Information matching method and device, computer device and storage medium

Abstract

The application can be applied to various scenes such as cloud technology, artificial intelligence, intelligent transportation and auxiliary driving, and relates to an information matching method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises: obtaining information features of target information, object features of a target object, and direct interaction behavior features and indirect interaction behavior features of interaction behaviors generated by the target object for historical information; performing information correlation degree matching on the information features and the indirect interaction behavior features to determine indirect information interaction features of the target information, and performing information correlation degree matching on the information features and the direct interaction behavior features to determine direct information interaction features of the target information; and determining a matching result of the target information and the target object based on the indirect information interaction features, the direct information interaction features and object interaction features determined based on the object features. The information matching method has the advantage of high accuracy.

Description

Technical Field

[0001] This application relates to the field of computer application technology, and in particular to an information matching method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of computer technology, more and more users are obtaining information through the internet. Driven by the strong information demands of internet users, information service platforms strive to prevent user churn by continuously providing massive amounts of information resources. However, faced with this vast amount of information, users struggle to quickly and effectively extract the information that meets their needs, leading to information overload. To alleviate this problem, information matching methods have emerged.

[0003] Traditional information matching methods establish user preferences based on historical interactions such as positive reviews, likes, and purchases, which clearly indicate user biases. Information resources matching these preferences are then filtered. However, traditional methods often result in highly similar information to users' historical interactions, leading to a limited range of information types and failing to meet the diverse needs of users. Therefore, traditional information matching methods suffer from low accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide an information matching method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of matching results in response to the above-mentioned technical problems.

[0005] Firstly, this application provides an information matching method. The method includes:

[0006] The information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information are obtained; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features.

[0007] The information features and the indirect interaction behavior features are matched for information correlation to determine the indirect information interaction features of the target information; and the information features and the direct interaction behavior features are matched for information correlation to determine the direct information interaction features of the target information.

[0008] Based on the indirect information interaction features, the direct information interaction features, and the object interaction features, the target information and the target object are matched for similarity of interaction features to determine the matching result between the target information and the target object; the object interaction features are determined based on the object features.

[0009] Secondly, this application also provides an information matching device, characterized in that the device comprises:

[0010] The acquisition module is used to acquire the information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features.

[0011] The information interaction feature determination module is used to perform information correlation matching between the information features and the indirect interaction behavior features to determine the indirect information interaction features of the target information, and to perform information correlation matching between the information features and the direct interaction behavior features to determine the direct information interaction features of the target information.

[0012] The matching module is used to perform interaction feature similarity matching between the target information and the target object based on the indirect information interaction features, the direct information interaction features, and the object interaction features, and to determine the matching result between the target information and the target object; the object interaction features are determined based on the object features.

[0013] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0014] The information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information are obtained; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features.

[0015] The information features and the indirect interaction behavior features are matched for information correlation to determine the indirect information interaction features of the target information; and the information features and the direct interaction behavior features are matched for information correlation to determine the direct information interaction features of the target information.

[0016] Based on the indirect information interaction features, the direct information interaction features, and the object interaction features, the target information and the target object are matched for similarity of interaction features to determine the matching result between the target information and the target object; the object interaction features are determined based on the object features.

[0017] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0018] The information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information are obtained; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features.

[0019] The information features and the indirect interaction behavior features are matched for information correlation to determine the indirect information interaction features of the target information; and the information features and the direct interaction behavior features are matched for information correlation to determine the direct information interaction features of the target information.

[0020] Based on the indirect information interaction features, the direct information interaction features, and the object interaction features, the target information and the target object are matched for similarity of interaction features to determine the matching result between the target information and the target object; the object interaction features are determined based on the object features.

[0021] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0022] The information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information are obtained; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features.

[0023] The information features and the indirect interaction behavior features are matched for information correlation to determine the indirect information interaction features of the target information; and the information features and the direct interaction behavior features are matched for information correlation to determine the direct information interaction features of the target information.

[0024] Based on the indirect information interaction features, the direct information interaction features, and the object interaction features, the target information and the target object are matched for similarity of interaction features to determine the matching result between the target information and the target object; the object interaction features are determined based on the object features.

[0025] The aforementioned information matching method, apparatus, computer equipment, computer-readable storage medium, and computer program product acquire information features of target information, object features of target objects, and direct and indirect interaction features of the target objects' interactions with historical information. Then, information correlation matching is performed between information features and indirect interaction features to determine the indirect information interaction features of the target information, and information correlation matching is performed between information features and direct interaction features to determine the direct information interaction features of the target information. Finally, based on the indirect information interaction features, direct information interaction features, and object interaction features determined based on object features, interaction feature similarity matching is performed between the target information and the target object to determine the matching result. Because the process of matching interaction features between target information and target objects considers both indirect and direct information interaction features determined based on indirect interaction features, it ensures that the information interaction features of the target information are fully represented during the similarity matching process. Mining the potential preferences of the target object from the perspective of information interaction features is beneficial to the accuracy of the interaction feature similarity matching results, thereby improving the accuracy of the matching results between target information and target objects. Attached Figure Description

[0026] Figure 1 This is a diagram illustrating the application environment of the information matching method in one embodiment;

[0027] Figure 2 This is a flowchart illustrating an information matching method in one embodiment;

[0028] Figure 3 This is a flowchart illustrating the information matching method in another embodiment;

[0029] Figure 4 This is a schematic diagram illustrating the process of determining indirect information interaction features in one embodiment;

[0030] Figure 5 This is a schematic diagram illustrating the process of determining the sub-weights of sub-information features in one embodiment;

[0031] Figure 6 This is a schematic diagram illustrating the process of determining target matching information of a target object based on indirect information interaction features, direct information interaction features, and object interaction features in one embodiment.

[0032] Figure 7 This is a flowchart illustrating the information matching method in yet another embodiment;

[0033] Figure 8 This is a schematic diagram illustrating the information matching process in an information push application scenario in one embodiment;

[0034] Figure 9 This is a structural block diagram of an information matching device in one embodiment;

[0035] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0037] In one embodiment, the information matching method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on the cloud or other servers. Specifically, during the information matching process, server 104: acquires the information features of the target information, the object features of the target object, and the direct and indirect interaction behavior features of the target object in response to historical information; performs information correlation matching between information features and indirect interaction behavior features to determine the indirect information interaction features of the target information, and performs information correlation matching between information features and direct interaction behavior features to determine the direct information interaction features of the target information; based on the indirect information interaction features, direct information interaction features, and the object interaction features determined based on object features, performs interaction feature similarity matching between the target information and the target object to determine the matching result between the target information and the target object.

[0038] In one embodiment, if the data processing capability of terminal 102 meets the data processing requirements, the information method provided in this application embodiment may be applied only to terminal 102. Specifically, terminal 102 acquires the information features of target information, the object features of target object, and the direct and indirect interaction behavior features of the target object in response to historical information. Then, it performs information correlation matching between the information features and the indirect interaction behavior features to determine the indirect information interaction features of the target information, and performs information correlation matching between the information features and the direct interaction behavior features to determine the direct information interaction features of the target information. Finally, based on the indirect information interaction features, the direct information interaction features, and the object interaction features determined based on the object features, terminal 102 performs interaction feature similarity matching between the target information and the target object to determine the matching result between the target information and the target object.

[0039] The terminal 102 includes, but is not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, and aircraft. This invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving. The server 104 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal 102 and the server 104 can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0040] In one embodiment, such as Figure 2 As shown, an information matching method is provided. This embodiment illustrates the application of this method to server 104. It is understood that this method can also be applied to terminal 102, and to a system including terminal 102 and server 104, and is implemented through the interaction between terminal 102 and server 104. In this embodiment, the method includes the following steps:

[0041] Step S202: Obtain the information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information.

[0042] In this context, the target object refers to the object to which information recommendation needs to be made based on the information matching results. The target information refers to information whose recommendation to the target object needs to be determined. The types of target information include, but are not limited to, advertisements, articles, news, short videos, and applications. The information features of the target information refer to the features used to characterize the characteristics of the target information, specifically including sub-information features such as identifier features, category features, brand features, and item features. Item features include, but are not limited to, the semantic features and image features contained in the target information. In some embodiments, information features also include context features, including but not limited to the context information of the access request associated with the target information, time, and terminal device features. The object features of the target object refer to the features used to characterize the characteristics of the target object, specifically including, but not limited to, the identifier features, basic attribute features, and behavioral interest features of the target object. Basic attribute features include features such as name, gender, age, and city, while behavioral interest features include browsing behavior interest features and click behavior interest features. It should be noted that the information (including but not limited to identification information, feature information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data 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.

[0043] Specifically, the server can obtain the object parameters of the target object, perform mapping processing on each object parameter to obtain the low-dimensional features corresponding to each object parameter, and then obtain the object features of the target object based on each low-dimensional feature. For example, the server can concatenate the low-dimensional features to obtain the object features of the target object, or it can determine the set of low-dimensional features as the object features of the target object. The object parameters of the target object refer to the parameters corresponding to the object features, used to characterize the features of the target object, including but not limited to object identification parameters, basic attribute parameters, and behavioral interest parameters. For example, basic attribute parameters include name, gender, age, and city. Object parameters can be multi-valued parameters with non-unique values, such as behavioral interest parameters, or single-valued parameters with unique values, such as object identification parameters.

[0044] Similarly, the server can obtain the information parameters of the target information and process each information parameter to obtain the information features of the target information. The specific process for determining information features is detailed in the section on determining object features above, and will not be repeated here. The information parameters of the target information refer to the parameters corresponding to the information features, used to characterize the features of the target information. These parameters include, but are not limited to, information identification parameters, category parameters, brand parameters, item parameters, and context parameters. Similar to object parameters, these information parameters can be multi-valued parameters with non-unique values, such as brand parameters in a joint production scenario, or single-valued parameters with unique values, such as information identification parameters.

[0045] Taking information identifier parameters as an example, the server can convert the information identifier parameter into a hash value (hash id) using a hash function, then use the hash value as the key and the corresponding feature embedding value as the value to store it in a lookup table (embedding table). During the mapping process, the server retrieves the corresponding feature embedding value from the lookup table based on the target information's information identifier parameter; this is the embedding vector corresponding to that information identifier parameter. It should be noted that in the process of obtaining the embedding vectors corresponding to each information parameter, for multi-valued parameters, the server can obtain the embedding vector corresponding to each multi-valued parameter by performing a weighted summation of the embedding sub-vectors corresponding to each parameter value.

[0046] Furthermore, interactive behavior refers to the actions generated by a target object during its interaction with historical information. Historical information refers to information pushed to the target object by the server during its historical interactions with the terminal, and is related to the type of target information. For example, if the target information is a short video, the historical information related to the target information may include the short video itself, advertisements promoting the short video, and applications that play the short video. Specifically, interactive behavior can include direct interactive behavior involving actions with a clear preference for historical information, and indirect interactive behavior involving actions without a clear preference for historical information. Indirect interactive behavior can include browsing and clicking. Direct interactive behavior can include positive interactive behavior involving positive feedback on historical information, such as actions like saving, liking, commenting, downloading, and purchasing, and negative interactive behavior involving negative feedback on historical information, such as rejecting, leaving negative reviews, and uninstalling. Correspondingly, interactive behavior features refer to features used to characterize the characteristics of interactive behavior. Interactive behavior features can include direct interactive behavior features used to characterize the characteristics of direct interactive behavior, and indirect interactive behavior features used to characterize the characteristics of indirect interactive behavior. These interactive behavior features can specifically be in the form of vectors or matrices.

[0047] It is understandable that each independent interactive behavior carries data information such as a timestamp, object data, information data, and behavior type. The timestamp represents the time the interactive behavior occurred, the object data represents the object initiating the interaction, the information data represents the information associated with the interaction, and the behavior type represents the type of interaction. Based on this, the server can perform feature mapping processing on the interaction information of multiple interactive behaviors generated by the target object in response to historical information to obtain multiple sub-interactive behavior features. Then, feature fusion processing is performed on the sub-interactive behavior features of the same type to obtain the direct and indirect interactive behavior features of the target object. The specific algorithm for feature fusion processing can include at least one of the following: summation, averaging, or concatenation. It is understandable that in determining direct interactive behaviors, feature fusion processing on the sub-interactive behavior features corresponding to each positive interactive behavior can obtain the positive interactive behavior features of the target object, and feature fusion processing on the sub-interactive behavior features corresponding to each negative interactive behavior can obtain the negative interactive behavior features of the target object.

[0048] Furthermore, the server can also use a vectorization processing layer to map the features of each sub-interaction behavior into low-dimensional embedding vectors to improve data processing efficiency. The specific method by which the server obtains the information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information can be either active acquisition or passive reception.

[0049] In addition, a candidate information dataset can be pre-set, which includes multiple candidate information items. When the server receives an access request, it obtains the object characteristics of the target object corresponding to the access request and the interactive behavior of the target object in response to historical information. It then retrieves the candidate information corresponding to the target object from the candidate information dataset, determines it as the target information for the target object, and further obtains the information characteristics of the target information.

[0050] Step S204: Perform information correlation matching between information features and indirect interaction behavior features to determine the indirect information interaction features of the target information, and perform information correlation matching between information features and direct interaction behavior features to determine the direct information interaction features of the target information.

[0051] Indirect information interaction features characterize the correlation between target information and indirect interaction behaviors in the information dimension, while direct interaction behavior features characterize the correlation between target information and direct interaction behaviors in the information dimension. That is, indirect information interaction features can be used to characterize the features of indirect interaction behaviors associated with target information, and direct interaction behavior features can be used to characterize the features of direct interaction behaviors associated with target information. In practical applications, target objects generate numerous indirect interaction behaviors related to historical information, and these indirect interaction behaviors can reflect the target object's preferences to a certain extent. For example, the more times a target object has viewed a certain historical piece of information, the greater the likelihood that the target object is interested in that historical information. Based on this, from the perspective of information interaction features, both direct and indirect information interaction features can be comprehensively considered to uncover the target object's potential preferences.

[0052] Specifically, since interactive behavior information carries data, interactive behavior features can characterize the information dimension of interactive behavior. Based on this, the server can perform information correlation matching between information features and indirect interactive behavior features to determine the degree of information correlation between the information features and indirect interactive behavior features, thereby determining the indirect information interaction features of the target information. Similarly, the server can perform information correlation matching between information features and direct interactive behavior features to determine the degree of information correlation between the information features and direct interactive behavior features, thereby determining the direct information interaction features of the target information.

[0053] Furthermore, the server performs information correlation matching between information features and indirect interaction behavior features, and the specific method for determining the information correlation is not unique. For example, the server can calculate the correlation between information features and indirect interaction behavior features in the information dimension based on Convolutional Neural Networks (CNN) or Attention Networks; the server can also determine the correlation between information features and indirect interaction behavior features in the information dimension by calculating the cosine similarity or Euclidean distance between them. It can be understood that the higher the cosine similarity or the smaller the Euclidean distance, the higher the information correlation between the information features and indirect interaction behavior features.

[0054] The specific methods by which a server determines direct and indirect information interaction features are not unique. Taking indirect information interaction features as an example: For instance, the server can match each sub-information feature with the indirect interaction behavior feature to determine the degree of association for each sub-information feature, and then perform a weighted sum based on the degree of association to determine the indirect information interaction features of the target information. Alternatively, the server can match each sub-information feature with the indirect interaction behavior feature to determine the degree of association for each sub-information feature, and then determine the indirect information interaction features of the target information based on the sub-information features whose degree of association satisfies the information association condition.

[0055] Regarding the specific methods by which the server matches information features with direct interaction behavior features to determine the degree of information correlation, and then determines the direct information interaction features, please refer to the process for determining indirect information interaction features above; it will not be repeated here. Furthermore, the specific methods by which the server determines direct and indirect information interaction features can be the same or different.

[0056] Step S208: Based on indirect information interaction features, direct information interaction features, and object interaction features, perform interaction feature similarity matching between target information and target object to determine the matching result between target information and target object.

[0057] Among them, object interaction features are used to characterize the target object's interest in information, and these features can be determined based on object features. Since object features include the target object's behavioral interest features, the server can determine object interaction features based on these features. For example, the server can use the target object's behavioral interest features as its object interaction features; alternatively, the server can match each sub-object feature with the positive interaction behavior features to determine the sub-object correlation degree corresponding to each sub-object feature, then select target sub-object features whose sub-object correlation degrees satisfy the object correlation conditions from among the sub-object features, and finally sum the target sub-object features to determine the target object's object interaction features.

[0058] Specifically, the server can perform interaction feature similarity matching between target information and target object based on indirect information interaction features, direct information interaction features, and object interaction features to obtain the matching degree between target information and target object, and then determine the matching result between target information and target object. Target information with a matching degree higher than a set threshold is identified as the proposed recommendation information for the target object.

[0059] It should be noted that the specific method by which the server performs interaction feature similarity matching between target information and target object based on indirect information interaction features, direct information interaction features, and object interaction features is not unique. For example, the server can use a neural network model to perform feature matching on indirect information interaction features, direct information interaction features, and object interaction features to obtain the matching degree between target information and target object; alternatively, the server can first perform feature fusion processing on indirect and direct information interaction features to obtain comprehensive information interaction features of the target information, and then perform similarity matching on this comprehensive information interaction features and object interaction features to determine the matching degree between target information and target object.

[0060] The aforementioned information matching method acquires the information features of the target information, the object features of the target object, and the direct and indirect interaction features of the target object's interaction behavior with historical information. Then, it performs information correlation matching between the information features and the indirect interaction features to determine the indirect information interaction features of the target information, and performs information correlation matching between the information features and the direct interaction features to determine the direct information interaction features of the target information. Finally, based on the indirect information interaction features, the direct information interaction features, and the object interaction features determined based on the object features, it performs interaction feature similarity matching between the target information and the target object to determine the matching result. Because the process of matching the interaction features between the target information and the target object considers both the indirect information interaction features determined by the indirect interaction features and the direct information interaction features determined by the direct interaction features, it ensures that the interaction features of the target information are fully represented during the similarity matching process. By mining the potential preferences of the target object from the perspective of information interaction features, it improves the accuracy of the interaction feature similarity matching results, thereby enhancing the accuracy of the matching results between the target information and the target object.

[0061] As mentioned above, the specific method by which the server determines the characteristics of indirect information interaction is not unique. In one embodiment, the information characteristics include multiple sub-information characteristics. In the case of this embodiment, such as Figure 3 As shown, information correlation is matched between information features and indirect interaction behavior features to determine the indirect information interaction features of the target information, including:

[0062] Step S303: Match each sub-information feature with the indirect interaction behavior feature to determine the sub-information correlation degree corresponding to each sub-information feature.

[0063] The sub-information correlation degree corresponding to each sub-information feature refers to the information correlation degree between each sub-information feature and the indirect interaction behavior feature in terms of information dimension. For the specific method by which the server determines the sub-information correlation degree, please refer to the section above on the server's method of matching information features with indirect interaction behavior features to determine the information correlation degree; it will not be repeated here.

[0064] Specifically, such as Figure 4 As shown, the server obtains n information parameters of the target information, including information parameter I1, information parameter I2, ..., information parameter I... n The information parameters are mapped to obtain the sub-information features corresponding to each information parameter, including the sub-information features. Sub-information features ...sub-information features Then, the server matches each sub-information feature with the indirect interaction behavior feature to determine the sub-information relevance degree corresponding to each sub-information feature, including sub-information relevance degree k1, sub-information relevance degree k2, ..., sub-information relevance degree k n .

[0065] Step S304: Based on the sub-information features that satisfy the information association conditions, determine the indirect information interaction features of the target information.

[0066] The information association condition can be that the sub-information association degree is greater than a set information association degree threshold, or that the sub-information association degree is greater than or equal to the set information association degree threshold, or that the sub-information association degree ranking is higher than a set ranking. The specific data form of the set ranking can be an integer or a percentage.

[0067] Specifically, such as Figure 4 As shown, the server determines the indirect information interaction features of the target information based on the features of sub-information whose sub-information relevance satisfies the information association condition. Furthermore, the server can determine the indirect information interaction features of the target information by performing feature concatenation and dimensionality reduction on the features of sub-information whose sub-information relevance satisfies the information association condition; alternatively, the server can determine the indirect information interaction features of the target information by performing summation or averaging operations on the features of sub-information whose sub-information relevance satisfies the information association condition.

[0068] In one embodiment, step S304 includes: selecting target sub-information features from the sub-information features whose sub-information correlation degree meets the information correlation condition; and performing a summation operation on each target sub-information feature to determine the indirect information interaction features of the target information.

[0069] Specifically, the server can filter target sub-information features that satisfy the information association condition from among the sub-information features based on the sub-information correlation degree corresponding to each sub-information feature. Then, it can perform a summation operation on each target sub-information feature to determine the indirect information interaction features of the target information. Furthermore, the summation operation on each target sub-information feature can refer to directly superimposing each target sub-information feature, or it can refer to performing a weighted summation on each target sub-information feature based on the sub-information correlation degree corresponding to each target sub-information feature.

[0070] In the above embodiments, the correlation degree of each sub-information feature is first determined, and then the indirect information interaction features of the target information are determined based on the sub-information features whose correlation degree satisfies the information association condition. This can filter out sub-information features with low correlation degree with indirect interaction behavior, which is beneficial to improving the accuracy of indirect information interaction features.

[0071] Furthermore, as mentioned above, the specific method by which the server determines the direct information interaction characteristics is not unique. In one embodiment, the information characteristics include multiple sub-information characteristics, and the direct interaction behavior characteristics include positive interaction behavior characteristics and negative interaction behavior characteristics. In the case of this embodiment, please continue to refer to... Figure 3 Information features are matched with direct interaction behavior features to determine the direct information interaction features of the target information, including:

[0072] Step S305: Calculate the first sub-information correlation degree between each sub-information feature and the positive interaction behavior feature, and the second sub-information correlation degree between each sub-information feature and the negative interaction behavior feature.

[0073] The first sub-information correlation degree refers to the information correlation between sub-information features and positive interaction behavior features in the information dimension. Correspondingly, the second sub-information correlation degree refers to the information correlation between sub-information features and negative interaction behavior features in the information dimension. For example, if a brand appears in multiple positive interaction behaviors of a target object, then the corresponding positive interaction behavior feature has a high information correlation degree with the brand information corresponding to that brand. Regarding the specific methods by which the server calculates the first and second sub-information correlation degrees, please refer to the section above on the server's method of matching information features with indirect interaction behavior features to determine information correlation degrees; this will not be repeated here.

[0074] Specifically, the server can match each sub-information feature with the positive interaction behavior feature to determine the first sub-information correlation degree corresponding to each sub-information feature, and match each sub-information feature with the negative interaction behavior feature to determine the second sub-information correlation degree corresponding to each sub-information feature.

[0075] Step S306: Based on the correlation degree of the first sub-information and the correlation degree of the second sub-information, determine the sub-weights corresponding to each sub-information feature.

[0076] The sub-weights of sub-information features are used to characterize the degree of correlation between the sub-information feature and the target object's interest tendencies. For example, the identifier feature has a low correlation with the target object's interest tendencies, while the brand feature has a high correlation.

[0077] Specifically, the server determines the sub-weights corresponding to each sub-information feature based on the correlation between the first and second sub-information features in a different way.

[0078] In one embodiment, the server subtracts the relevance of the second sub-information from the relevance of the first sub-information to determine the comprehensive relevance of each sub-information feature, and then determines the sub-weight corresponding to the comprehensive relevance. It can be understood that a higher comprehensive relevance indicates a greater relevance between the sub-information feature and the positive interaction behavior relative to negative interaction behavior, and thus a larger corresponding sub-weight.

[0079] In another embodiment, step S306 includes: determining the positive sub-weights corresponding to each sub-information feature based on the first sub-information correlation degree; determining the negative sub-weights corresponding to each sub-information feature based on the second sub-information correlation degree; and superimposing the positive and negative sub-weights corresponding to the same sub-information feature to obtain the sub-weights corresponding to each sub-information feature.

[0080] Among them, the higher the relevance of the first sub-information, the larger the corresponding positive sub-weight; the higher the relevance of the second sub-information, the larger the absolute value of the corresponding negative sub-weight. Specifically, for example... Figure 5 As shown, on one hand, the server determines the first sub-information correlation degree between the sub-information feature and the positive interaction behavior feature, and then determines the corresponding positive sub-weight based on the first sub-information correlation degree; on the other hand, the server determines the second sub-information correlation degree between the sub-information feature and the negative interaction behavior feature, and then determines the corresponding negative sub-weight based on the second sub-information correlation degree. Then, the server superimposes the positive and negative sub-weights corresponding to the same sub-information feature to obtain the sub-weights corresponding to each sub-information feature.

[0081] Because user preferences fluctuate, the same target information may be associated with both positive and negative interaction behaviors. In this embodiment, based on the first sub-information correlation degree, the positive sub-weights corresponding to each sub-information feature are determined, and based on the second sub-information correlation degree, the negative sub-weights corresponding to each sub-information feature are determined. Then, the positive and negative sub-weights corresponding to the same sub-information feature are superimposed to obtain the sub-weights corresponding to each sub-information feature. This can offset the correlation degree between each sub-information feature and the negative interaction behavior feature, which is equivalent to taking into account the impact of user preference fluctuations. This helps to improve the accuracy of the sub-weights corresponding to the sub-information features, thereby improving the accuracy of the direct information interaction features of the target information.

[0082] Step S307: Based on the sub-weights corresponding to each sub-information feature, perform a weighted summation of each sub-information feature to determine the direct information interaction features of the target information.

[0083] Specifically, the server performs a weighted summation of each sub-information feature based on its corresponding sub-weight, thereby determining the direct information interaction features of the target information.

[0084] In the above embodiments, based on the first sub-information correlation degree between each sub-information feature and the positive interaction behavior feature, and the second sub-information correlation degree between each sub-information feature and the negative interaction behavior feature, the sub-weights corresponding to each sub-information feature are determined. Then, based on the sub-weights, each sub-information feature is weighted and summed to determine the direct information interaction feature of the target information. This is equivalent to matching the information features with different types of direct interaction behavior features based on their information correlation degree, which helps to improve the accuracy of the direct information interaction feature.

[0085] In one embodiment, please refer to [link / reference]. Figure 3 Step S208 includes:

[0086] Step S308: Perform feature fusion processing on the indirect information interaction features and the direct information interaction features to obtain the comprehensive information interaction features of the target information.

[0087] Among them, the comprehensive information interaction features of the target information are used to characterize the correlation between the target information and the interaction behavior in the information dimension. That is, the comprehensive information interaction features can characterize the features of the interaction behavior associated with the target information.

[0088] Specifically, the server can directly perform feature operations on indirect and direct information interaction features to obtain a feature fusion result, which is then determined as the comprehensive information interaction feature of the target information. Alternatively, the server can perform feature operations on indirect and direct information interaction features, obtain the result, concatenate it with the indirect and direct information interaction features, and finally output the concatenated feature fusion result through a multilayer perceptron (MLP). After dimensionality reduction processing, the comprehensive information interaction feature of the target information is obtained. The specific content of the feature operations can be one or more combinations of various operations such as addition, subtraction, dot product, and Hadamard product.

[0089] Step S309: Perform similarity matching on the comprehensive information interaction features and object interaction features to determine the matching result between the target information and the target object.

[0090] Specifically, the server performs similarity matching on comprehensive information interaction features and object interaction features to determine the matching result between target information and target object in various ways. For example, the server can perform similarity matching on comprehensive information interaction features and object interaction features based on the target neural network model to determine the matching result between target information and target object; alternatively, the server can calculate the cosine similarity or Euclidean distance between comprehensive information interaction features and object interaction features to obtain a similarity score, and then determine the matching result between target information and target object based on this similarity score. It can be understood that the higher the cosine similarity or the smaller the Euclidean distance, the higher the similarity score between comprehensive information interaction features and object interaction features, indicating a higher degree of matching between target information and target object.

[0091] In the above embodiments, feature fusion processing is performed on indirect information interaction features and direct information interaction features to obtain comprehensive information interaction features of target information. Then, information matching is performed between the comprehensive information interaction features and object interaction features to determine the matching result between target information and target object, which helps to improve data processing efficiency.

[0092] In one embodiment, step S309 includes: determining object interaction features based on object features; performing similarity matching on the comprehensive information interaction features and object interaction features based on the target neural network model to obtain a matching degree calculation value between the target information and the target object; and determining the matching result between the target information and the target object based on the matching degree calculation value.

[0093] The specific algorithm for determining object interaction characteristics based on object features by the server is described above and will not be repeated here. The target neural network model refers to a model obtained based on neural networks and machine learning algorithms. It reflects many fundamental characteristics of human brain function and is a highly complex nonlinear dynamic learning system capable of processing information under multiple factors and conditions.

[0094] Specifically, the server may train a model using a training sample set containing sample information interaction features and sample object interaction features. This model yields a target neural network model used to determine the similarity between sample information interaction features and sample object interaction features, thereby determining the matching degree calculation value between target information and target object. Based on this target neural network model, similarity matching is performed on the comprehensive information interaction features and object interaction features to obtain the matching degree calculation value between target information and target object. The matching result between target information and target object is determined based on the matching degree calculation value. Target information whose matching degree calculation value meets the matching condition is identified as the proposed recommended information for the target object. This matching condition can be that the matching degree calculation value is greater than a set matching degree threshold, or that the matching degree calculation value is greater than or equal to a set matching degree threshold.

[0095] In the above embodiments, based on the target neural network model, similarity matching is performed on the comprehensive information interaction features and object interaction features to obtain the matching degree calculation value between the target information and the target object, and then the matching result between the target information and the target object is determined. Thanks to the advantages of the neural network model, such as large-scale parallelism, distributed storage and processing, self-organization, self-adaptation and self-learning capabilities, it is beneficial to improve the data processing capability of the information matching method.

[0096] In one embodiment, the process of training a target neural network model includes: determining the loss weight of the sample information based on the number of interaction behaviors generated by the sample objects in response to the sample information; weighting and summing the differences between the calculated matching degree values ​​of each sample object and each sample information and the actual matching degree values ​​of each sample object and each sample information according to the loss weights corresponding to each sample information, to obtain the loss function of the target neural network model; and training the model using a training sample set including the sample information interaction features of the sample information and the sample object interaction features of the sample objects, based on the loss function, to obtain the target neural network model used to determine the calculated matching degree values ​​between the target information and the target object.

[0097] Among them, sample information interaction features are used to characterize the interactive behaviors associated with sample information. Sample object interaction features are used to characterize the interactive behaviors associated with sample objects. Further, during model training, the difference between the calculated matching degree and the actual matching degree is calculated using a loss function. This loss function can specifically be a mean squared error loss function, an L1 loss function, or a cross-entropy loss function, etc. Correspondingly, the difference between the calculated matching degree and the actual matching degree can be the square or absolute value of their difference, or it can be the cross-entropy loss. In summary, this embodiment does not limit the specific type of loss function used to calculate the difference between the calculated matching degree and the actual matching degree.

[0098] Specifically, the server trains the model using a training sample set that includes sample information interaction features and sample object interaction features, based on a loss function, to obtain a target neural network model used to determine the matching degree calculation value between target information and target object. The process of determining the loss function includes: first, the server determines the loss weight of the sample information based on the number of interaction behaviors generated by the sample object towards the sample information; then, based on the respective loss weights of each sample information, the server performs a weighted summation of the difference between the calculated matching degree values ​​of each sample object and each sample information, and the actual matching degree values ​​of each sample object and each sample information, to obtain the loss function of the target neural network model.

[0099] It is understandable that the more interactions a sample object makes with the sample information, the greater the loss weight. In one embodiment, this loss weight c ui The calculation formula is:

[0100] c ui =1+α*C

[0101] Where α is a coefficient, ranging from [0, 100], and C represents the number of interaction behaviors generated by the sample object in response to the sample information. This coefficient can be dynamically adjusted based on segmentation testing (A / B testing) to improve the accuracy of the neural network model.

[0102] Furthermore, a regularization term can be constructed into the loss function of the target neural network model. This regularization term can be the product of the sum of squares of sample information interaction features, sample object interaction features, and the error term, plus a set parameter. Similarly, this set parameter can be dynamically adjusted based on segmentation tests to further improve the accuracy of the neural network model.

[0103] In the above embodiments, the loss weight of the sample information is determined based on the number of interaction behaviors generated by the sample object in relation to the sample information. Then, based on the loss weight corresponding to each sample information, the difference between the calculated matching degree value of each sample object and each sample information and the actual matching degree value of each sample object and each sample information is weighted and summed to obtain the loss function of the target neural network model. This is beneficial to improve the accuracy of the target neural network model trained based on the loss function, thereby improving the accuracy of the matching results.

[0104] In one embodiment, there are multiple target information items. In this embodiment, step S309 includes: performing clustering processing on the comprehensive information interaction features corresponding to each target information item to obtain clustered information interaction features corresponding to each target information item; performing similarity matching between each clustered information interaction feature and object interaction features to determine the matching degree corresponding to each clustered information interaction feature; and filtering target matching information of target objects from each target information item based on the matching degree corresponding to each clustered information interaction feature.

[0105] Clustering is the process of dividing multiple data features into clusters composed of similar features. A cluster generated by clustering is a collection of data objects. After clustering, data objects within the same cluster are highly similar, while data objects in different clusters are highly dissimilar. Specifically, in this application, after clustering the comprehensive information interaction features corresponding to each target information, multiple clusters composed of comprehensive information interaction features can be obtained. The comprehensive information interaction features within the same cluster are similar to each other, while the comprehensive information interaction features in different clusters are dissimilar to each other. Clustered information interaction features refer to information interaction features that can characterize the features of each comprehensive information interaction feature within the same cluster.

[0106] Specifically, such as Figure 6As shown, the server fuses indirect and direct information interaction features to obtain comprehensive information interaction features. Then, the server clusters these comprehensive information interaction features for each target information into multiple distinct clusters. The specific algorithm used for clustering these features can be one or a combination of various clustering algorithms, such as prototype clustering, density clustering, hierarchical clustering, model clustering, and spectral clustering. Next, the server calculates the clustering information interaction features for each cluster based on the comprehensive information interaction features within that cluster. For example, the server can use the average value of the comprehensive information interaction features within the same cluster as the clustering information interaction feature, or it can use the comprehensive information interaction feature closest to the cluster center within the same cluster as the clustering information interaction feature. Finally, the server performs similarity matching between each clustering information interaction feature and the object interaction features to determine the matching degree for each clustering information interaction feature. Based on the matching degree of each clustering information interaction feature, the server filters out the target matching information for the target object from the target information.

[0107] Furthermore, the server performs similarity matching between the cluster information interaction features and the object interaction features, and the specific method for determining the matching degree corresponding to each cluster information interaction feature is not unique. For example, the server can calculate the matching degree between the cluster information interaction features and the object interaction features based on the target neural network model; the server can also determine the matching degree by calculating the cosine similarity or Euclidean distance between the cluster information interaction features and the object interaction features.

[0108] In the above embodiments, the clustered information interaction features obtained after clustering the comprehensive information interaction features are similar to the object interaction features, and the target matching information of the target object is obtained from multiple target information. Since the number of clustered information interaction features obtained after clustering is necessarily less than the number of comprehensive information interaction features, the workload in the similarity matching process can be reduced, which is conducive to improving data processing efficiency.

[0109] It should be noted that the specific method by which the server filters out the target matching information of the target object from each target information based on the matching degree corresponding to the interaction features of each cluster information is not unique.

[0110] In one embodiment, the server determines the target information corresponding to the cluster information interaction features with the highest matching degree as the target matching information of the target object.

[0111] In another embodiment, target matching information of the target object is obtained from each target information based on the matching degree corresponding to each cluster information interaction feature, including: when the matching degree corresponding to the cluster information interaction feature meets the first matching condition, each target information corresponding to the cluster information interaction feature is determined as candidate matching information; the comprehensive information interaction feature of each candidate matching information is matched with the object interaction feature to obtain the matching degree corresponding to each candidate matching information; when the matching degree corresponding to the candidate matching information meets the second matching condition, the candidate matching information is determined as the target matching information of the target object.

[0112] The first matching condition can refer to a matching degree greater than a first matching degree threshold, or a matching degree greater than or equal to the first matching degree threshold. The second matching condition can also refer to a matching degree greater than a second matching degree threshold, or a matching degree greater than or equal to the second matching degree threshold. The first matching degree threshold is less than the second matching degree threshold. Further, the second matching condition can also refer to the ranking of the matching degree corresponding to the candidate matching information within the same cluster being before a set ranking threshold. The specific data form of the set ranking threshold can be an integer or a percentage. For example, the set ranking threshold can be 4 or 5, or 10% or 20%, etc. Using a set ranking threshold to limit the second matching condition ensures that the target matching information is distributed in different clusters, which is beneficial to improving the diversity of the target matching information.

[0113] Specifically, when the matching degree corresponding to the clustering information interaction feature meets the first matching condition, the server determines each target information corresponding to the clustering information interaction feature as candidate matching information. Then, it performs similarity matching between the comprehensive information interaction feature of each candidate matching information and the object interaction feature to obtain the matching degree corresponding to each candidate matching information. When the matching degree corresponding to the candidate matching information meets the second matching condition, the candidate matching information is determined as the target matching information of the target object.

[0114] In the above embodiments, the target matching information of the target object is determined by two screenings. First, clusters with relatively high matching degree are selected, and the target information corresponding to the cluster is determined as candidate matching information. Then, the candidate matching information and the target object are matched by interaction feature similarity to further screen and obtain the target matching information. This can reduce the data scale of similarity matching in the second screening process and improve data processing efficiency while ensuring the accuracy of the matching results.

[0115] In one embodiment, such as Figure 7 As shown, the information matching methods include:

[0116] Step S701: Obtain multiple object parameters of the target object;

[0117] Step S702: Map the parameters of each object to obtain multiple sub-object features of the target object;

[0118] Step S703: Obtain multiple information parameters of the target information;

[0119] Step S704: Map each information parameter to obtain multiple sub-information features of the target information;

[0120] Step S705: Obtain the positive and negative interaction behavior information of the target object in response to historical information;

[0121] Step S706: Map the positive and negative interaction behavior information respectively to obtain the positive and negative interaction behavior features of the target object's direct interaction behavior in response to historical information.

[0122] Step S707: Obtain indirect interaction behavior information of the target object in response to historical information;

[0123] Step S708: Map the indirect interaction behavior information to obtain the indirect interaction behavior characteristics of the target object in response to historical information.

[0124] Step S709: Match the object association degree of each sub-object feature with the positive interaction behavior feature to determine the sub-object association degree corresponding to each sub-object feature.

[0125] Step S710: From the features of each sub-object, select the target sub-object features whose sub-object correlation degree meets the object correlation condition;

[0126] Step S711: Sum the features of each target sub-object to determine the object interaction features of the target object;

[0127] Step S712: Calculate the first sub-information correlation degree between each sub-information feature and the positive interaction behavior feature, and the second sub-information correlation degree between each sub-information feature and the negative interaction behavior feature.

[0128] Step S713: Based on the first sub-information correlation degree, determine the positive sub-weights corresponding to each sub-information feature; based on the second sub-information correlation degree, determine the negative sub-weights corresponding to each sub-information feature; superimpose the positive and negative sub-weights corresponding to the same sub-information feature to obtain the sub-weights corresponding to each sub-information feature.

[0129] Step S714: Based on the sub-weights corresponding to each sub-information feature, perform a weighted summation of each sub-information feature to determine the direct information interaction features of the target information.

[0130] Step S715: Match each sub-information feature with the indirect interaction behavior feature to determine the sub-information correlation degree corresponding to each sub-information feature.

[0131] Step S716: From each sub-information feature, select the target sub-information feature whose sub-information correlation degree meets the information correlation condition;

[0132] Step S717: Summation operation is performed on the features of each target sub-information to determine the indirect information interaction features of the target information;

[0133] Step S718: Perform feature fusion processing on the indirect information interaction features and the direct information interaction features to obtain the comprehensive information interaction features of the target information;

[0134] Step S719: Perform clustering processing on the comprehensive information interaction features corresponding to each target information to obtain the clustered information interaction features corresponding to each target object.

[0135] Step S720: Determine the loss weight of the sample information based on the number of interaction behaviors generated by the sample object in response to the sample information.

[0136] Step S721: Based on the loss weights corresponding to each sample information, the difference between the calculated matching degree between each sample object and each sample information and the actual matching degree between each sample object and each sample information is weighted and summed to obtain the loss function of the target neural network model.

[0137] Step S722: Based on the loss function, the model is trained using a training sample set including the sample object interaction features and the sample information interaction features of the sample information to obtain a target neural network model used to determine the matching degree calculation value between the target information and the target object.

[0138] Step S723: Based on the target neural network model, perform similarity matching between the information interaction features of each cluster and the object interaction features to determine the matching degree corresponding to each cluster information interaction feature.

[0139] Step S724: If the matching degree corresponding to the clustering information interaction feature satisfies the first matching condition, each target information corresponding to the clustering information interaction feature is determined as candidate matching information.

[0140] Step S725: Based on the target neural network model, perform similarity matching between the comprehensive information interaction features and object interaction features of each candidate matching information to obtain the matching degree corresponding to each candidate matching information.

[0141] Step S726: If the matching degree corresponding to the candidate matching information meets the second matching condition, the candidate matching information is determined as the target matching information of the target object.

[0142] This application also provides an application scenario for application push notifications, which utilizes the aforementioned information matching method. In this scenario, the target user accesses an application marketplace via a control terminal to search for applications of interest. During the process of determining the application to be pushed, the server first acquires the information features of the target application to be matched, the object features of the target user, and the direct and indirect interaction behavior features of the target user's interactions with previously pushed applications. Then, it performs information correlation matching between the information features and the indirect interaction behavior features to determine the indirect information interaction features of the target application, and performs information correlation matching between the information features and the direct interaction behavior features to determine the direct information interaction features of the target application. Finally, based on the indirect information interaction features, the direct information interaction features, and the object interaction features determined based on the object features, the server performs interaction feature similarity matching between the target application and the target user to determine the matching result. This matching result characterizes the matching degree between the target application and the target user. The server can push the target application with the highest matching degree to the target user based on multiple matching degrees between target applications and target users.

[0143] This application also provides an application scenario for information push, in which the server executes the above-mentioned information matching method to determine the matching result between the target information and the target object, so as to determine the information to be pushed to the target object based on the matching result.

[0144] Specifically, such as Figure 8 As shown, the server first constructs the information characteristics of the target information, the object characteristics of the target object, and the interaction behavior characteristics of the target object in response to historical information. The interaction behavior of the target object includes direct interaction behavior and indirect interaction behavior. Correspondingly, the interaction behavior characteristics include direct interaction behavior characteristics and indirect interaction behavior characteristics.

[0145] In constructing information features and object features, the server obtains the information parameters of the target information and the object parameters of the target object. The embedding layer maps each information parameter and object parameter separately, obtaining a low-dimensional embedding vector corresponding to each parameter. The vector set formed by the low-dimensional vectors corresponding to each object parameter is then used as the object feature of the target object, and the vector set formed by the low-dimensional vectors corresponding to each information parameter is used as the information feature of the target information. Specifically, object parameters include, but are not limited to, the target object's identifier parameters, basic attribute parameters (e.g., name, gender, age, and city), and behavioral interest parameters (e.g., browsing behavior interest parameters and click behavior interest parameters). Information parameters include, but are not limited to, the target information's identifier parameters, category parameters (e.g., news, entertainment), brand parameters (e.g., the platform on which the information is published), item parameters (e.g., semantic parameters and image parameters), and context parameters (e.g., the context information of the access request associated with the target information, time, and terminal device parameters). Furthermore, in obtaining the low-dimensional feature vectors, a corresponding feature bias can be determined for each feature vector to increase the fitting ability. Specifically, the i-th object feature vector can be represented as... The feature bias corresponding to the feature vector of this object can be expressed as... The i-th information feature vector can be represented as: The feature bias corresponding to this information feature vector can be expressed as follows:

[0146] In constructing interactive behavior features, the server, based on the embedding layer, maps interactive behavior information of the same type into low-dimensional embedding vectors. Then, feature fusion processing is performed on each embedding vector to obtain the direct and indirect interactive behavior features of the target object. The specific algorithm for feature fusion processing can include at least one of the following: summation, averaging, or concatenation.

[0147] After constructing the information features, object features, and interaction behavior features, the server determines the object interaction vector based on the object features. The object interaction vector can be the sum of the object feature vectors within the object features, i.e.:

[0148]

[0149] In the formula, q u F is the object interaction vector. U This represents the number of object feature vectors contained in the object's features. The corresponding object bias is then b. u It can be represented as:

[0150]

[0151] On the other hand, the server matches the information feature vectors contained in the information features with the direct interaction behavior features to determine the information correlation degree, determines the sub-weights corresponding to each information feature vector, and then performs a weighted summation of each information feature vector based on the sub-weights to determine the direct information interaction vector of the target information.

[0152] Specifically, direct interaction behaviors include positive interaction behaviors that provide positive feedback to historical information, and negative interaction behaviors that provide negative feedback to historical information. The server calculates the first sub-information correlation degree between each information feature vector and the positive interaction behavior feature, and the second sub-information correlation degree between each information feature vector and the negative interaction behavior feature. Based on the first sub-information correlation degree, the positive sub-weights corresponding to each information feature vector are determined, and based on the second sub-information correlation degree, the negative sub-weights corresponding to each information feature vector are determined. Then, the positive and negative sub-weights corresponding to the same information feature vector are superimposed to obtain the sub-weights corresponding to each information feature vector. Finally, based on the sub-weights corresponding to each information feature vector, the information feature vectors are weighted and summed to obtain the direct information interaction vector of the target information, i.e.:

[0153]

[0154] In the formula, p I A is a direct information interaction vector. i Information feature vector The corresponding sub-weights, F I Let b be the number of information feature vectors contained in the information feature. Then the corresponding information bias b is... I It can be represented as:

[0155]

[0156] Furthermore, the server determines the indirect information interaction vector of the target information by matching the information features and indirect interaction behavior features based on information relevance. Specifically, the server matches each information feature vector contained in the information features with the indirect interaction behavior features to determine the sub-information relevance corresponding to each information feature vector, and then assigns information feature vectors whose sub-information relevance satisfies the information relevance condition. Perform a summation operation to obtain the indirect information interaction vector X of the target information. I ,Right now:

[0157]

[0158] In the formula, F I2 This represents the number of information feature vectors whose sub-information correlation degree satisfies the information correlation condition.

[0159] After determining the object interaction vector, indirect information interaction vector, and direct information interaction vector, the server performs interaction feature similarity matching on the target information and the target object based on these three interaction vectors to determine the matching result. The specific method by which the server performs interaction feature similarity matching on the target information and the target object based on the object interaction vector, indirect information interaction vector, and direct information interaction vector is not unique.

[0160] In one embodiment, the server uses a target neural network model based on the object interaction vector q. u Direct information interaction vector p I and indirect information interaction vector X I The interaction feature similarity matching of target information and target object is performed to determine the matching degree score r between target information and target object. u :

[0161]

[0162] Where f is the sigmod function:

[0163]

[0164] Furthermore, the loss function L of the target neural network model is:

[0165]

[0166] In the formula, k is the number of sample pairs consisting of the sample object and the sample information, and r ui Let Y be the matching score between the i-th sample object and the sample information calculated by the target neural network model, and let Y be the actual matching score between the i-th sample object and the sample information. ui Let λ be the loss weight corresponding to the i-th sample group, and λ be a set parameter. u || 2 +||p i || 2 +b i 2 +b u 2 ) is a regular term.

[0167] Wherein, the loss weight c ui The calculation formula is:

[0168] c ui =1+α*C

[0169] In the formula, α is a coefficient with a value between [0, 100], and C represents the number of interaction behaviors generated by the sample object in response to the sample information. This coefficient can be dynamically adjusted based on segmentation tests to improve the accuracy of the neural network model.

[0170] In another embodiment, the server first processes the direct information interaction vector p I and indirect information interaction vector X I Feature fusion processing is performed to obtain the comprehensive information interaction vector Q of the target information. I Then, for each target information, the comprehensive information interaction vector Q is... I Clustering is performed to form N clusters, each containing a comprehensive information interaction vector of M target information, and the clustering information interaction vector of each cluster is K. i This is the average of the comprehensive information interaction vectors of this cluster, i.e.:

[0171]

[0172] Then, the server calculates the object interaction vector q respectively. u Interaction vector K with each cluster information i vector dot product:

[0173] R ui =q u *K i

[0174] Next, the server sorts the vector dot products of each cluster and selects the n clusters with the highest scores. Finally, the server multiplies each comprehensive information interaction vector and object interaction vector in the n highest-scoring clusters pairwise to obtain the cosine similarity between each comprehensive information interaction vector and the object interaction vector. The target information corresponding to the m comprehensive information interaction vectors with the highest cosine similarity in each cluster is then identified as the proposed recommended information for the target object. This method not only significantly reduces the computational load in the matching process but also ensures that the target information is distributed across multiple clusters, thus improving the diversity of the proposed recommended information.

[0175] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0176] Based on the same inventive concept, this application also provides an information matching apparatus for implementing the information matching method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more information matching apparatus embodiments provided below can be found in the limitations of the information matching method described above, and will not be repeated here.

[0177] In one embodiment, such as Figure 9 As shown, an information matching device 900 is provided, including: an acquisition module 901, an information interaction feature determination module 902, and a matching module 903, wherein:

[0178] The acquisition module 901 is used to acquire the information features of the target information, the object features of the target object, and the interaction behavior features of the target object in response to historical information; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features.

[0179] The information interaction feature determination module 902 is used to match the information features with the indirect interaction behavior features to determine the indirect information interaction features of the target information, and to match the information features with the direct interaction behavior features to determine the direct information interaction features of the target information.

[0180] The matching module 903 is used to perform interaction feature similarity matching between target information and target object based on indirect information interaction features, direct information interaction features, and object interaction features, and to determine the matching result between target information and target object; the object interaction features are determined based on object features.

[0181] In one embodiment, the information features include multiple sub-information features. In this embodiment, the information interaction feature determination module 902 includes: an indirect information correlation degree determination unit, used to match each sub-information feature with an indirect interaction behavior feature to determine the sub-information correlation degree corresponding to each sub-information feature; and an indirect information interaction feature determination unit, used to determine the indirect information interaction features of the target information based on the sub-information features whose sub-information correlation degrees satisfy the information correlation conditions.

[0182] In one embodiment, the indirect information interaction feature determination unit is specifically used to: select target sub-information features whose sub-information correlation degree meets the information correlation condition from each sub-information feature; and perform a summation operation on each target sub-information feature to determine the indirect information interaction feature of the target information.

[0183] In one embodiment, the information feature includes multiple sub-information features. In this embodiment, the information interaction feature determination module 902 includes: a direct information correlation degree determination unit, configured to calculate the first sub-information correlation degree between each sub-information feature and the positive interaction behavior feature, and the second sub-information correlation degree between each sub-information feature and the negative interaction behavior feature; a sub-weight determination unit, configured to determine the sub-weight corresponding to each sub-information feature based on the first and second sub-information correlation degrees; and a direct information interaction feature determination unit, configured to perform a weighted summation of each sub-information feature based on its corresponding sub-weight to determine the direct information interaction feature of the target information.

[0184] In one embodiment, the sub-weight determination unit is specifically used to: determine the positive sub-weight corresponding to each sub-information feature based on the first sub-information correlation degree; determine the negative sub-weight corresponding to each sub-information feature based on the second sub-information correlation degree; and superimpose the positive and negative sub-weights corresponding to the same sub-information feature to obtain the sub-weight corresponding to each sub-information feature.

[0185] In one embodiment, the matching module 903 includes: a comprehensive information interaction feature determination unit, used to perform feature fusion processing on indirect information interaction features and direct information interaction features to obtain comprehensive information interaction features of the target information; and a matching result determination unit, used to perform similarity matching on the comprehensive information interaction features and object interaction features to determine the matching result between the target information and the target object.

[0186] In one embodiment, the matching result determination unit is specifically used to: determine object interaction features based on object features; perform similarity matching on comprehensive information interaction features and object interaction features based on the target neural network model to obtain a matching degree calculation value between the target information and the target object; and determine the matching result between the target information and the target object based on the matching degree calculation value.

[0187] In one embodiment, the information matching device 900 further includes a training module, configured to: determine the loss weight of the sample information based on the number of interaction behaviors generated by the sample object in relation to the sample information; calculate the difference between the matching degree calculation value of each sample object and each sample information and the actual matching degree value of each sample object and each sample information based on the loss weight corresponding to each sample information, thereby obtaining the loss function of the target neural network model; and train the model using a training sample set including the sample information interaction features of the sample information and the sample object interaction features of the sample object, based on the loss function, thereby obtaining the target neural network model used to determine the matching degree calculation value between the target information and the target object.

[0188] In one embodiment, the number of target information items is multiple. In this embodiment, the matching result determination unit includes: a clustering information interaction feature determination component, used to perform clustering processing on the comprehensive information interaction features corresponding to each target information item to obtain clustering information interaction features corresponding to each target information item; a matching degree determination component, used to perform similarity matching between each clustering information interaction feature and object interaction features respectively to determine the matching degree corresponding to each clustering information interaction feature; and a target matching information determination component, used to filter target matching information of target objects from each target information item according to the matching degree corresponding to each clustering information interaction feature.

[0189] In one embodiment, the target matching information determination component is specifically used to: determine each target information corresponding to the clustering information interaction feature as candidate matching information when the matching degree corresponding to the clustering information interaction feature meets the first matching condition; perform similarity matching between the comprehensive information interaction feature and the object interaction feature of each candidate matching information to obtain the matching degree corresponding to each candidate matching information; and determine the candidate matching information as the target matching information of the target object when the matching degree corresponding to the candidate matching information meets the second matching condition.

[0190] Each module in the aforementioned information matching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0191] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10As shown. The computer device includes a processor, memory, input / output interface (I / O), and communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores object parameters, information parameters, and interactive behavior records. The I / O interface allows the processor to exchange information with external devices. The communication interface allows communication with external terminals via a network connection. When executed by the processor, the computer program implements an information matching method. Those skilled in the art will understand that... Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specifically, the computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0192] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0193] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0194] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0195] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one type of non-volatile or volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database or non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0196] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0197] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An information matching method, characterized in that, The method includes: The method involves acquiring information features of target information, object features of the target object, and interaction behavior features of the target object in response to historical information. The information features include sub-information features corresponding to multiple information parameters. The object features include sub-object features corresponding to multiple object parameters. The historical information refers to information pushed to the target object during historical interactions and associated with the type of the target information. The interaction behavior features are obtained based on the interaction behavior information of the target object's interactions with the historical information. The interaction behavior features include direct interaction behavior features and indirect interaction behavior features. The direct interaction behavior features include positive interaction behavior features and negative interaction behavior features. The positive interaction behavior features are obtained based on interaction behavior information of positive feedback operations. The negative interaction behavior features are obtained based on interaction behavior information of negative feedback operations. The indirect interaction behavior features are obtained based on interaction behavior information of operations without a clear preference. Each of the sub-information features is matched with the indirect interaction behavior feature to determine the sub-information correlation degree corresponding to each of the sub-information features. Based on the sub-information features that satisfy the information association conditions, the indirect information interaction features of the target information are determined. Calculate the first sub-information correlation degree between each of the sub-information features and the positive interaction behavior feature, and the second sub-information correlation degree between each of the sub-information features and the negative interaction behavior feature; Based on the correlation degree of the first sub-information and the correlation degree of the second sub-information, the sub-weights corresponding to each sub-information feature are determined; Based on the sub-weights corresponding to each of the sub-information features, the direct information interaction features of the target information are determined. The indirect information interaction features and the direct information interaction features are fused together to obtain the comprehensive information interaction features of the target information. Similarity matching is performed on the comprehensive information interaction features and object interaction features to determine the matching result between the target information and the target object; the object interaction features are determined based on at least one sub-object feature.

2. The method according to claim 1, characterized in that, The determination of the indirect information interaction features of the target information based on the sub-information features that satisfy the information association conditions based on the sub-information association degree includes: From the sub-information features described above, target sub-information features whose sub-information correlation degree meets the information correlation conditions are selected; The indirect information interaction features of the target information are determined by summing the features of each target sub-information feature.

3. The method according to claim 1, characterized in that, The step of determining the sub-weights corresponding to each of the sub-information features based on the first sub-information correlation degree and the second sub-information correlation degree includes: Based on the correlation degree of the first sub-information, determine the positive sub-weight corresponding to each of the sub-information features; Based on the correlation degree of the second sub-information, the negative sub-weights corresponding to each of the sub-information features are determined; By superimposing the positive and negative sub-weights corresponding to the same sub-information feature, the sub-weights corresponding to each sub-information feature are obtained.

4. The method according to claim 1, characterized in that, The step of performing similarity matching on the comprehensive information interaction features and object interaction features to determine the matching result between the target information and the target object includes: Determine object interaction characteristics based on the object characteristics; Based on the target neural network model, similarity matching is performed on the comprehensive information interaction features and the object interaction features to obtain the matching degree calculation value between the target information and the target object; The matching result between the target information and the target object is determined based on the calculated matching degree value.

5. The method according to claim 4, characterized in that, The process of training the target neural network model includes: The loss weight of the sample information is determined based on the number of interaction behaviors generated by the sample object in response to the sample information. Based on the loss weights corresponding to each of the sample information, the difference between the calculated matching degree between each sample object and each sample information and the actual matching degree between each sample object and each sample information is weighted and summed to obtain the loss function of the target neural network model. Based on the loss function, a training sample set including sample information interaction features and sample object interaction features of sample objects is used to train the model, thereby obtaining a target neural network model used to determine the matching degree calculation value between target information and target object.

6. The method according to claim 1, characterized in that, The number of target information items is multiple; The step of performing similarity matching on the comprehensive information interaction features and object interaction features to determine the matching result between the target information and the target object includes: Clustering processing is performed on the comprehensive information interaction features corresponding to each of the target information to obtain the clustered information interaction features corresponding to each of the target information; Each clustering information interaction feature is matched with the object interaction feature for similarity to determine the matching degree of each clustering information interaction feature. Based on the matching degree corresponding to each of the clustering information interaction features, the target matching information of the target object is obtained by filtering from each of the target information.

7. The method according to claim 6, characterized in that, The step of filtering target matching information of the target object from each of the target information based on the matching degree corresponding to each of the clustering information interaction features includes: If the matching degree corresponding to the clustering information interaction feature satisfies the first matching condition, each of the target information corresponding to the clustering information interaction feature is determined as candidate matching information. The comprehensive information interaction features of each candidate matching information are compared with the object interaction features to obtain the matching degree corresponding to each candidate matching information. If the matching degree corresponding to the candidate matching information satisfies the second matching condition, the candidate matching information is determined as the target matching information of the target object.

8. The method according to any one of claims 1 to 7, characterized in that, The process of determining the object interaction features includes: Each of the sub-object features is matched with the positive interaction behavior features to determine the sub-object association degree corresponding to each of the sub-object features. From the features of each sub-object, select the target sub-object features whose sub-object correlation degree meets the object correlation condition; The object interaction features of the target object are determined by summing the features of each target sub-object.

9. An information matching device, characterized in that, The device includes: The acquisition module is used to acquire information features of target information, object features of target objects, and interaction behavior features of the target objects in response to historical information. The information features include sub-information features corresponding to multiple information parameters; the object features include sub-object features corresponding to multiple object parameters; the historical information refers to information pushed to the target object during historical interactions and associated with the type of the target information; the interaction behavior features are obtained based on the interaction behavior information of the target object's interactions with the historical information; the interaction behavior features include direct interaction behavior features and indirect interaction behavior features; the direct interaction behavior features include positive interaction behavior features and negative interaction behavior features; the positive interaction behavior features are obtained based on interaction behavior information of positive feedback operations; the negative interaction behavior features are obtained based on interaction behavior information of negative feedback operations; and the indirect interaction behavior features are obtained based on interaction behavior information of operations without obvious preference. An indirect information correlation degree determination unit is used to match each of the sub-information features with the indirect interaction behavior features to determine the sub-information correlation degree corresponding to each of the sub-information features. An indirect information interaction feature determination unit is used to determine the indirect information interaction features of the target information based on the features of the sub-information whose correlation degree satisfies the information association condition. The direct information correlation degree determination unit is used to calculate the first sub-information correlation degree between each of the sub-information features and the positive interaction behavior feature, and the second sub-information correlation degree between each of the sub-information features and the negative interaction behavior feature; The sub-weight determination unit is used to determine the sub-weight corresponding to each of the sub-information features based on the first sub-information correlation degree and the second sub-information correlation degree. The direct information interaction feature determination unit is used to determine the direct information interaction features of the target information based on the sub-weights corresponding to each of the sub-information features. A comprehensive information interaction feature determination unit is used to perform feature fusion processing on the indirect information interaction features and the direct information interaction features to obtain the comprehensive information interaction features of the target information. The matching result determination unit is used to perform similarity matching on the comprehensive information interaction features and object interaction features to determine the matching result between the target information and the target object; the object interaction features are determined based on at least one sub-object feature.

10. The apparatus according to claim 9, characterized in that, The indirect information interaction feature determination unit is specifically used for: From the sub-information features described above, target sub-information features whose sub-information correlation degree meets the information correlation conditions are selected; The indirect information interaction features of the target information are determined by summing the features of each target sub-information feature.

11. The apparatus according to claim 9, characterized in that, The sub-weight determination unit is specifically used for: Based on the correlation degree of the first sub-information, determine the positive sub-weight corresponding to each of the sub-information features; Based on the correlation degree of the second sub-information, the negative sub-weights corresponding to each of the sub-information features are determined; By superimposing the positive and negative sub-weights corresponding to the same sub-information feature, the sub-weights corresponding to each sub-information feature are obtained.

12. The apparatus according to claim 9, characterized in that, The matching result determination unit is specifically used for: Determine object interaction characteristics based on the object characteristics; Based on the target neural network model, similarity matching is performed on the comprehensive information interaction features and the object interaction features to obtain the matching degree calculation value between the target information and the target object; The matching result between the target information and the target object is determined based on the calculated matching degree value.

13. The apparatus according to claim 12, characterized in that, The device further includes a training module for: The loss weight of the sample information is determined based on the number of interaction behaviors generated by the sample object in response to the sample information. Based on the loss weights corresponding to each of the sample information, the difference between the calculated matching degree between each sample object and each sample information and the actual matching degree between each sample object and each sample information is weighted and summed to obtain the loss function of the target neural network model. Based on the loss function, a training sample set including sample information interaction features and sample object interaction features of sample objects is used to train the model, thereby obtaining a target neural network model used to determine the matching degree calculation value between target information and target object.

14. The apparatus according to claim 9, characterized in that, The number of target information items is multiple; The matching result determination unit includes: A clustering information interaction feature determination component is used to perform clustering processing on the comprehensive information interaction features corresponding to each of the target information to obtain the clustering information interaction features corresponding to each of the target information. The matching degree determination component is used to perform similarity matching between each cluster information interaction feature and the object interaction feature, and to determine the matching degree corresponding to each cluster information interaction feature. The target matching information determination component is used to filter out the target matching information of the target object from each of the target information based on the matching degree corresponding to the interaction features of each of the clustering information.

15. The apparatus according to claim 14, characterized in that, The target matching information determination component is specifically used for: If the matching degree corresponding to the clustering information interaction feature satisfies the first matching condition, each of the target information corresponding to the clustering information interaction feature is determined as candidate matching information. The comprehensive information interaction features of each candidate matching information are compared with the object interaction features to obtain the matching degree corresponding to each candidate matching information. If the matching degree corresponding to the candidate matching information satisfies the second matching condition, the candidate matching information is determined as the target matching information of the target object.

16. The apparatus according to any one of claims 9 to 15, characterized in that, The object features include sub-object features corresponding to multiple object parameters; the process of determining the object interaction features includes: Each of the sub-object features is matched with the positive interaction behavior feature to determine the sub-object association degree corresponding to each sub-object feature; from each of the sub-object features, target sub-object features whose sub-object association degree meets the object association condition are selected; the target sub-object features are summed to determine the object interaction feature of the target object.

17. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

19. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images