes and the providing object nodes represent the weight between the receiving object and the providing object. For example, the edge between receiving object node U1 and providing object node P5 represents the weight between the receiving object and the providing object, and this weight (i.e., the edge between U1 and P5) is denoted by W15.

[0070] In one possible implementation, step 2014 includes: if the number of interactions between any receiving object and any providing object is greater than a second threshold, then based on the number of interactions between any receiving object and any providing object, determine the edge between the node of any receiving object and the node of any providing object; if the number of interactions between any receiving object and any providing object is not greater than the second threshold, then determine that there is no edge between the node of any receiving object and the node of any providing object.

[0071] In this embodiment, if the number of interactions between any receiving object and any providing object is greater than a second threshold, then according to step 2014, the edge between the node of the receiving object and the node of the providing object can be determined based on the number of interactions between the receiving object and the providing object. The method for determining the edge will not be elaborated further here. If the number of interactions between any receiving object and any providing object is not greater than the second threshold, then it is determined that there is no edge between the node of the receiving object and the node of the providing object. This embodiment does not limit the value of the second threshold. For example, the second threshold is a value set based on human experience, or the number of interactions between each receiving object and each providing object can be sorted, and the sorted number of interactions can be used as the second threshold.

[0072] In this embodiment of the application, the reference interaction network includes nodes of each receiving object and nodes of each providing object. There may or may not be an edge between any node of a receiving object and any node of a providing object.

[0073] Please see Figure 4 , Figure 4 This is a schematic diagram of another reference interactive network provided in the embodiments of this application. Figure 4 The reference interaction network is equivalent to the Figure 3 The reference interaction network is obtained by edge filtering, that is, by... Figure 3 Filter some edges in the reference interaction network, such as filtering out Figure 3 The reference interactive networks W24, W32, and W51 are obtained. Figure 4 The reference interactive network.

[0074] Step 202: Based on the reference interaction network, determine the relevant provider from multiple providers. The probability that the sample receiving object interacts with the relevant provider is greater than a first threshold.

[0075] The reference interaction network includes nodes for each receiving object, nodes for each providing object, and edges between the nodes of the receiving objects and the nodes of the providing objects, wherein each receiving object includes a sample receiving object. Since the edges between the nodes of the receiving objects and the nodes of the providing objects represent the number of interactions between the receiving objects and the providing objects, and the greater the number of interactions between the receiving objects and the providing objects, the higher the probability of interaction between the receiving objects and the providing objects, the electronic device can, based on the reference interaction network, determine from multiple providing objects that the probability of interaction with the sample receiving object is greater than a first threshold, and designate the determined providing objects as relevant providing objects. This application embodiment does not limit the value of the first threshold; exemplarily, the first threshold is a value set based on human experience.

[0076] In one possible implementation, step 202 includes steps 2021 to 2023.

[0077] Step 2021: Feature extraction is performed on the reference interaction network to obtain the first features of each providing object.

[0078] In this embodiment of the application, the nodes of the providing object included in the reference interaction network can represent the basic information of the providing object. Therefore, by performing feature extraction on any node of the providing object, the first feature of the providing object can be obtained.

[0079] Optionally, any node providing an object can be the basic information of the providing object, or it can be the basic feature of the providing object obtained after feature extraction from the basic information of the providing object. The basic feature of the providing object can be used as the first feature of the providing object, or further feature extraction can be performed on the basic feature of the providing object to obtain the first feature of the providing object. This application does not limit the method of feature extraction of the basic feature of the providing object. For example, the basic feature of the providing object can be mapped to the first feature of the providing object through feature mapping. The feature dimension of the first feature of the providing object can be lower than the feature dimension of the basic feature of the providing object. Alternatively, based on some prior knowledge, the first feature of the providing object can be selected from the basic feature of the providing object.

[0080] Since the reference interaction network includes nodes of each receiving object, nodes of each providing object, and edges between nodes of receiving objects and nodes of providing objects, for any providing object, the first feature of the providing object can be determined based on the nodes of the providing object and the nodes of the receiving objects that have edges with the nodes of the providing object.

[0081] Optionally, any node providing the object and any node receiving the object that has an edge with the node providing the object are considered as a composite node. This composite node represents the basic information of the providing object and the basic information of the receiving object interacting with the providing object. Feature extraction of this composite node yields the first feature of the providing object. The method for feature extraction of the composite node is similar to that for feature extraction of the nodes providing the object, and will not be elaborated further here.

[0082] Optionally, the first feature of any providing object is obtained by feature extraction from the reference interaction network using a feature extraction model.

[0083] In this embodiment, a second network model can be trained to obtain a feature extraction model. A reference interaction network is input into the feature extraction model, which then extracts features from the reference interaction network to obtain the first features of each provided object. This embodiment does not limit the model structure of the feature extraction model. For example, the model structure of the feature extraction model is the same as that of the second network model, differing only in the model parameters. The feature extraction model can be any network such as a graph neural network or a multi-layer perceptron (MLP).

[0084] Optionally, after inputting the reference interaction network into the feature extraction model, the feature extraction model can determine the first feature of any providing object based on the nodes of the providing object and the nodes of the receiving object that have edges with the nodes of the providing object.

[0085] In the case of using a feature extraction model to extract features from a reference interaction network to obtain the first features of each providing object, the training method of the object matching model in this application embodiment further includes steps 205 to 208.

[0086] Step 205: Extract features from the reference interaction network using the second network model to obtain the second features of each receiving object and the second features of each providing object.

[0087] In this embodiment, after inputting the reference interaction network into the second network model, the second network model can extract features from any node of the providing object to obtain the second feature of the providing object. Alternatively, it can determine the second feature of the providing object based on any node of the providing object and the nodes of the receiving objects that have edges with the node of the providing object. The method for determining the second feature of the providing object is similar to the method for determining the first feature of the providing object, and will not be described again here.

[0088] Similarly, the second network model can extract features from any node of the receiving object to obtain the second feature of that receiving object. Alternatively, it can determine the second feature of the receiving object based on any node of the receiving object and the nodes of the providing object that have edges with that node. The method for determining the second feature of the receiving object is similar to the method for determining the first feature of the providing object, and will not be elaborated further here.

[0089] Step 206: For any receiving object, determine the loss of any receiving object based on the second feature of any receiving object, the second features of each receiving object and the second features of each providing object.

[0090] The loss of a receiving object can be determined based on the loss function of the receiving object, the second features of any receiving object, the second features of all other receiving objects, and the second features of all providing objects. This application does not limit the loss function of the receiving object. For example, the feature similarity between the second feature of any receiving object and the second features of all other receiving objects can be calculated. Furthermore, the feature similarity between the second feature of any receiving object and the second features of all providing objects can also be calculated. The loss of any receiving object is determined based on the feature similarity between the second feature of any receiving object and the second features of all other receiving objects, and the feature similarity between the second feature of any receiving object and the second features of all providing objects. Here, "other receiving objects" refers to receiving objects other than any given receiving object.

[0091] This application does not limit the method of determining feature similarity. For example, the inner product or cosine value of the included angle between the second feature of any receiving object and the second feature of any other receiving object is calculated, and this information is used as the feature similarity between the second feature of the receiving object and the second feature of the other receiving object.

[0092] Optionally, step 206 includes steps 2061 to 2063.

[0093] Step 2061: Based on the reference interaction network, sample the nodes that provide positive samples of any receiving object from the nodes that have edges with any receiving object.

[0094] First, identify the nodes of each providing object that have edges with any node of the receiving object from the nodes included in the reference interaction network. Then, randomly sample a set number of nodes from the nodes of each providing object that have edges with any node of the receiving object, and use these sampled nodes as the positive sample nodes of the receiving object. Since there are edges between the nodes of the receiving object and the nodes of the providing object when the receiving object interacts with the providing object, and there are edges between the nodes of the positive sample nodes of the receiving object and the nodes of the receiving object, the positive sample nodes of the receiving object are the providing objects that interact with the receiving object.

[0095] Step 2062: Based on the reference interaction network, sample the nodes of negative samples of any receiving object from the nodes of other objects that have no edge between them and the nodes of any receiving object. The other objects are either receiving objects or providing objects.

[0096] We can first identify, from the nodes included in the reference interaction network, all nodes that have no edge connection with any receiving object node. Any node that has no edge connection with a receiving object node can be either the receiving object node or the providing object node. For ease of description, we will denote any node that has no edge connection with a receiving object node as a node of another object that has no edge connection with a receiving object node.

[0097] Next, from the nodes of other objects that have no edge between them and any receiving object, a set number of nodes are randomly sampled, and these sampled nodes are used as the negative sample nodes of the receiving object.

[0098] On the one hand, since there is no interaction between the receiving object and the providing object, there are no edges between the nodes of the receiving object and the nodes of the providing object. Furthermore, there are no edges between the nodes of the negative sample of the receiving object and the nodes of the receiving object. Therefore, the negative sample of a receiving object may be a providing object that has not interacted with the receiving object. On the other hand, since there is no interaction between any two receiving objects (i.e., no edges between the nodes of any two receiving objects), and there are no edges between the nodes of the negative sample of a receiving object and the nodes of the receiving object, the negative sample of one receiving object may be another receiving object, and these two receiving objects do not interact.

[0099] Step 2063: Determine the loss of any receiving object based on the second feature of any receiving object, the second feature of any positive sample of any receiving object, and the second feature of any negative sample of any receiving object.

[0100] In this embodiment, each receiving object corresponds to at least one positive sample. Information such as the product and cosine of the angle between the second feature of the receiving object and the second feature of any positive sample of the receiving object can be calculated, and this information is determined as the similarity between the receiving object and this positive sample of the receiving object. Based on the similarity between the receiving object and each of the positive samples of the receiving object, the positive sample loss of the receiving object is determined.

[0101] Optionally, the positive sample loss of any receiving object can be calculated according to formula (1) as described below.

[0102]

[0103] Where L1 is the positive sample loss of the i-th receiving object, E i It is the second characteristic of the i-th receiving object, E j E is the second feature of the j-th positive sample of the i-th receiving object. i *E jIt represents the similarity between the i-th receiving object and the j-th positive sample of the i-th receiving object. exp is the function symbol for the exponential function, and ln is the function symbol for the logarithmic function.

[0104] For each receiving object, there is at least one negative sample. Information such as the product and cosine of the angle between the second feature of the receiving object and the second feature of any negative sample of that receiving object can be calculated. This information is then used to determine the similarity between the receiving object and its negative sample. Based on the similarity between the receiving object and each of its negative samples, the negative sample loss for that receiving object is determined.

[0105] Optionally, the negative sample loss of any receiving object can be calculated according to formula (2) as described below.

[0106]

[0107] Where L2 is the negative sample loss for the i-th receiving object, E i It is the second characteristic of the i-th receiving object, E l E is the second feature of the l-th negative sample of the i-th receiving object. i *E l It represents the similarity between the i-th receiving object and the l-th negative sample of the i-th receiving object. exp is the function symbol for the exponential function, and ln is the function symbol for the logarithmic function.

[0108] Next, the sum of the positive sample loss and the negative sample loss of any receiving object is taken as the loss of any receiving object. That is, the loss of the i-th receiving object is L = L1 + L2.

[0109] Step 207: For any provider, determine the loss of any provider based on the second feature of the provider, the second feature of each receiver, and the second feature of each provider.

[0110] The loss of a providing object can be determined based on the second features of any providing object, the second features of each receiving object, and the second features of each providing object, according to the loss function of the providing object. This application does not limit the loss function of the providing object. For example, the feature similarity between the second features of any providing object and the second features of each other providing object can be calculated. Furthermore, the feature similarity between the second features of any providing object and the second features of each receiving object can also be calculated. The loss of any providing object is determined based on the feature similarity between the second features of any providing object and the second features of each other providing object, and the feature similarity between the second features of any providing object and the second features of each receiving object. Here, "other providing objects" refers to providing objects other than any one providing object.

[0111] This application does not limit the method of determining feature similarity. For example, the inner product or cosine value of the included angle between the second feature of any providing object and the second feature of any other providing object is calculated, and this information is used as the feature similarity between the second feature of the providing object and the second feature of the other providing object.

[0112] Optionally, step 207 includes: sampling nodes of positive samples of any provider object from nodes of receiving objects that have edges with nodes of any provider object, based on the reference interaction network; sampling nodes of negative samples of any provider object from nodes of other objects that do not have edges with nodes of any provider object, wherein the other objects are receiving objects or provider objects; and determining the loss of any provider object based on the second feature of any provider object, the second feature of the positive samples of any provider object, and the second feature of the negative samples of any provider object.

[0113] The method for determining the loss of the providing object is similar to the method for determining the loss of the receiving object, as can be seen in the descriptions of steps 2061 to 2063, and will not be repeated here.

[0114] Step 208: Based on the loss of each receiving object and the loss of each providing object, train the second network model to obtain the feature extraction model.

[0115] The loss of the second network model can be determined based on the losses of each receiving object and each providing object. For example, the sum or average of the losses of each receiving object and each providing object can be used as the loss of the second network model.

[0116] The second network model can be trained using the loss of the second network model to obtain the trained second network model. If the trained second network model meets the first training termination condition, it is used as the feature extraction model. If the trained second network model does not meet the first training termination condition, it is used as the second network model for the next training iteration, and the training can be repeated according to steps 205 to 208 until the feature extraction model is obtained. After obtaining the feature extraction model, the reference interactive network can be input into the feature extraction model, and the feature extraction model will extract features from the reference interactive network to obtain the first features of each provided object.

[0117] Step 2022: Determine the first similarity between the features of the sample receiving object and the first features of each providing object. The first similarity between the features of the sample receiving object and the first features of any providing object is used to characterize the likelihood of the sample receiving object interacting with any providing object.

[0118] Feature extraction can be performed on the basic information of the sample receiving object to obtain its features. The method for feature extraction on the basic information of the sample receiving object is similar to the method for feature extraction on the basic information of the receiving object, and will not be elaborated further here.

[0119] It can calculate information such as the product and cosine of the angle between the features of the sample receiving object and the first feature of any providing object, and determine this information as the first similarity between the features of the sample receiving object and the first feature of any providing object.

[0120] In this embodiment, the first feature of the providing object can characterize the basic information of the providing object, and also characterize the basic information of each receiving object interacting with the providing object. Therefore, the first feature of the providing object can characterize the information shared by the receiving objects interacting with the providing object. When calculating the first similarity using the features of the sample receiving object and the first feature of the providing object, the larger the first similarity, the closer the sample receiving object is to the receiving object interacting with the providing object; that is, the more likely the sample receiving object is to interact with the providing object. Conversely, the smaller the first similarity, the further the sample receiving object is from the receiving object interacting with the providing object; that is, the less likely the sample receiving object is to interact with the providing object. Therefore, the first similarity can characterize the probability of the sample receiving object interacting with the providing object.

[0121] Step 2023: If the first similarity between the features of the sample receiving object and the first feature of any providing object is greater than the first threshold, then any providing object is regarded as the relevant providing object.

[0122] Since the first similarity between the features of the sample receiving object and the first feature of the providing object indicates the possibility of interaction between the sample receiving object and the providing object, the first similarity can be binarized based on a first threshold to predict whether the sample receiving object and the providing object will interact.

[0123] Optionally, when the first similarity is greater than the first threshold, it indicates that the sample receiving object is likely to interact with the providing object, and it can be predicted that the sample receiving object will interact with the providing object. In this case, the providing object is regarded as the relevant providing object of the sample receiving object. When the first similarity is not greater than the first threshold, it indicates that the sample receiving object is unlikely to interact with the providing object, and it can be predicted that the sample receiving object will not interact with the providing object. In this case, the providing object cannot be regarded as the relevant providing object of the sample receiving object.

[0124] Step 203: Using the first network model, based on the basic information of the sample receiving object, the basic information of the sample providing object, and the basic information of the relevant providing object, determine the prediction matching result between the sample receiving object and the sample providing object.

[0125] The basic information of the sample receiving object, the sample providing object, and related providing objects can be input into the first network model. The first network model determines and outputs the predicted matching result between the sample receiving object and the sample providing object. This predicted matching result represents the probability that the sample receiving object and the sample providing object match. That is, the predicted matching result characterizes the likelihood that the sample receiving object and the sample providing object match. The larger the predicted matching result, the higher the likelihood that the sample receiving object and the sample providing object match. This application does not limit the model structure of the first network model. For example, the first network model includes an encoder, a splicing network, a feature processing network, and an activation network, etc. The functions of the encoder, the splicing network, the feature processing network, and the activation network are described below and will not be repeated here.

[0126] Optionally, step 203 includes steps 2031 to 2033.

[0127] Step 2031: Based on the basic information of the sample receiving object and the basic information of the relevant providing object, determine the first weight of the relevant providing object.

[0128] The first network model includes an encoder that can extract features from the basic information of the sample receiving object to obtain the features of the sample receiving object. The encoder may include a feature extraction model, which extracts features from the basic information of the relevant providing object to obtain the first feature of the relevant providing object. Alternatively, the encoder may directly obtain the first feature of the relevant providing object from the feature extraction model.

[0129] Information such as the product and cosine of the angle between the features of the sample receiving object and the first features of the related providing object can be calculated. This information is then used to determine the second similarity between the features of the sample receiving object and the first features of the related providing object. The second similarity can serve as the first weight of the related providing object, or it can be used to determine the first weight of the related providing object. The second similarity characterizes the probability of interaction between the sample receiving object and the related providing object; therefore, the greater the second similarity, the greater the first weight of the related providing object.

[0130] Step 2032: Determine the first weighted feature based on the first weight and the basic information of the relevant provided object.

[0131] The number of relevant providers is at least one. For any given relevant provider, the encoder can determine the first feature of the relevant provider based on its basic information, and multiply the first weight of the relevant provider with the first feature of the relevant provider to obtain the first weighted feature of the relevant provider.

[0132] For example, the first weights of the 1st to kth related providing objects are W1, W2, ..., W... k The first features of the first to k related objects are P1, P2, ..., Pk, respectively. k Then the first weighted features of the 1st to kth related providing objects are W1*P1, W2*P2, ..., W k *P k .

[0133] Step 2033: Based on the first weighted feature, the basic information of the sample receiving object, and the basic information of the sample providing object, determine the predicted matching result.

[0134] The encoder can determine the first weighted feature of any relevant providing object. Furthermore, the encoder can determine the features of the sample receiving object and the sample providing object based on the basic information of the sample receiving object. The concatenation network included in the first network model concatenates the first weighted features of each relevant providing object, the features of the sample receiving object, and the features of the sample providing object to obtain a first concatenated feature. The feature processing network included in the first network model performs fusion processing on the first concatenated feature to obtain a first fusion result. The activation network included in the first network model performs activation processing on the first fusion result to obtain a predicted matching result.

[0135] Optionally, step 203 includes steps 2034 to 2036.

[0136] Step 2034: Based on the basic information of the sample provider and the basic information of the related provider, determine the second weight of the related provider.

[0137] The encoder in the first network model can extract features from the basic information of the sample provided object to obtain the features of the sample provided object. Alternatively, the encoder in the first network model can directly extract features from the basic information of the relevant provided object to obtain the first features of the relevant provided object, or it can directly obtain the first features of the relevant provided object from the feature extraction model.

[0138] Information such as the product and cosine of the angle between the features of the sample object and the first features of the related objects can be calculated. This information is then used to determine the third similarity between the features of the sample object and the first features of the related objects. This third similarity can serve as the second weight of the related objects, or it can be used to determine the second weight of the related objects. Since the third similarity characterizes the similarity between the sample object and the related objects, a larger third similarity results in a larger second weight for the related objects.

[0139] For example, the second weight of any relevant providing object is determined according to the formula (3) shown below.

[0140]

[0141] Among them, S i The second weight characterizes the i-th relevant providing object. Target poi The characterization sample provides features of the object. i The first feature representing the i-th related object. P j The first characteristic of the j-th related providing object is represented. k represents the number of related providing objects. exp represents the sign of the exponential function, and ∑ is the sign of the summation function.

[0142] Step 2035: Determine the second weighted feature based on the second weight and the basic information of the relevant provided object.

[0143] The number of relevant providers is at least one. For any given relevant provider, the encoder can determine a first feature of the relevant provider based on its basic information, and multiply the second weight of the relevant provider with the first feature to obtain a second weighted feature of the relevant provider.

[0144] For example, the second weights of the first to k related providing objects are S1, S2, ..., S... k The first features of the first to k related objects are P1, P2, ..., Pk, respectively.k Then the second weighted features of the first to k related providing objects are S1*P1, S2*P2, ..., S k *P k .

[0145] Step 2036: Determine the predicted matching result based on the second weighted feature, the basic information of the sample receiving object, and the basic information of the sample providing object.

[0146] The first network model includes a concatenation network that concatenates the second weighted features of each relevant providing object, the features of the receiving object, and the features of the providing object to obtain a second concatenated feature. The feature processing network in the first network model then fuses this second concatenated feature to obtain a second fusion result. Finally, the activation network in the first network model activates this second fusion result to obtain a predicted matching result.

[0147] Optionally, for any relevant provider, the first weight, the second weight, and the first feature of the relevant provider can be multiplied to obtain the third weighted feature of the relevant provider. Then, the concatenation network included in the first network model can concatenate the third weighted features of each relevant provider, the features of the sample receiving object, and the features of the sample providing object to obtain the third concatenated feature.

[0148] For example, please see Figure 5 , Figure 5 This is a schematic diagram illustrating the determination of a third splicing feature according to an embodiment of this application. The first features of the 1st to kth related providing objects are P1, P2...P... k The first weights of the 1st to kth related objects are W1, W2...W k The second weights of the first to k related objects are S1, S2...S... k Then the third weighted features of the first to k related providing objects are W1*S1*P1, W2*S2*P2, ..., W k *S k *P k The third concatenated feature is obtained by concatenating the third weighted features of each relevant providing object. The third concatenated feature can be characterized as follows: Where k is the number of related provided objects, and j is the sequence number.

[0149] The first network model includes a feature processing network that fuses the third concatenated features to obtain a third fusion result. The first network model also includes an activation network that activates the third fusion result to obtain a predicted matching result.

[0150] Optionally, step 203 includes steps 2037 to 2038.

[0151] Step 2037: Obtain target information. Target information includes at least one of the following: basic information of the historical provider, search terms input by the sample receiver, and environmental information. The historical provider is the provider that has interacted with the sample receiver in the past.

[0152] This application does not limit the method of obtaining basic information about historical providers. For example, an electronic device can retrieve basic information about historical providers that have historically interacted with the sample receiving object from the network; alternatively, a server can record interaction information between the receiving object and the provider. The electronic device can then determine the historical providers that have historically interacted with the sample receiving object based on this interaction information, thereby obtaining the basic information of the historical providers. The basic information of the historical providers includes information about the resources provided by the historical providers.

[0153] This application does not limit the method of obtaining search terms and environmental information. For example, the electronic device can retrieve the search terms and environmental information input by the sample receiving object from the network, or the electronic device can determine the search terms and environmental information input by the sample receiving object from the interaction information recorded by the server.

[0154] Step 2038: Using the first network model, based on the target information, the basic information of the sample receiving object, the basic information of the sample providing object, and the basic information of the relevant providing object, determine the predicted matching result.

[0155] The first network model includes an encoder that can extract features from target information to obtain the features of the target information. The method for determining the features of the target information is similar to the method for determining the basic features of the receiving object, and will not be elaborated here.

[0156] Next, the splicing network included in the first network model can splice together the features of the target information, the features of the sample receiving object, the features of the sample providing object, and the first weighted features (or the second weighted features or the third weighted features) of each relevant providing object to obtain the fourth spliced ​​feature.

[0157] The first network model includes a feature processing network that fuses the fourth concatenated features to obtain the fourth fusion result. The first network model also includes an activation network that activates the fourth fusion result to obtain the predicted matching result.

[0158] Please see Figure 6 , Figure 6This is a schematic diagram illustrating the determination of a predicted matching result according to an embodiment of this application. In this embodiment, the basic information of the historical providing object, the basic information of the sample providing object, the basic information of the related providing object, the basic information of the sample receiving object, the search term input by the sample receiving object, and the environmental information are all input into a first network model. The encoder of the first network model can determine the features of the historical providing object based on the basic information of the historical providing object, determine the features of the sample providing object based on the basic information of the sample providing object, determine the first feature of the related providing object based on the basic information of the related providing object, determine the features of the sample receiving object based on the basic information of the sample receiving object, determine the features of the search term based on the search term input by the sample receiving object, and determine the environmental features based on the environmental information. The features of the target information mentioned above include at least one of the features of the historical providing object, the features of the search term, and the environmental features.

[0159] The number of historical objects provided is at least one; therefore, the characteristics of historical objects provided are at least one, and can be V1, V2...V m The encoder is used to characterize the features of the first to m historical provided objects. It also includes a first aggregation network, which receives the features of each historical provided object and the features of the sample provided objects together, and determines the first aggregated features.

[0160] The number of relevant providing objects is at least one; therefore, the first characteristic of the relevant providing objects is at least one, which can be P1, P2...P k The encoder also includes a second aggregation network. The first features of each relevant providing object and the features of the sample providing objects are input into the second aggregation network, which determines the second aggregated features. These second aggregated features are equivalent to the first weighted features, second weighted features, or third weighted features mentioned above.

[0161] The concatenation network concatenates the first aggregated features, the features of the sample provider, the second aggregated features, the features of the sample receiver, the features of the search term, and the environmental features to obtain the target concatenated features. The feature processing network then fuses these target concatenated features to obtain the target fusion result. Finally, the activation network activates the target fusion result to obtain the predicted matching result.

[0162] Step 204: Based on the predicted matching results and the labeled matching results, train the first network model to obtain the object matching model. The labeled matching results are the matching results between the sample receiving object and the sample providing object obtained through labeling. The object matching model is used to determine the matching result between the target receiving object and the target providing object.

[0163] The loss of the first network model can be determined based on the predicted matching results and the labeled matching results. For example, the loss of the first network model can be determined based on the predicted matching results and the labeled matching results according to the formula of the cross-entropy loss function. Optionally, the formula of the cross-entropy loss function is shown in formula (4) below.

[0164]

[0165] Here, Loss represents the loss of the first network model. i The labeling matching result is represented by N, where N is a positive integer. Optionally, if N is 2, then y1 represents the interaction between the sample receiving object and the sample providing object, and y2 represents the interaction between the sample receiving object and the sample providing object. P(x) represents the predicted matching result. log represents the sign of the logarithmic function, and ∑ represents the sign of the summation function.

[0166] The first network model can be trained using the loss of the first network model to obtain the trained first network model. If the trained first network model meets the second training termination condition, then the trained first network model is used as the object matching model. If the trained first network model does not meet the second training termination condition, then the trained first network model is used as the first network model for the next training iteration, and the first network model can be trained again according to steps 201 to 204, until the object matching model is obtained.

[0167] It should be noted that all information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the basic information of the sample recipients, the basic information of the sample providers, and the basic information of the relevant providers involved in this application were all obtained with full authorization.

[0168] The reference interaction network in the above method includes nodes for receiving objects and nodes for providing objects, and the edges between the nodes of receiving objects and providing objects represent the number of interactions between them. Therefore, the reference interaction network records the interaction information between the receiving object and the providing object. Based on the reference interaction network, relevant providing objects with a high probability of interacting with the sample receiving object are identified from multiple providing objects. This ensures that the basic information of the relevant providing objects includes information common to all providing objects interacting with the sample receiving object. The predicted matching results between the sample receiving object and the sample providing object determined based on this common information, the basic information of the sample receiving object, and the basic information of the sample providing object have high accuracy. Therefore, the object matching model trained based on the predicted matching results and the labeled matching results has high accuracy, thereby improving the accuracy of the object matching results.

[0169] This application also provides an object matching method, which can be applied in the above-described implementation environment and can improve the accuracy of object matching results. Figure 7 The flowchart of an object matching method provided in this application embodiment is shown as an example. For ease of description, the terminal device 101 or server 102 executing the object matching method in this application embodiment is referred to as an electronic device. This method can be executed by an electronic device. Figure 7 As shown, the method includes the following steps.

[0170] Step 701: Obtain the reference interactive network.

[0171] The reference interaction network includes multiple nodes, each of which can be either a receiving object node or a providing object node. The edges between any receiving object node and any providing object node represent the number of interactions between any receiving object and any providing object. The multiple receiving objects include the target receiving object.

[0172] The content of the reference interaction network has been described in step 201 and will not be repeated here.

[0173] Step 702: Based on the reference interaction network, determine the reference provider from multiple providers.

[0174] The probability that the target receiving object interacts with the reference providing object is greater than the first threshold. The method for determining the relevant providing object is described in step 202, and the method for determining the reference providing object is similar to that for determining the relevant providing object. Therefore, the description of step 701 can be found in the description of step 202, and will not be repeated here.

[0175] Step 703: Using an object matching model, based on the basic information of the target receiving object, the basic information of the target providing object, and the basic information of the reference providing object, determine the matching result between the target receiving object and the target providing object.

[0176] Among them, the object matching model is based on and Figure 2 The relevant object matching model was trained using the training method. The method for determining the matching result between the target receiving object and the target providing object is similar to the method for determining the predicted matching result between the sample receiving object and the sample providing object, as described in step 203, and will not be repeated here.

[0177] It should be noted that all information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the basic information of the target receiving object, the basic information of the target providing object, and the basic information of the reference providing object involved in this application were all obtained with full authorization.

[0178] The reference interaction network described above includes nodes for receiving objects and nodes for providing objects, with edges between these nodes representing the number of interactions between them. Therefore, the reference interaction network records the interaction information between the receiving object and the providing object. Based on this network, a reference providing object with a high probability of interacting with the target receiving object is determined from multiple providing objects. This ensures that the basic information of the reference providing object contains information common to all providing objects interacting with the target receiving object. The matching results between the target receiving object and the target providing object determined based on this common information, the basic information of the target receiving object, and the basic information of the target providing object are highly accurate.

[0179] Figure 8 The diagram shown is a structural schematic of a training device for an object matching model provided in an embodiment of this application. Figure 8 As shown, the device includes:

[0180] The acquisition module 801 is used to acquire a reference interaction network. The reference interaction network includes multiple nodes, any node is a receiving object node or a providing object node, and the edge between any receiving object node and any providing object node represents the number of interactions between any receiving object and any providing object. The multiple receiving objects include sample receiving objects.

[0181] The determination module 802 is used to determine a relevant providing object from multiple providing objects based on a reference interaction network, wherein the probability of the sample receiving object interacting with the relevant providing object is greater than a first threshold.

[0182] The determination module 802 is also used to determine the prediction matching result between the sample receiving object and the sample providing object based on the basic information of the sample receiving object, the basic information of the sample providing object and the basic information of the relevant providing object through the first network model;

[0183] Training module 803 is used to train the first network model based on the predicted matching results and the labeled matching results to obtain the object matching model. The labeled matching results are the matching results between the sample receiving object and the sample providing object obtained through labeling. The object matching model is used to determine the matching results between the target receiving object and the target providing object.

[0184] In one possible implementation, the acquisition module 801 is used to acquire basic information of multiple receiving objects, basic information of multiple providing objects, and the number of interactions between each receiving object and each providing object; determine the nodes of the multiple receiving objects included in the reference interaction network based on the basic information of the multiple receiving objects; determine the nodes of the multiple providing objects included in the reference interaction network based on the basic information of the multiple providing objects; and for any receiving object and any providing object, determine the edge between the node of any receiving object and the node of any providing object based on the number of interactions between the two objects.

[0185] In one possible implementation, the acquisition module 801 is used to determine the edge between the node of any receiving object and the node of any providing object based on the number of interactions between any receiving object and any providing object if the number of interactions between any receiving object and any providing object is greater than a second threshold; if the number of interactions between any receiving object and any providing object is not greater than the second threshold, it is determined that there is no edge between the node of any receiving object and the node of any providing object.

[0186] In one possible implementation, the determining module 802 is used to extract features from the reference interaction network to obtain first features of each providing object; determine the first similarity between the features of the sample receiving object and the first features of each providing object, wherein the first similarity between the features of the sample receiving object and the first features of any providing object is used to characterize the possibility of the sample receiving object interacting with any providing object; if the first similarity between the features of the sample receiving object and the first features of any providing object is greater than a first threshold, then any providing object is regarded as a relevant providing object.

[0187] In one possible implementation, the first feature of any providing object is obtained by feature extraction from the reference interaction network using a feature extraction model; the device further includes:

[0188] The extraction module is used to extract features from the reference interaction network through the second network model to obtain the second features of each receiving object and the second features of each providing object.

[0189] The determining module 802 is further configured to, for any receiving object, determine the loss of any receiving object based on the second feature of any receiving object, the second features of each receiving object and the second features of each providing object;

[0190] The determining module 802 is further configured to, for any given providing object, determine the loss of any given providing object based on the second characteristics of any given providing object, the second characteristics of each receiving object and the second characteristics of each providing object;

[0191] The training module 803 is also used to train the second network model based on the loss of each receiving object and the loss of each providing object, so as to obtain the feature extraction model.

[0192] In one possible implementation, the determining module 802 is further configured to, based on the reference interaction network, sample nodes of positive samples of any receiving object from nodes of providing objects that have edges with nodes of any receiving object; sample nodes of negative samples of any receiving object from nodes of other objects that do not have edges with nodes of any receiving object, wherein the other objects are either receiving objects or providing objects; and determine the loss of any receiving object based on the second feature of any receiving object, the second feature of the positive samples of any receiving object, and the second feature of the negative samples of any receiving object.

[0193] In one possible implementation, the determining module 802 is further configured to determine a first weight of the relevant providing object based on the basic information of the sample receiving object and the basic information of the relevant providing object; determine a first weighted feature based on the first weight and the basic information of the relevant providing object; and determine a predicted matching result based on the first weighted feature, the basic information of the sample receiving object and the basic information of the sample providing object.

[0194] In one possible implementation, the determining module 802 is further configured to determine a second weight of the relevant providing object based on the basic information of the sample providing object and the basic information of the relevant providing object; determine a second weighted feature based on the second weight and the basic information of the relevant providing object; and determine a predicted matching result based on the second weighted feature, the basic information of the sample receiving object and the basic information of the sample providing object.

[0195] In one possible implementation, the determining module 802 is further configured to acquire target information, which includes at least one of the following: basic information of historical providers, search terms input by sample recipients, and environmental information. Historical providers are providers that have historically interacted with sample recipients. Based on the target information, basic information of sample recipients, basic information of sample providers, and basic information of related providers, the predictive matching result is determined through the first network model.

[0196] The reference interaction network in the aforementioned device includes nodes for receiving objects and nodes for providing objects, with edges between these nodes representing the number of interactions between them. Therefore, the reference interaction network records the interaction information between the receiving and providing objects. Based on this network, relevant providing objects with a high probability of interacting with the sample receiving object are identified from among multiple providing objects. This ensures that the basic information of these relevant providing objects includes information common to all providing objects interacting with the sample receiving object. The predicted matching results between the sample receiving object and the sample providing object, determined based on this common information, the basic information of the sample receiving object, and the basic information of the sample providing object, have high accuracy. Therefore, the object matching model trained based on the predicted matching results and the labeled matching results has high accuracy, thereby improving the accuracy of the object matching results.

[0197] It should be understood that the above Figure 8 The provided device, in implementing its functions, is only illustrated by the division of the above-described functional modules. In practical applications, the functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the device and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation processes are detailed in the method embodiments, and will not be repeated here.

[0198] Figure 9 The diagram shown is a structural schematic of an object matching device provided in an embodiment of this application. Figure 9 As shown, the device includes:

[0199] The acquisition module 901 is used to acquire a reference interaction network. The reference interaction network includes multiple nodes, any node is a receiving object node or a providing object node, and the edge between any receiving object node and any providing object node represents the number of interactions between any receiving object and any providing object. The multiple receiving objects include a target receiving object.

[0200] The determination module 902 is used to determine a reference provider from multiple providers based on a reference interaction network, wherein the probability of the target receiving object interacting with the reference provider is greater than a first threshold.

[0201] The determining module 902 is also used to determine the matching result between the target receiving object and the target providing object based on the basic information of the target receiving object, the basic information of the target providing object, and the basic information of the reference providing object, using an object matching model. The object matching model is based on... Figure 2 The relevant object matching model was trained using training methods.

[0202] The reference interaction network of the aforementioned device includes nodes for receiving objects and nodes for providing objects, and the edges between the nodes for receiving objects and nodes for providing objects represent the number of interactions between the receiving object and the providing object. Therefore, the reference interaction network records the interaction information between the receiving object and the providing object. Based on the reference interaction network, a reference providing object with a high probability of interacting with the target receiving object is determined from multiple providing objects. This ensures that the basic information of the reference providing object contains information common to all providing objects interacting with the target receiving object. The accuracy of the matching result between the target receiving object and the target providing object determined based on this common information, the basic information of the target receiving object, and the basic information of the target providing object is high.

[0203] It should be understood that the above Figure 9 The provided device, in implementing its functions, is only illustrated by the division of the above-described functional modules. In practical applications, the functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the device and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation processes are detailed in the method embodiments, and will not be repeated here.

[0204] Figure 10 This diagram illustrates a structural block diagram of a terminal device 1000 provided in an exemplary embodiment of this application. The terminal device 1000 includes a processor 1001 and a memory 1002.

[0205] Processor 1001 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1001 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 1001 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1001 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the screen. In some embodiments, processor 1001 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0206] The memory 1002 may include one or more computer-readable storage media, which may be non-transitory. The memory 1002 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1002 are used to store at least one computer program, which is executed by the processor 1001 to implement the training method or object matching method of the object matching model provided in the method embodiments of this application.

[0207] In some embodiments, the terminal device 1000 may also optionally include: a peripheral device interface 1003 and at least one peripheral device. The processor 1001, memory 1002, and peripheral device interface 1003 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 1003 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of: a radio frequency circuit 1004, a display screen 1005, a camera assembly 1006, an audio circuit 1007, and a power supply 1008.

[0208] Peripheral device interface 1003 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 1001 and memory 1002. In some embodiments, processor 1001, memory 1002 and peripheral device interface 1003 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 1001, memory 1002 and peripheral device interface 1003 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0209] The radio frequency (RF) circuit 1004 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 1004 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 1004 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 1004 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 1004 can communicate with other terminals via at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 1004 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0210] Display screen 1005 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 1005 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 1001 for processing. In this case, display screen 1005 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, display screen 1005 may be a single screen, disposed on the front panel of terminal device 1000; in other embodiments, display screen 1005 may be at least two, disposed on different surfaces of terminal device 1000 or in a folded design; in still other embodiments, display screen 1005 may be a flexible display screen, disposed on a curved or folded surface of terminal device 1000. Furthermore, display screen 1005 may also be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. The display screen 1005 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0211] The camera assembly 1006 is used to acquire images or videos. Optionally, the camera assembly 1006 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the terminal, and the rear-facing camera is located on the back of the terminal. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments, the camera assembly 1006 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm-light flash and a cool-light flash, which can be used for light compensation at different color temperatures.

[0212] The audio circuit 1007 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, converting the sound waves into electrical signals that are input to the processor 1001 for processing, or input to the radio frequency circuit 1004 for voice communication. For stereo sound acquisition or noise reduction purposes, multiple microphones may be used, each located at a different part of the terminal device 1000. The microphone may also be an array microphone or an omnidirectional microphone. The speaker is used to convert electrical signals from the processor 1001 or the radio frequency circuit 1004 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into audible sound waves but also into inaudible sound waves for purposes such as distance measurement. In some embodiments, the audio circuit 1007 may also include a headphone jack.

[0213] The power supply 1008 is used to power the various components in the terminal device 1000. The power supply 1008 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When the power supply 1008 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged via a wired line, while a wireless rechargeable battery is a battery that is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0214] In some embodiments, the terminal device 1000 further includes one or more sensors 1009. The one or more sensors 1009 include, but are not limited to: an acceleration sensor 1011, a gyroscope sensor 1012, a pressure sensor 1013, an optical sensor 1014, and a proximity sensor 1015.

[0215] Accelerometer 1011 can detect the magnitude of acceleration along the three coordinate axes of a coordinate system established by terminal device 1000. For example, accelerometer 1011 can be used to detect the components of gravitational acceleration along the three coordinate axes. Processor 1001 can control display screen 1005 to display the user interface in either a landscape or portrait view based on the gravitational acceleration signal acquired by accelerometer 1011. Accelerometer 1011 can also be used for games or for acquiring user motion data.

[0216] The gyroscope sensor 1012 can detect the orientation and rotation angle of the terminal device 1000. The gyroscope sensor 1012 can work in conjunction with the accelerometer sensor 1011 to collect the user's 3D movements on the terminal device 1000. Based on the data collected by the gyroscope sensor 1012, the processor 1001 can perform the following functions: motion sensing (e.g., changing the UI based on the user's tilt), image stabilization during shooting, game control, and inertial navigation.

[0217] The pressure sensor 1013 can be disposed on the side bezel of the terminal device 1000 and / or on the lower layer of the display screen 1005. When the pressure sensor 1013 is disposed on the side bezel of the terminal device 1000, it can detect the user's grip signal on the terminal device 1000, and the processor 1001 can perform left / right hand recognition or quick operation based on the grip signal collected by the pressure sensor 1013. When the pressure sensor 1013 is disposed on the lower layer of the display screen 1005, the processor 1001 can control the operable controls on the UI interface based on the user's pressure operation on the display screen 1005. The operable controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

[0218] An optical sensor 1014 is used to collect ambient light intensity. In one embodiment, the processor 1001 can control the display brightness of the display screen 1005 based on the ambient light intensity collected by the optical sensor 1014. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1005 is increased; when the ambient light intensity is low, the display brightness of the display screen 1005 is decreased. In another embodiment, the processor 1001 can also dynamically adjust the shooting parameters of the camera assembly 1006 based on the ambient light intensity collected by the optical sensor 1014.

[0219] The proximity sensor 1015, also known as a distance sensor, is typically installed on the front panel of the terminal device 1000. The proximity sensor 1015 is used to detect the distance between the user and the front of the terminal device 1000. In one embodiment, when the proximity sensor 1015 detects that the distance between the user and the front of the terminal device 1000 is gradually decreasing, the processor 1001 controls the display screen 1005 to switch from a screen-on state to a screen-off state; when the proximity sensor 1015 detects that the distance between the user and the front of the terminal device 1000 is gradually increasing, the processor 1001 controls the display screen 1005 to switch from a screen-off state to a screen-on state.

[0220] Those skilled in the art will understand that Figure 10 The structure shown does not constitute a limitation on the terminal device 1000, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0221] Figure 11This is a schematic diagram of the server structure provided in the embodiments of this application. The server 1100 can vary considerably due to different configurations or performance. It may include one or more processors 1101 and one or more memories 1102. The one or more memories 1102 store at least one computer program, which is loaded and executed by the one or more processors 1101 to implement the training method or object matching method of the object matching model provided in the above-described method embodiments. For example, the processor 1101 is a CPU. Of course, the server 1100 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server 1100 may also include other components for implementing device functions, which will not be described in detail here.

[0222] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one computer program that is loaded and executed by a processor to enable an electronic device to implement the training method or object matching method of any of the above-described object matching models.

[0223] Optionally, the aforementioned computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0224] In an exemplary embodiment, a computer program or computer program product is also provided, which stores at least one computer program that is loaded and executed by a processor to enable an electronic device to implement the training method or object matching method of any of the above-described object matching models.

[0225] It should be understood that "multiple" as used in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0226] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0227] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A training method for an object matching model, characterized in that, The method comprises: obtaining a reference interaction network, the reference interaction network comprising a plurality of nodes, any one node being a receiving object node or a providing object node, an edge between any one receiving object node and any one providing object node representing a number of interactions between the any one receiving object and the any one providing object, a plurality of receiving objects comprising a sample receiving object; determining, based on the reference interaction network, a relevant providing object from the plurality of providing objects, a possibility of the sample receiving object interacting with the relevant providing object being greater than a first threshold value; determining, by a first network model, a predicted matching result between the sample receiving object and a sample providing object based on basic information of the sample receiving object, basic information of the sample providing object and basic information of the relevant providing object; training the first network model based on the predicted matching result and a labeled matching result to obtain an object matching model, the labeled matching result being a matching result between the sample receiving object and the sample providing object obtained by labeling, the object matching model being used to determine a matching result between a target receiving object and a target providing object.

2. The method of claim 1, wherein, The obtaining of the reference interaction network comprises: obtaining basic information of the plurality of receiving objects, basic information of the plurality of providing objects, and a number of interactions between each receiving object and each providing object; determining, based on the basic information of the plurality of receiving objects, nodes of the plurality of receiving objects included in the reference interaction network; determining, based on the basic information of the plurality of providing objects, nodes of the plurality of providing objects included in the reference interaction network; for any one receiving object and any one providing object, determining an edge between a node of the any one receiving object and a node of the any one providing object based on the number of interactions between the any one receiving object and the any one providing object.

3. The method of claim 2, wherein, The determining of the edge between the node of the any one receiving object and the node of the any one providing object based on the number of interactions between the any one receiving object and the any one providing object comprises: if the number of interactions between the any one receiving object and the any one providing object is greater than a second threshold value, determining the edge between the node of the any one receiving object and the node of the any one providing object based on the number of interactions between the any one receiving object and the any one providing object; if the number of interactions between the any one receiving object and the any one providing object is not greater than the second threshold value, determining that there is no edge between the node of the any one receiving object and the node of the any one providing object.

4. The method of claim 1, wherein, The determining of the relevant providing object from the plurality of providing objects based on the reference interaction network comprises: performing feature extraction on the reference interaction network to obtain first features of each providing object; determining a first similarity between a feature of the sample receiving object and the first features of each providing object, the first similarity between the feature of the sample receiving object and the first feature of any one providing object being used to represent a possibility of the sample receiving object interacting with the any one providing object; If a first similarity between a feature of the sample receiving object and a first feature of the any one providing object is greater than a first threshold, the any one providing object is taken as the relevant providing object.

5. The method of claim 4, wherein, The first feature of the any one providing object is obtained by performing feature extraction on the reference interaction network through a feature extraction model; and the method further comprises: performing feature extraction on the reference interaction network through a second network model to obtain second features of each receiving object and second features of each providing object; for any one receiving object, determining a loss of the any one receiving object based on the second feature of the any one receiving object, the second features of each receiving object, and the second features of each providing object; for any one providing object, determining a loss of the any one providing object based on the second feature of the any one providing object, the second features of each receiving object, and the second features of each providing object; training the second network model based on the losses of each receiving object and the losses of each providing object to obtain the feature extraction model.

6. The method of claim 5, wherein, The determining of the loss of the any one receiving object based on the second feature of the any one receiving object, the second features of each receiving object, and the second features of each providing object comprises: sampling, based on the reference interaction network, a node of a positive sample of the any one receiving object from nodes of providing objects having an edge with a node of the any one receiving object; sampling, based on the reference interaction network, a node of a negative sample of the any one receiving object from nodes of other objects having no edge with the node of the any one receiving object, the other objects being the receiving objects or the providing objects; determining the loss of the any one receiving object based on the second feature of the any one receiving object, the second feature of the positive sample of the any one receiving object, and the second feature of the negative sample of the any one receiving object.

7. The method of claim 1, wherein, The determining of the predicted matching result between the sample receiving object and the sample providing object based on the basic information of the sample receiving object, the basic information of the sample providing object, and the basic information of the relevant providing object through the first network model comprises: determining a first weight of the relevant providing object based on the basic information of the sample receiving object and the basic information of the relevant providing object; determining a first weighted feature based on the first weight and the basic information of the relevant providing object; determining the predicted matching result based on the first weighted feature, the basic information of the sample receiving object, and the basic information of the sample providing object.

8. The method of claim 1, wherein, The determining of the predicted matching result between the sample receiving object and the sample providing object based on the basic information of the sample receiving object, the basic information of the sample providing object, and the basic information of the relevant providing object through the first network model comprises: determining a second weight of the relevant providing object based on the basic information of the sample providing object and the basic information of the relevant providing object; determine a second weighted feature based on the second weight and the correlation of the basic information of the related providing object; determine the predicted matching result based on the second weighted feature, the basic information of the sample receiving object and the basic information of the sample providing object.

9. The method of claim 1, wherein, The method comprises: obtain target information, the target information comprising at least one of the basic information of the historical providing object, the search term input by the sample receiving object, and the environmental information, the historical providing object being a providing object that has historically interacted with the sample receiving object; determine the predicted matching result based on the target information, the basic information of the sample receiving object, the basic information of the sample providing object and the basic information of the related providing object through the first network model.

10. An object matching method characterized by, The method comprises: obtain a reference interaction network, the reference interaction network comprising a plurality of nodes, any one node being a receiving object node or a providing object node, an edge between any one receiving object node and any one providing object node representing the number of interactions between the any one receiving object and the any one providing object, the plurality of receiving objects comprising a target receiving object; determine a reference providing object from the plurality of providing objects based on the reference interaction network, the target receiving object having a possibility of interacting with the reference providing object greater than a first threshold value; determine a matching result between the target receiving object and a target providing object based on the basic information of the target receiving object, the basic information of the target providing object and the basic information of the reference providing object through an object matching model, the object matching model being trained according to the training method of the object matching model of any one of claims 1 to 9.

Images