Interactive indicator recognition model training, object recommendation method and device

Abstract

This disclosure relates to a method and apparatus for training an interaction indicator recognition model and recommending objects. The method includes acquiring sample indicator association features, sample interaction tags, and historical resource interaction indicators corresponding to multiple historical exposure accounts; configuring labels for multiple preset indicator intervals based on the historical resource interaction indicators to obtain sample interaction indicator labels, whereby the sample interaction indicator labels represent the probability that the historical resource interaction indicator is greater than or equal to the lower limit of multiple preset indicator intervals; and jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, sample interaction indicator labels, and sample interaction labels to obtain a target interaction indicator recognition model. Utilizing embodiments of this disclosure can improve the accuracy of interaction indicator recognition.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a method and apparatus for training an interactive indicator recognition model and recommending objects. Background Technology

[0002] With the development of internet technology, recommending products, applications, stores, live streams, and other objects via the internet has gradually become the main form of object recommendation. Metrics such as GMV (Gross Merchandise Volume) that reflect the revenue of object providers are important indicators to focus on during the object recommendation process.

[0003] In related technologies, deep learning models are often used for indicator identification. During the modeling process, multiple bins (multiple indicator intervals) are often set from a classification perspective. The indicator label is set based on the bin in which the sample data corresponds to the indicator (1 for the bin in which the indicator is located, and 0 otherwise). In the identification process, regression models such as softmax are used to predict the indicator. However, the classification-based modeling approach in the above-mentioned technologies does not recognize the relationship between different categories, making it impossible to identify the size relationship between different indicator intervals. The model predictions are too concentrated, resulting in low indicator identification accuracy, which cannot effectively assist in object recommendation and leads to a poor user experience. Summary of the Invention

[0004] This disclosure provides a method and apparatus for training an interactive indicator recognition model and recommending objects, to at least solve the technical problems in related technologies, such as the inability to identify the magnitude relationship between different indicator ranges, overly concentrated model prediction distribution, low indicator recognition accuracy, ineffective assistance in object recommendation, and poor user experience. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a method for training an interactive indicator recognition model is provided, comprising:

[0006] The system acquires sample indicator association features corresponding to multiple historical exposure accounts, sample interaction tags corresponding to multiple historical exposure accounts, and historical resource interaction indicators corresponding to positive sample accounts among the multiple historical exposure accounts. The historical resource interaction indicators are the virtual resource amount brought by the corresponding object provider for the object acquisition interaction operation triggered by the positive sample account. The sample indicator association features represent the virtual resource consumption and virtual resource acquisition of the corresponding object providers for multiple historical exposure accounts. The sample interaction tags represent the probability of multiple historical exposure accounts triggering object acquisition interaction operations.

[0007] Based on the historical resource interaction indicators, labels are configured for multiple preset indicator intervals to obtain sample interaction indicator labels. The sample interaction indicator labels represent the probability that the historical resource interaction indicator is greater than or equal to the lower limit of multiple preset indicator intervals. The multiple preset indicator intervals are multiple adjacent indicator intervals.

[0008] Based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model.

[0009] In an optional embodiment, the plurality of historical exposure accounts further include negative sample accounts; the step of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels to obtain the target interaction indicator recognition model includes:

[0010] Randomly sample the first indicator association feature corresponding to the negative sample account in the sample indicator association features to obtain the second indicator association feature; the difference between the data volume corresponding to the second indicator association feature and the data volume corresponding to the third indicator association feature is less than a preset threshold, and the third indicator association feature is the indicator association feature corresponding to the positive sample account in the sample indicator association features.

[0011] The third indicator association feature is input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label corresponding to the positive sample account.

[0012] The second indicator association feature and the third indicator association feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label.

[0013] Based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model.

[0014] In an optional embodiment, the method further includes:

[0015] Obtain associated task training features corresponding to multiple historical exposure accounts; the associated task training features are extracted task training features during the process of training at least one task model associated with the interaction index recognition model to be trained.

[0016] The step of inputting the third indicator association feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and obtaining the predicted interaction indicator label corresponding to the positive sample account, includes:

[0017] The first training feature corresponding to the positive sample account in the third indicator association feature and the association task training feature is input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label.

[0018] The step of inputting the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label includes:

[0019] The second indicator association feature, the third indicator association feature, the first training feature, and the second training feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second training feature is the training feature of the target negative sample account in the association task training feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0020] In an optional embodiment, the method further includes:

[0021] Obtain sample identifier features corresponding to multiple historical exposure accounts;

[0022] The step of inputting the third indicator association feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and obtaining the predicted interaction indicator label corresponding to the positive sample account, includes:

[0023] The first identifier feature corresponding to the positive sample account in the third indicator association feature and the sample identifier feature are input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label.

[0024] The step of inputting the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label includes:

[0025] The second indicator association feature, the third indicator association feature, the first identifier feature, and the second identifier feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second identifier feature is the identifier feature of the target negative sample account in the sample identifier feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0026] In an optional embodiment, obtaining the sample identifier features corresponding to the multiple historical exposure accounts includes:

[0027] Obtain sample identification information corresponding to multiple historical exposure accounts;

[0028] The sample identification information is input into the identification feature extraction model to be trained for feature extraction processing to obtain the sample identification features;

[0029] The step of jointly training the interaction recognition model and the interaction recognition model to be trained in the interaction recognition model to be trained, based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, to obtain the target interaction indicator recognition model includes:

[0030] Based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, the target interaction indicator recognition model, the target identifier feature extraction model, the target interaction recognition model, and the target interaction indicator recognition model in the target interaction indicator recognition model are jointly trained to obtain the identifier feature extraction model corresponding to the target interaction indicator recognition model and the target identifier feature extraction model.

[0031] In an optional embodiment, obtaining the sample metric association features corresponding to multiple historical exposure accounts includes:

[0032] Obtain sample indicator association information corresponding to multiple historical exposure accounts. The sample indicator association information includes information on virtual resource consumption corresponding to multiple historical exposure accounts and information on virtual resource acquisition by the object providers corresponding to multiple historical exposure accounts.

[0033] The sample indicator association information is input into the indicator feature extraction model to be trained for feature extraction processing to obtain the sample indicator association features.

[0034] The step of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels to obtain the target interaction indicator recognition model includes:

[0035] Based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels, the indicator feature extraction model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained are jointly trained to obtain the indicator feature extraction model corresponding to the target interaction indicator recognition model and the indicator feature extraction model to be trained.

[0036] In an optional embodiment, the plurality of historical exposure accounts further includes negative sample accounts; the method further includes:

[0037] Obtain associated task training features and / or sample identification features corresponding to multiple historical exposure accounts; the associated task training features are task training features extracted during the training of at least one task model associated with the interaction index recognition model to be trained.

[0038] The sample indicator association features, the association task training features, and / or the sample identification features are input into the feature fusion model to be trained for fusion processing to obtain sample fusion features;

[0039] The step of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels to obtain the target interaction indicator recognition model includes:

[0040] Based on the sample fusion features, the sample interaction index labels, and the sample interaction labels, the feature fusion model to be trained, the interaction recognition model to be trained, and the interaction index recognition model to be trained are jointly trained to obtain the feature fusion model corresponding to the target interaction index recognition model and the feature fusion model to be trained.

[0041] In an optional embodiment, the step of configuring labels for multiple preset indicator ranges based on the historical resource interaction indicators to obtain sample interaction indicator labels includes:

[0042] Determine the target indicator interval where the historical resource interaction indicator is located from multiple preset indicator intervals;

[0043] Based on the first preset label, the first indicator interval is labeled to obtain the first interactive indicator label; the first indicator interval is the interval in which the upper limit value of the multiple preset indicator intervals is less than the target indicator and greater than or equal to the lower limit value of the multiple preset indicator intervals.

[0044] Based on the second preset label, the target indicator interval and the second indicator interval are labeled to obtain the second interactive indicator label; the second indicator interval is the interval among the plurality of preset indicator intervals whose lower limit is greater than the upper limit of the target indicator interval;

[0045] The sample interaction indicator label is generated based on the first interaction indicator label and the interaction indicator label.

[0046] According to a second aspect of the present disclosure, an object recommendation method is provided, comprising:

[0047] Obtain the target indicator association features; the target indicator association features are features that characterize the virtual resource consumption of the target account and the virtual resource acquisition of at least one target object provider, wherein the at least one target object provider is the provider of at least one object to be recommended;

[0048] The target indicator association features are input into the target interaction indicator recognition model obtained based on the interaction indicator recognition model training method provided in the first aspect for interaction indicator recognition processing to obtain target interaction indicator labels. The target interaction indicator labels represent the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals. The predicted resource interaction indicator is the predicted amount of virtual resources brought to at least one target object provider by the target account triggering object acquisition interaction for at least one of the recommended objects. The multiple preset indicator intervals are multiple adjacent indicator intervals.

[0049] The predicted resource interaction index is determined based on the target interaction index label;

[0050] Based on the predicted resource interaction metrics, at least one target object from the list of objects to be recommended is recommended to the target account.

[0051] In an optional embodiment, the method further includes:

[0052] Obtain the target-related task features and / or the target identifier features corresponding to the target account; the target-related task features are the features required to identify the task indicators associated with the predicted resource interaction indicators.

[0053] The target indicator association features, the target association task features, and / or target identification features are input into a feature fusion model for fusion processing to obtain target fusion features; the feature fusion model is jointly trained with the target interaction indicator recognition model.

[0054] The step of inputting the target indicator association features into the target interaction indicator recognition model obtained based on the interaction indicator recognition model training method provided in the first aspect for interaction indicator recognition processing to obtain the target interaction indicator label includes:

[0055] The target fusion features are input into the target interaction indicator recognition model for interaction indicator recognition processing to obtain the target interaction indicator label.

[0056] In an optional embodiment, determining the predicted resource interaction indicator based on the target interaction indicator label includes:

[0057] The mean value of the indicators corresponding to multiple preset indicator intervals and the target indicator difference corresponding to each preset indicator interval are determined. The target indicator difference corresponding to each preset indicator interval is the difference between the probability corresponding to each preset indicator interval and the probability corresponding to the previous preset indicator interval. The previous preset indicator interval is the indicator interval whose upper limit value is adjacent to the lower limit value of each preset indicator interval. The predicted indicator data corresponding to the preset indicator region preceding the first preset indicator interval is zero. The first preset indicator interval is the interval with the smallest upper limit value among multiple preset indicator intervals.

[0058] The predicted resource interaction index is generated based on the difference between the mean of the index and the target index.

[0059] In an optional embodiment, obtaining the target indicator association features includes:

[0060] Obtain the target indicator association information corresponding to the target account. The target indicator association information is information that characterizes the virtual resource consumption of the target account and the virtual resource acquisition of at least one object to be recommended.

[0061] The target indicator association information is input into the indicator feature extraction model for feature extraction processing to obtain the target indicator association features. The indicator feature extraction model is jointly trained with the target interaction indicator recognition model.

[0062] According to a third aspect of the present disclosure, an interactive indicator recognition model training apparatus is provided, comprising:

[0063] The sample data acquisition module is configured to acquire sample indicator association features corresponding to multiple historical exposure accounts, sample interaction tags corresponding to multiple historical exposure accounts, and historical resource interaction indicators corresponding to positive sample accounts among the multiple historical exposure accounts. The historical resource interaction indicators are the amount of virtual resources brought by the corresponding object provider for the object acquisition interaction operation triggered by the positive sample account. The sample indicator association features represent the virtual resource consumption and virtual resource acquisition of the object providers corresponding to the multiple historical exposure accounts. The sample interaction tags represent the probability of the multiple historical exposure accounts triggering object acquisition interaction operations.

[0064] The tag configuration module is configured to perform tag configuration on multiple preset indicator intervals based on the historical resource interaction indicators to obtain sample interaction indicator tags. The sample interaction indicator tags represent the probability that the historical resource interaction indicators are greater than or equal to the lower limit values ​​of multiple preset indicator intervals. The multiple preset indicator intervals are multiple adjacent indicator intervals.

[0065] The joint training module is configured to perform joint training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels and the sample interaction labels, so as to obtain the target interaction indicator recognition model.

[0066] In an optional embodiment, the plurality of historical exposure accounts further includes negative sample accounts; the joint training module includes:

[0067] The random sampling unit is configured to perform random sampling on the first indicator association feature corresponding to the negative sample account in the sample indicator association features to obtain the second indicator association feature; the difference between the data volume corresponding to the second indicator association feature and the data volume corresponding to the third indicator association feature is less than a preset threshold, and the third indicator association feature is the indicator association feature corresponding to the positive sample account in the sample indicator association features.

[0068] The interaction indicator recognition and processing unit is configured to input the third indicator association feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and obtain the predicted interaction indicator label corresponding to the positive sample account.

[0069] The interaction recognition processing unit is configured to perform interaction recognition processing by inputting the second indicator association feature and the third indicator association feature into the interaction recognition model to be trained, and to obtain the predicted interaction label.

[0070] The joint training unit is configured to perform joint training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label and the interaction label corresponding to the predicted interaction label in the sample interaction label, so as to obtain the target interaction indicator recognition model.

[0071] In an optional embodiment, the apparatus further includes:

[0072] The associated task training feature acquisition module is configured to acquire associated task training features corresponding to multiple historical exposure accounts; the associated task training features are the extracted task training features in the process of training at least one task model associated with the interaction index recognition model to be trained.

[0073] The first interaction indicator recognition and processing module is specifically configured to input the first training feature corresponding to the positive sample account in the third indicator association feature and the association task training feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label.

[0074] The interaction recognition processing module is specifically configured to perform interaction recognition processing by inputting the second indicator association feature, the third indicator association feature, the first training feature, and the second training feature into the interaction recognition model to be trained, and obtain the predicted interaction label; the second training feature is the training feature of the target negative sample account in the association task training feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0075] In an optional embodiment, the apparatus further includes:

[0076] The sample identifier feature acquisition module is configured to acquire sample identifier features corresponding to multiple historical exposure accounts;

[0077] The first interactive indicator recognition processing module is specifically configured to input the first identifier feature corresponding to the positive sample account in the third indicator association feature and the sample identifier feature into the interactive indicator recognition model to be trained for interactive indicator recognition processing to obtain the predicted interactive indicator label.

[0078] The interaction recognition processing module is specifically configured to perform interaction recognition processing by inputting the second indicator association feature, the third indicator association feature, the first identifier feature, and the second identifier feature into the interaction recognition model to be trained, and obtain the predicted interaction label; the second identifier feature is the identifier feature of the target negative sample account in the sample identifier feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0079] In an optional embodiment, the sample identifier feature acquisition module includes:

[0080] The sample identification information acquisition unit is configured to acquire sample identification information corresponding to multiple historical exposure accounts;

[0081] The first feature extraction processing unit is configured to perform feature extraction processing by inputting the sample identification information into the identification feature extraction model to be trained, and to obtain the sample identification features.

[0082] The interaction recognition processing module is specifically configured to perform joint training on the target interaction indicator recognition model, the target identifier feature extraction model, the target interaction recognition model, and the target interaction indicator recognition model based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, so as to obtain the identifier feature extraction model corresponding to the target interaction indicator recognition model and the target identifier feature extraction model.

[0083] In an optional embodiment, the sample data acquisition module includes:

[0084] The sample indicator association information acquisition unit is configured to acquire sample indicator association information corresponding to multiple historical exposure accounts, wherein the sample indicator association information includes virtual resource consumption information corresponding to multiple historical exposure accounts and virtual resource acquisition information of object providers corresponding to multiple historical exposure accounts.

[0085] The second feature extraction processing unit is configured to input the sample indicator association information into the indicator feature extraction model to be trained for feature extraction processing to obtain the sample indicator association features.

[0086] The joint training module is specifically configured to perform joint training on the indicator feature extraction model, the interaction recognition model, and the interaction indicator recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels, so as to obtain the indicator feature extraction model corresponding to the target interaction indicator recognition model and the indicator feature extraction model to be trained.

[0087] In an optional embodiment, the plurality of historical exposure accounts further includes negative sample accounts; the device further includes:

[0088] The feature acquisition module is configured to acquire associated task training features and / or sample identification features corresponding to multiple historical exposure accounts; the associated task training features are task training features extracted during the training of at least one task model associated with the interaction index recognition model to be trained.

[0089] The first fusion processing module is configured to perform fusion processing by inputting the sample index association features, the association task training features and / or the sample identification features into the feature fusion model to be trained, and obtain sample fusion features;

[0090] The joint training module is specifically configured to perform joint training on the feature fusion model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained based on the sample fusion features, the sample interaction indicator labels, and the sample interaction labels, so as to obtain the feature fusion model corresponding to the target interaction indicator recognition model and the feature fusion model to be trained.

[0091] In an optional embodiment, the label configuration module includes:

[0092] The target indicator interval determination unit is configured to determine the target indicator interval where the historical resource interaction indicator is located from a plurality of preset indicator intervals;

[0093] The first label configuration unit is configured to perform label configuration on the first indicator interval based on the first preset label to obtain the first interactive indicator label; the first indicator interval is the interval in which the upper limit value of the multiple preset indicator intervals is less than the target indicator and greater than or equal to the lower limit value of the multiple preset indicator intervals.

[0094] The second tag configuration unit is configured to perform tag configuration on the target indicator range and the second indicator range based on the second preset tag to obtain the second interactive indicator tag; the second indicator range is the range among the plurality of preset indicator ranges whose lower limit value is greater than the upper limit value of the target indicator range;

[0095] The sample interaction indicator label generation unit is configured to generate the sample interaction indicator label based on the first interaction indicator label and the interaction indicator label.

[0096] According to a fourth aspect of the present disclosure, an object recommendation apparatus is provided, comprising:

[0097] The target indicator association feature acquisition module is configured to acquire target indicator association features; the target indicator association features are features that characterize the virtual resource consumption of the target account and the virtual resource acquisition of at least one target object provider, wherein the at least one target object provider is the provider of at least one object to be recommended;

[0098] The second interaction indicator recognition and processing module is configured to perform interaction indicator recognition processing by inputting the target indicator association features into a target interaction indicator recognition model obtained based on the interaction indicator recognition model training method provided in the first aspect, and obtaining a target interaction indicator label. The target interaction indicator label represents the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals. The predicted resource interaction indicator is the predicted amount of virtual resources brought to at least one target object provider by the target account triggering an object acquisition interaction for at least one of the recommended objects. The multiple preset indicator intervals are multiple adjacent indicator intervals.

[0099] The predictive resource interaction index determination module is configured to determine the predictive resource interaction index based on the target interaction index label.

[0100] The object push module is configured to recommend at least one target object from the objects to be recommended to the target account based on the predicted resource interaction metrics.

[0101] In an optional embodiment, the apparatus further includes:

[0102] The feature acquisition module is configured to acquire the target-related task features and / or the target identifier features corresponding to the target account; the target-related task features are the features required to identify the task indicators associated with the predicted resource interaction indicators;

[0103] The second fusion processing module is configured to perform fusion processing by inputting the target indicator association features, the target association task features, and / or target identification features into a feature fusion model to obtain target fusion features; the feature fusion model is jointly trained with the target interaction indicator recognition model.

[0104] The second interaction indicator recognition processing module is specifically configured to input the target fusion features into the target interaction indicator recognition model to perform interaction indicator recognition processing, and obtain the target interaction indicator label.

[0105] In an optional embodiment, the predictive resource interaction index determination module includes:

[0106] The calculation unit is configured to determine the mean value of indicators corresponding to multiple preset indicator intervals and the target indicator difference corresponding to each preset indicator interval. The target indicator difference corresponding to each preset indicator interval is the difference between the probability corresponding to each preset indicator interval and the probability corresponding to the previous preset indicator interval. The previous preset indicator interval is an indicator interval whose upper limit value is adjacent to the lower limit value of each preset indicator interval. The predicted indicator data corresponding to the previous preset indicator region of the first preset indicator interval is zero. The first preset indicator interval is the interval with the smallest upper limit value among the multiple preset indicator intervals.

[0107] The predictive resource interaction index generation unit is configured to generate the predictive resource interaction index based on the difference between the index mean and the target index.

[0108] In an optional embodiment, the target indicator association feature acquisition module includes:

[0109] The target indicator association information acquisition unit is configured to acquire the target indicator association information corresponding to the target account, wherein the target indicator association information is information representing the virtual resource consumption status of the target account and the virtual resource acquisition status of at least one object to be recommended.

[0110] The third feature extraction processing unit is configured to perform feature extraction processing by inputting the target indicator association information into the indicator feature extraction model to obtain the target indicator association features. The indicator feature extraction model is jointly trained with the target interaction indicator recognition model.

[0111] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method as described in any one of the first or second aspects above.

[0112] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the method described in any one of the first or second aspects of the present disclosure.

[0113] According to a fifth aspect of the present disclosure, a computer program product comprising instructions is provided that, when run on a computer, causes the computer to perform the method as described in any one of the first or second aspects above.

[0114] The technical solutions provided by the embodiments of this disclosure bring at least the following beneficial effects:

[0115] During the training of the interaction indicator recognition model, historical resource interaction indicators, reflecting the virtual resource volume brought by positive sample accounts from multiple historical exposure accounts, are combined with the historical resource interaction indicators for the corresponding object providers. Labels are configured for multiple adjacent preset indicator intervals to generate sample interaction indicator labels. This avoids the introduction of zero-value indicators during the resource interaction indicator learning process, which could lead to the underestimation of resource interaction indicators. Furthermore, these sample interaction indicator labels can characterize the probability that the historical resource interaction indicator is greater than or equal to the lower limit of multiple preset indicator intervals. This ensures that the sample interaction indicator labels, while indicating the indicator interval where the historical resource interaction indicator for the positive sample account falls, also indicate the probability of the historical resource interaction indicator falling within multiple preset indicator intervals. The size relationship ensures the balanced distribution of learned resource interaction indicators. Furthermore, by combining sample indicator association features representing the virtual resource consumption of multiple historically exposed objects and the virtual resource acquisition of corresponding object providers, along with sample interaction labels, joint training of the interaction indicator recognition model and the training interaction indicator recognition model allows the training interaction indicator recognition model to comprehensively learn from positive and negative samples from the dimension of whether object acquisition interactions are triggered during the learning process of resource interaction indicators. This avoids overestimation of predicted resource interaction indicators, effectively improving the indicator recognition accuracy of the trained target interaction indicator recognition model, effectively assisting object recommendation, and greatly enhancing the user experience.

[0116] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0117] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0118] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment;

[0119] Figure 2 This is a flowchart illustrating an interactive indicator recognition model training method according to an exemplary embodiment;

[0120] Figure 3 This is a flowchart illustrating a method for configuring labels for multiple preset indicator ranges based on historical resource interaction indicators to obtain sample interaction indicator labels according to an exemplary embodiment.

[0121] Figure 4 This is a flowchart illustrating an exemplary embodiment of a method for jointly training an interaction indicator recognition model to be trained and an interaction recognition model to be trained, based on sample indicator association features, sample interaction indicator labels, and sample interaction labels, to obtain a target interaction indicator recognition model.

[0122] Figure 5 This is a schematic diagram illustrating the training process of an interactive indicator recognition model based on an example.

[0123] Figure 6 This is a flowchart illustrating an object recommendation method according to an exemplary embodiment;

[0124] Figure 7 This is a block diagram of an interactive indicator recognition model training device according to an exemplary embodiment;

[0125] Figure 8 This is a block diagram illustrating an object recommendation device according to an exemplary embodiment;

[0126] Figure 9 This is a block diagram illustrating an electronic device for training an interactive indicator recognition model according to an exemplary embodiment;

[0127] Figure 10 This is a block diagram illustrating an electronic device for object recommendation according to an exemplary embodiment. Detailed Implementation

[0128] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0129] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0130] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0131] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment, which may include a terminal 100 and a server 200.

[0132] In an optional embodiment, terminal 100 can be used to provide object recommendation services to any user. Specifically, terminal 100 can be, but is not limited to, electronic devices such as smartphones, desktop computers, tablets, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices, or software running on the aforementioned electronic devices, such as applications. Optionally, the operating system running on the electronic device can be, but is not limited to, Android, iOS, Linux, Windows, etc.

[0133] In an optional embodiment, server 200 can be used to train an interaction indicator recognition model and provide backend services to terminal 100 based on the trained interaction indicator recognition model. Specifically, server 200 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services.

[0134] In addition, it should be noted that, Figure 1 The example shown is merely one application environment provided by this disclosure. In practical applications, other application environments may also be included, such as more terminals.

[0135] In the embodiments described in this specification, the terminal 100 and the server 200 can be directly or indirectly connected through wired or wireless communication, and this disclosure does not impose any restrictions.

[0136] Figure 2 This is a flowchart illustrating an interactive indicator recognition model training method according to an exemplary embodiment. Optionally, the above-described interactive indicator recognition model training method can be applied to electronic devices such as servers or terminals. Figure 2 As shown, the above-mentioned interaction indicator recognition model training method may include the following steps:

[0137] In step S201, sample indicator association features corresponding to multiple historical exposure accounts, sample interaction tags corresponding to multiple historical exposure accounts, and historical resource interaction indicators corresponding to positive sample accounts among multiple historical exposure accounts are obtained.

[0138] In one specific embodiment, multiple historical exposure accounts can be exposure accounts corresponding to recommended objects in the object recommendation platform within a preset historical time period. Optionally, the recommended objects can be products, stores, live streams, etc., and the object can be a user account. Optionally, when pushing multimedia data corresponding to a certain recommended object to any user account, the user account can be a historical exposure account corresponding to that recommended object. Correspondingly, the recommended object can be a historical recommended object corresponding to that user account, and the provider of the historical recommended object is the object provider corresponding to that user account. Specifically, the multimedia data corresponding to the recommended object can be data used to introduce the recommended object, such as the details page content corresponding to a product, the homepage content corresponding to a store, the live stream corresponding to a live stream, etc.

[0139] In a specific embodiment, the aforementioned sample indicator association features can characterize the virtual resource consumption of multiple historical exposure accounts and the virtual resource acquisition of the object providers corresponding to multiple historical exposure accounts. Specifically, the aforementioned sample indicator association features can be obtained by extracting features from the sample indicator association information based on a corresponding indicator feature extraction model. Optionally, the indicator feature extraction model can be pre-trained independently or trained together with the interaction indicator recognition model to be trained (joint training). Optionally, taking the indicator feature extraction model trained together with the interaction indicator recognition model to be trained as an example to obtain sample indicator association features, the above-mentioned acquisition of sample indicator association features corresponding to multiple historical exposure accounts can include:

[0140] Obtain the sample metric correlation information corresponding to multiple historical exposure accounts;

[0141] The sample indicator association information is input into the indicator feature extraction model to be trained for feature extraction processing to obtain the sample indicator association features.

[0142] In a specific embodiment, the aforementioned sample indicator association information can be information on the virtual resource consumption of multiple historical exposure accounts and information on the virtual resource acquisition of multiple historical exposure accounts' corresponding object providers. In practical applications, each recommendation corresponds to one sample (information on the virtual resource consumption of a historical exposure account and information on the virtual resource acquisition of the object provider corresponding to that historical exposure account); optionally, the sample indicator association information can include the cumulative virtual resource consumption information of historical exposure accounts within at least one first preset historical time period (e.g., the virtual resource consumption in the last 7 days, 30 days, and 90 days), the single virtual resource consumption information of historical exposure accounts within at least one first preset historical time period, the average virtual resources obtained by the object providers of historical recommended objects based on the recommended objects within a second preset historical time period (e.g., the average amount of virtual resources obtained in the live broadcast room based on the recommended objects in the last 30 days, e.g., the average amount of virtual resources corresponding to a single recommended object in the live broadcast room in the last 30 days, etc.), and the cumulative virtual resources obtained by the object providers of historical recommended objects based on the recommended objects within a second preset historical time period, etc.

[0143] In a specific embodiment, the feature extraction model to be trained can be a feature extraction network to be trained, such as a convolutional neural network, and the network structure and number of layers can be set according to the actual application requirements.

[0144] In one specific embodiment, the aforementioned historical resource interaction metrics can represent the amount of virtual resources brought to the corresponding object provider by the interaction operation triggered by the positive sample account; the sample interaction tag represents the probability of multiple historical exposure accounts triggering the interaction operation of the object; specifically, the positive sample account can include multiple objects. Specifically, the multiple historical exposure objects also include negative sample accounts, and negative sample accounts can include multiple objects.

[0145] In practical applications, when a recommended object is exposed to a user account, if that user account triggers an interaction to retrieve the recommended object (object retrieval interaction operation), a historical resource interaction metric will be generated. Accordingly, the user account can be a positive sample account, and the sample interaction label corresponding to the positive sample account is 1. Conversely, if the user account does not trigger an interaction to retrieve the recommended object, no corresponding historical resource interaction metric will be generated (i.e., the historical resource interaction metric is 0). Accordingly, the user account can be a negative sample account, and the sample interaction label corresponding to the negative sample account is 0.

[0146] In step S203, based on historical resource interaction indicators, labels are configured for multiple preset indicator ranges to obtain sample interaction indicator labels.

[0147] In one specific embodiment, the aforementioned sample interaction indicator label can represent the probability that the historical resource interaction indicator corresponding to a positive sample account is greater than or equal to the lower limit of multiple preset indicator intervals. Optionally, the sample interaction indicator label can be a numerical value or a characterized representation of the probability that the historical resource interaction indicator corresponding to a positive sample account is greater than or equal to the lower limit of multiple preset indicator intervals. The multiple preset indicator intervals are multiple adjacent indicator intervals. Specifically, since multiple preset indicator intervals are adjacent, and when a historical resource interaction indicator is located within a certain preset indicator interval (target indicator interval), the probability that the historical resource interaction indicator is greater than or equal to the lower limit of the target indicator interval is 1. The probability that the historical resource interaction indicator is greater than or equal to the lower limit of the indicator interval after the target indicator interval (the indicator interval whose lower limit is greater than the upper limit of the target indicator interval) is also 1. Furthermore, the probability that the historical resource interaction indicator is greater than or equal to the lower limit of the indicator interval before the target indicator interval (the indicator interval whose upper limit is less than the lower limit of the target indicator interval) is 0. Accordingly, it can be determined that when the sample interaction indicator label represents the probability that the historical resource interaction indicator corresponding to the positive sample account is greater than or equal to the lower limit of multiple preset indicator intervals, the sample interaction indicator label can indicate the size relationship of multiple preset indicator intervals based on indicating the interval to which the historical resource interaction indicator belongs.

[0148] In an optional embodiment, multiple preset index ranges can be set in combination with the distribution range of resource interaction indicators in actual applications. For example, multiple preset index ranges include (0, 50], (50, 100], (100, 150], (150, 200], (200, 500], (500, 1000] and (1000, 10000], etc.

[0149] In an optional embodiment, such as Figure 3 As shown, based on historical resource interaction metrics, the above-mentioned label configuration for multiple preset indicator ranges yields sample interaction indicator labels that can include:

[0150] In step S301, the target indicator interval where the historical resource interaction indicator is located is determined from multiple preset indicator intervals;

[0151] In step S303, based on the first preset label, the first indicator range is labeled to obtain the first interactive indicator label;

[0152] In step S305, based on the second preset label, the target indicator interval and the second indicator interval are labeled to obtain the second interactive indicator label;

[0153] In step S307, sample interaction indicator labels are generated based on the first interaction indicator label and the interaction indicator label.

[0154] In a specific embodiment, taking the sample interaction index label as an example, the first preset label is 0 and the second preset label is 1; the above-mentioned first index interval is the interval in which the upper limit value of the multiple preset index intervals is less than the target index and greater than or equal to the lower limit value of the multiple preset index intervals; the above-mentioned second index interval is the interval in which the lower limit value of the multiple preset index intervals is greater than the upper limit value of the target index interval.

[0155] In a specific embodiment, it is assumed that multiple preset indicator intervals can be set in combination with the resource benefit distribution range in actual applications. For example, multiple preset indicator intervals include (0, 10], (10, 20], (20, 30], (30, 50], (50, 100], and (100, 300). Optionally, it is assumed that the historical resource interaction indicator corresponding to a certain object in the positive sample account is 45; correspondingly, the target indicator interval is (30, 50], the first indicator interval includes (0, 10], (10, 20], and (20, 30], and the second indicator interval includes (50, 100] and (100, 300]. Optionally, the labels can be configured sequentially from large to small based on the numerical ranges corresponding to the multiple preset indicator intervals; correspondingly, the first interaction indicator label is 000, the second interaction indicator label is 111, and the sample interaction indicator label is 000111.

[0156] In the above embodiments, the target indicator interval and the second indicator interval whose lower limit is greater than the upper limit of the target indicator interval are configured by the second preset label; and the first indicator region whose upper limit is less than the lower limit of the target indicator interval is configured by the first preset label. This allows the obtained sample interaction indicator label to indicate the size relationship of multiple preset indicator intervals on the basis of indicating the indicator interval where the historical resource interaction indicator of the positive sample account is located, thereby ensuring the balanced distribution of resource interaction indicators learned by the subsequent model.

[0157] In step S205, based on the sample indicator association features, sample interaction indicator labels, and sample interaction labels, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model.

[0158] In a specific embodiment, the training data for the interaction indicator recognition model to be trained can be the third sample indicator association feature corresponding to the positive sample account in the sample indicator association feature and the sample interaction indicator label.

[0159] In one specific embodiment, the interaction indicator recognition model to be trained can be a deep learning network to be trained, and the network structure and number of layers can be set according to the actual application requirements.

[0160] In an optional embodiment, the joint training of the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on sample indicator association features, sample interaction indicator labels, and sample interaction labels to obtain the target interaction indicator recognition model may include:

[0161] The third indicator association feature is input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label corresponding to the positive sample account.

[0162] Input the sample index association features into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label;

[0163] Based on the predicted interaction indicator labels, sample interaction indicator labels, predicted interaction labels, and sample interaction labels, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model.

[0164] In a specific embodiment, the predicted interaction indicator label can be the interaction indicator recognition model to be trained, combined with the third indicator association feature to perform interaction indicator recognition processing, and the predicted interaction indicator label. Specifically, the predicted interaction indicator label represents the probability that the historical resource interaction indicator corresponding to the positive sample account predicted by the interaction indicator recognition model to be trained is greater than or equal to the lower limit of multiple preset indicator intervals.

[0165] In a specific embodiment, the predicted interaction label can be the interaction recognition model to be trained, which combines the sample index association features to perform interaction recognition processing, and the predicted interaction label can specifically represent the probability of multiple historical exposure accounts triggering objects to obtain interaction operations, as predicted by the interaction recognition model to be trained.

[0166] In a specific embodiment, the above-mentioned joint training of the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the sample interaction label to obtain the target interaction indicator recognition model may include: determining first loss information based on the predicted interaction indicator label and the sample interaction indicator label; determining second loss information based on the predicted interaction label and the sample interaction label; determining first target loss information based on the first loss information and the second loss information; and performing backpropagation training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the first target loss information to obtain the target interaction indicator recognition model.

[0167] In one specific embodiment, the aforementioned first loss information can characterize the accuracy of the interaction indicator recognition model under training in recognizing interaction indicator labels. Optionally, the predicted interaction indicator labels and sample interaction indicator labels can be substituted into the corresponding preset loss function to obtain the aforementioned first loss information. The aforementioned second loss information can characterize the accuracy of the interaction recognition model under training in recognizing interaction labels. Optionally, the loss functions corresponding to the first loss information and the second loss information can be set according to actual application requirements. Optionally, the predicted interaction labels and sample interaction labels can be substituted into the corresponding preset loss function to obtain the aforementioned second loss information. Optionally, the first loss information and the second loss information can be added together to obtain the first target loss information, or the first loss information and the second loss information can be weighted and summed to obtain the aforementioned first target loss information. Specifically, the weights corresponding to the first loss information and the second loss information can be set according to actual application requirements.

[0168] In a specific embodiment, backpropagation training is performed on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the first target loss information to obtain the target interaction indicator recognition model. This may include: updating the model parameters in the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the first target loss information (specifically, gradient descent may be used to update the model parameters), and repeating the above interaction indicator recognition processing to update the model parameters based on the updated interaction indicator recognition model and the interaction recognition model to be trained, until a preset convergence condition is met, and the interaction indicator recognition model to be trained when the preset convergence condition is met is taken as the target interaction indicator recognition model.

[0169] In a specific embodiment, the aforementioned satisfying of the preset convergence condition can be that the first target loss information is less than or equal to a preset loss threshold, or that the number of training iterations reaches a preset number, etc. Specifically, the preset loss threshold and the preset number can be set in combination with the model accuracy and training speed requirements in actual applications.

[0170] In practical applications, the aforementioned sample indicator association features are the indicator association features corresponding to multiple historical exposure objects. That is, the sample indicator association features are sample data from the exposure space. However, the interaction indicator recognition model to be trained is used for resource interaction indicator recognition, meaning it needs to use sample data from the exposure space to identify indicators in the interaction (transaction) space. Since the proportion of user accounts that trigger interaction operations in the exposure space is often small, there is a sample selection bias, which leads to the underestimation of indicators in the interaction space and inaccurate indicator recognition. If modeling is directly performed in the interaction space (i.e., directly using the indicator association features corresponding to user accounts that trigger interaction operations as sample data), there is a problem of inaccurate indicator recognition due to spatial inconsistency, since subsequent online predictions need to be made in the exposure space. Accordingly, in another optional embodiment, to solve the indicator recognition accuracy problem caused by spatial inconsistency, such as... Figure 4 As shown, the above-mentioned joint training of the interaction indicator recognition model to be trained and the interaction recognition model to be trained, based on sample indicator association features, sample interaction indicator labels, and sample interaction labels, to obtain the target interaction indicator recognition model may include the following steps:

[0171] In step S401, the first indicator association feature corresponding to the negative sample account in the sample indicator association feature is randomly sampled to obtain the second indicator association feature;

[0172] In step S403, the third indicator association features are input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label corresponding to the positive sample account.

[0173] In step S405, the second indicator association features and the third indicator association features are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label.

[0174] In step S407, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label to obtain the target interaction indicator recognition model.

[0175] In a specific embodiment, the difference between the amount of data corresponding to the second indicator association feature and the amount of data corresponding to the third indicator association feature is less than a preset threshold, and the third indicator association feature is the indicator association feature corresponding to the positive sample account in the sample indicator association features.

[0176] In a specific embodiment, the interaction label corresponding to the predicted interaction label in the above sample interaction label can be the interaction label corresponding to the positive sample account and the target negative sample account in the sample interaction label; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0177] In a specific embodiment, the joint training of the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label to obtain the target interaction indicator recognition model may include: determining third loss information based on the predicted interaction indicator label and the sample interaction indicator label; determining fourth loss information based on the predicted interaction label and the interaction label corresponding to the predicted interaction label in the sample interaction label; determining second target loss information based on the third loss information and the fourth loss information; and performing backpropagation training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the second target loss information to obtain the target interaction indicator recognition model.

[0178] In a specific embodiment, the detailed steps of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label to obtain the target interaction indicator recognition model can be found in the relevant detailed steps of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the sample interaction label to obtain the target interaction indicator recognition model. These details will not be repeated here.

[0179] In the above embodiments, by randomly sampling the first indicator association features corresponding to negative sample accounts in the sample indicator association features, the difference in data volume between the indicator association features corresponding to positive sample accounts and those corresponding to negative sample accounts participating in the interaction recognition process can be minimized. This effectively caches the sample selection bias and the underestimation of indicators in the interaction space caused by the inconsistency between the exposure space where the sample data is located during the training phase and the interaction space where the model prediction phase is located. Furthermore, by combining the third indicator association features corresponding to positive sample accounts and the sample interaction indicator labels, while modeling in the interaction space, the third indicator association features in the exposure space and the second indicator association features obtained from negative sampling of the exposure space are used to train the interaction recognition model to be trained. This can better alleviate the problem of space inconsistency, thereby effectively improving the indicator recognition accuracy of the trained target interaction indicator recognition model, effectively assisting in object recommendation, and greatly improving the user experience.

[0180] In practical applications, each recommendation records the corresponding behavioral path (i.e., post-recommendation operation information), which can then be used for multi-objective task learning (such as click-through rate, conversion rate, browsing time, etc.). Optionally, the task training features extracted by the corresponding task models during these multi-objective task learning processes can be used to enrich the sample data in the resource interaction indicator recognition process. Accordingly, the above method also includes:

[0181] Obtain training features for associated tasks corresponding to multiple historical exposure accounts;

[0182] Accordingly, the above-mentioned input of the third indicator association features into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and the resulting predicted interaction indicator labels for positive sample accounts may include:

[0183] Input the first training feature corresponding to the positive sample account in the third indicator association feature and the association task training feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label.

[0184] Accordingly, the above-mentioned input of the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing, and the resulting predicted interaction labels may include:

[0185] The second indicator association features, the third indicator association features, the first training features, and the second training features are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second training feature mentioned above is the training feature of the target negative sample account in the association task training features; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0186] In a specific embodiment, the aforementioned associated task training features can be extracted task training features during the process of training at least one task model associated with the interaction indicator recognition model to be trained; optionally, the at least one task model can include at least one of a click-through rate recognition model, a conversion rate recognition model, and a browsing duration recognition model. Specifically, the associated task training features can be extracted from associated task training information. Specifically, the associated task training information can be the training data required to train at least one task model. Optionally, the associated task training information can include account feature information (e.g., the gender of the user corresponding to the account, historical interaction operations, etc.), sample operation information of the multiple historical exposure accounts for their respective corresponding historical recommendation objects, object feature information corresponding to the sample recommendation objects, and interaction features of the sample recommendation objects within a preset time period. The preset time period includes the time point when the sample recommendation objects are recommended to the corresponding historical exposure accounts; the sample recommendation objects are the historical recommendation objects corresponding to the multiple historical exposure accounts.

[0187] In an optional embodiment, the associated task training features and sample indicator associated features can be fused together, and the features corresponding to positive sample accounts can be selected from the fused features and input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain predicted interaction indicator labels; and the features corresponding to positive sample accounts and the features corresponding to target negative sample accounts can be selected from the fused features and input into the interaction recognition model to be trained for interaction recognition processing to obtain predicted interaction labels.

[0188] In the above embodiments, during the interaction indicator recognition process, the first training feature corresponding to the positive sample account in the associated task training features is incorporated, and during the interaction recognition process, the task training features corresponding to the positive sample account and the target negative sample account in the associated task training features are incorporated. This can better enrich the sample data, thereby enriching the resource interaction indicator features learned during model training and improving the indicator recognition accuracy of the trained interaction indicator recognition model.

[0189] In an optional embodiment, to enhance the model's memorability, sample identifier features corresponding to historical exposure accounts can be introduced during model training. Accordingly, the above method further includes:

[0190] Obtain sample identifier features corresponding to multiple historical exposure accounts;

[0191] Accordingly, the above-mentioned input of the third indicator association features into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and the resulting predicted interaction indicator labels for positive sample accounts may include:

[0192] Input the first identifier feature corresponding to the positive sample account from the third indicator association feature and the sample identifier feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label.

[0193] Accordingly, the above-mentioned input of the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing, and the resulting predicted interaction labels may include:

[0194] The second indicator association feature, the third indicator association feature, the first identifier feature, and the second identifier feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second identifier feature is the identifier feature of the target negative sample account in the sample identifier feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0195] In one specific embodiment, the sample identification features include the account identification features of multiple historical exposure accounts, the object identification features of the historical recommended objects corresponding to the multiple historical exposure accounts, and the account identification features of the publishing accounts corresponding to the historical recommended objects.

[0196] In an optional embodiment, sample identifier features and sample indicator association features can be fused together, and features corresponding to positive sample accounts can be selected from the fused features and input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain predicted interaction indicator labels; and features corresponding to positive sample accounts and target negative sample accounts can be selected from the fused features and input into the interaction recognition model to be trained for interaction recognition processing to obtain predicted interaction labels.

[0197] In the above embodiments, during the interactive indicator recognition process, the first identifier feature corresponding to the positive sample account in the sample identifier features is incorporated, and during the interactive recognition process, the identifier features corresponding to the positive sample account and the target negative sample account in the sample identifier features are incorporated, which can enhance the model's memory and thus improve the indicator recognition accuracy of the trained interactive indicator recognition model.

[0198] In an optional embodiment, the aforementioned sample identification features can be obtained by extracting features from sample identification information based on a corresponding identification feature extraction model. Optionally, the identification feature extraction model can be pre-trained independently or trained together with the interaction indicator recognition model to be trained (joint training). Optionally, taking the acquisition of sample identification features by combining the identification feature extraction model trained together with the interaction indicator recognition model to be trained as an example, the acquisition of sample identification features corresponding to multiple historical exposure accounts includes:

[0199] Obtain sample identification information corresponding to multiple historical exposure accounts;

[0200] The sample identification information is input into the identification feature extraction model to be trained for feature extraction processing to obtain the sample identification features;

[0201] Accordingly, based on the predicted interaction indicator labels, sample interaction indicator labels, predicted interaction labels, and the interaction labels corresponding to the predicted interaction labels in the sample interaction labels, the joint training of the interaction recognition model to be trained and the interaction indicator recognition model to be trained in the interaction indicator recognition model to be trained can result in the following target interaction indicator recognition model:

[0202] Based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, the untrained identifier feature extraction model, the untrained interaction recognition model, and the untrained interaction indicator recognition model in the untrained interaction indicator recognition model are jointly trained to obtain the identifier feature extraction model corresponding to the target interaction indicator recognition model and the untrained identifier feature extraction model.

[0203] In one specific embodiment, the sample identification information includes account identification information of multiple historical exposure accounts, object identification information of historical recommended objects corresponding to multiple historical exposure accounts, and account identification information of publishing accounts corresponding to historical recommended objects.

[0204] In a specific embodiment, referring to the aforementioned steps, a second target loss information is determined based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label. Then, backpropagation training is performed on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the second target loss information, resulting in the target interaction indicator recognition model. This is replaced by backpropagation training on the identifier feature extraction model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained based on the second target loss information, resulting in the identifier feature extraction model to be trained (the trained identifier feature extraction model) and the target interaction indicator recognition model. Correspondingly, during the training process, the sample identifier features are continuously updated as the identifier feature extraction model to be trained is updated.

[0205] In a specific embodiment, the above-mentioned backpropagation training of the target identifier feature extraction model, the target interaction recognition model, and the target interaction index recognition model based on the second target loss information is used to obtain the specific refinement of the identifier feature extraction model and the target interaction index recognition model corresponding to the target identifier feature extraction model. For details, please refer to the above-mentioned backpropagation training of the target interaction index recognition model and the target interaction recognition model based on the second target loss information to obtain the specific refinement of the target interaction index recognition model, which will not be repeated here.

[0206] In the above embodiments, training the model used to extract sample identifier features together with the interaction indicator recognition model to be trained can greatly improve the accuracy and effectiveness of the extracted sample identifier features, thereby enabling better recognition of interaction indicators and improving the accuracy of interaction indicator recognition of the trained interaction indicator recognition model.

[0207] In an optional embodiment, the above-mentioned acquisition of sample metric association features corresponding to multiple historical exposure accounts includes:

[0208] Obtain the sample indicator association information corresponding to multiple historical exposure accounts. The sample indicator association information includes the virtual resource consumption information corresponding to multiple historical exposure accounts and the virtual resource acquisition information of the object providers corresponding to multiple historical exposure accounts.

[0209] The sample indicator association information is input into the indicator feature extraction model to be trained for feature extraction processing to obtain the sample indicator association features.

[0210] Accordingly, the target interaction indicator recognition model can be obtained by jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained, based on the sample indicator association features, sample interaction indicator labels, and sample interaction labels.

[0211] Based on sample indicator association features, sample interaction indicator labels, and sample interaction labels, the indicator feature extraction model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained are jointly trained to obtain the indicator feature extraction model corresponding to the target interaction indicator recognition model and the indicator feature extraction model to be trained.

[0212] In a specific embodiment, the above-mentioned joint training of the indicator feature extraction model, the interaction recognition model, and the interaction indicator recognition model based on sample indicator association features, sample interaction indicator labels, and sample interaction labels to obtain the indicator feature extraction model corresponding to the target interaction indicator recognition model and the indicator feature extraction model can be referred to in the above-mentioned joint training of the interaction indicator recognition model and the interaction recognition model based on sample indicator association features, sample interaction indicator labels, and sample interaction labels to obtain the target interaction indicator recognition model. It will not be repeated here. Specifically, in the backpropagation training process, backpropagation training of the indicator feature extraction model to be trained can be added.

[0213] In one specific embodiment, when the model used to extract the associated features of sample indicators is trained together with the interactive indicator recognition model to be trained, the aforementioned associated features of sample indicators can be continuously updated as the indicator feature extraction model to be trained is updated.

[0214] In the above embodiments, training the model used to extract the association features of sample indicators together with the interaction indicator recognition model to be trained can greatly improve the accuracy and effectiveness of the extracted association features of sample indicators, thereby enabling better recognition of interaction indicators and improving the accuracy of interaction indicator recognition of the trained interaction indicator recognition model.

[0215] In an optional embodiment, associated task training features corresponding to multiple historical exposure accounts and sample identification features corresponding to multiple historical exposure accounts can be introduced during model training. Accordingly, the above method may further include:

[0216] Obtain the associated task training features and / or sample identification features corresponding to multiple historical exposure accounts; the associated task training features are the task training features extracted during the training of at least one task model associated with the interaction index recognition model to be trained.

[0217] Input the sample indicator association features, association task training features and / or sample identification features into the feature fusion model to be trained for fusion processing to obtain sample fusion features;

[0218] Accordingly, based on the sample indicator association features, sample interaction indicator labels, and sample interaction labels, the above-mentioned joint training of the interaction indicator recognition model to be trained and the interaction recognition model to be trained yields the target interaction indicator recognition model, including:

[0219] Based on sample fusion features, sample interaction index labels, and sample interaction labels, the feature fusion model to be trained, the interaction recognition model to be trained, and the interaction index recognition model to be trained are jointly trained to obtain the feature fusion model corresponding to the target interaction index recognition model and the feature fusion model to be trained.

[0220] In one specific embodiment, the aforementioned feature fusion model to be trained may include a stitching module, an MLP (Multi-Layer Perception) and a non-linear activation layer.

[0221] In a specific embodiment, based on sample fusion features, sample interaction indicator labels, and sample interaction labels, the feature fusion model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model and the feature fusion model corresponding to the feature fusion model to be trained. This can be referred to in the above detailed description of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on sample indicator association features, sample interaction indicator labels, and sample interaction labels to obtain the target interaction indicator recognition model, which will not be repeated here. Specifically, the relevant sample indicator association features in the joint training process can be replaced with the corresponding sample fusion features, and backpropagation training of the feature fusion model to be trained can be added during backpropagation. Correspondingly, the sample fusion features will be continuously updated as the model is updated during training.

[0222] In the above embodiments, during the joint training of the interaction indicator recognition model and the interaction recognition model to be trained, the introduction of associated task training features and sample identification features can enhance the model's memory on the basis of rich sample data, thereby improving the accuracy of interaction indicator recognition of the trained interaction indicator recognition model.

[0223] In a specific embodiment, such as Figure 5 As shown, Figure 5This is a schematic diagram illustrating the training process of an interactive indicator recognition model based on an example. Optionally, assuming the task models associated with the interactive indicator recognition model to be trained include a click-through rate (CTR) recognition model and a browsing duration recognition model, the associated task training features extracted from the associated task training information during the training of the CTR recognition model and the browsing duration recognition model, the sample indicator association features extracted from the sample indicator association information based on the indicator feature extraction model to be trained, and the sample identifier features extracted from the sample identifier information based on the identifier feature extraction model under the guidance of the model, can be input into the feature fusion model to be trained for fusion processing to obtain sample fusion features. Then, the first fusion feature corresponding to the positive sample account in the sample fusion features can be input into the interactive indicator recognition model to be trained for interactive indicator recognition processing; and the first fusion feature in the sample fusion features, and the feature extracted from the sample identifier information based on the identifier feature extraction model under the guidance of the ... The second fusion feature corresponding to the negative sample accounts obtained by machine negative sampling (the difference in the data volume corresponding to the first fusion feature and the second fusion feature is less than a preset threshold) is input into the interaction recognition model to be trained for interaction recognition processing. Then, the sample interaction index label, the sample interaction label, the predicted interaction index label output by the interaction index recognition model to be trained and the predicted interaction label output by the interaction recognition model to be trained are combined to perform backpropagation training on the interaction recognition model to be trained, the interaction index recognition model to be trained, the feature fusion model to be trained, the index feature extraction model to be trained and the identifier feature extraction model to be trained until the corresponding convergence condition is reached. The trained interaction recognition model, the target interaction index recognition model, the feature fusion model, the index feature extraction model and the identifier feature extraction model can be obtained.

[0224] Furthermore, it should be noted that during the aforementioned backpropagation process, the training feature extraction model for extracting training features of related tasks is not backpropagated for training in order to avoid affecting the related task model.

[0225] As can be seen from the technical solutions provided in the embodiments of this specification above, in the training process of the interaction indicator recognition model, this specification combines the historical resource interaction indicators that reflect the virtual resource volume brought by positive sample accounts among multiple historical exposure accounts to the corresponding object provider, and configures labels for multiple adjacent preset indicator intervals to generate sample interaction indicator labels. This can avoid the problem of underestimation of resource interaction indicators caused by the introduction of zero-value indicators during the resource interaction indicator learning process. Moreover, the sample interaction indicator label can represent the probability that the historical resource interaction indicator is greater than or equal to the lower limit of multiple preset indicator intervals, so that the sample interaction indicator label indicates the indicator interval where the historical resource interaction indicator of the positive sample account is located. The above indicates the size relationship of multiple preset indicator intervals, which can ensure the balanced distribution of the learned resource interaction indicators. In addition, by combining the sample indicator correlation features that represent the virtual resource consumption of multiple historical exposure objects and the virtual resource acquisition of corresponding object providers, as well as sample interaction labels, the interaction indicator recognition model to be trained and the interaction recognition model to be trained can be jointly trained. This allows the interaction indicator recognition model to be trained to learn from positive and negative samples from the dimension of whether object acquisition interaction operations are triggered during the learning of resource interaction indicators, avoiding the overestimation of the estimated resource interaction indicators. This can effectively improve the indicator recognition accuracy of the trained target interaction indicator recognition model, effectively assist in object recommendation, and greatly improve the user experience.

[0226] The following describes an object recommendation method based on the aforementioned target interaction metric identification model, such as... Figure 6 As shown, Figure 6 This is a flowchart illustrating an object recommendation method according to an exemplary embodiment. Optionally, the object recommendation method can be applied to electronic devices such as servers or terminals. Specifically, the object recommendation method may include the following steps:

[0227] In step S601, the associated features of the target indicator are obtained.

[0228] In one specific embodiment, the aforementioned target indicator association features characterize the virtual resource consumption of the target account and the virtual resource acquisition of at least one target object provider. The at least one target object provider is the provider of at least one object to be recommended. Specifically, the target account can be the account to which the object needs to be exposed (recommended). The at least one object to be recommended can be an object within the object recommendation platform. Optionally, based on application requirements, objects within the platform can be pre-filtered using click-through rate identification models, conversion rate identification models, browsing time identification models, etc., to determine the at least one object to be recommended.

[0229] In an optional embodiment, obtaining the correlation features of the target indicator may include:

[0230] Obtain the target metric association information corresponding to the target account;

[0231] The target indicator association information is input into the indicator feature extraction model for feature extraction processing to obtain the target indicator association features.

[0232] In a specific embodiment, the target indicator association information may include the cumulative virtual resource consumption information of the target account in at least one first preset historical time period, the single virtual resource consumption information of the target account in at least one first preset historical time period, the average virtual resources obtained by at least one target object provider based on the recommended object in a second preset historical time period (e.g., the average amount of virtual resources obtained in the live room based on the recommended object in the past 30 days, e.g., the average amount of virtual resources corresponding to a single recommended object in the live room in the past 30 days), and the cumulative virtual resources obtained by at least one target object provider based on the recommended object in the second preset historical time period.

[0233] In an optional embodiment, the above-mentioned indicator feature extraction model can be trained independently or trained together with the target interaction indicator recognition model (i.e., the indicator feature extraction model is jointly trained with the target interaction indicator recognition model).

[0234] In the above embodiments, by combining the indicator feature extraction model trained together with the target interaction indicator recognition model, the target indicator association features can be extracted from the target indicator association information, which can greatly improve the accuracy and effectiveness of the extracted target indicator association features, thereby enabling better recognition of interaction indicators and improving the accuracy of interaction indicator recognition.

[0235] In step S603, the target indicator association features are input into the target interaction indicator recognition model for interaction indicator recognition processing to obtain the target interaction indicator label.

[0236] In a specific embodiment, the aforementioned target interaction indicator label can characterize the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals; the aforementioned predicted resource interaction indicator can trigger object acquisition interaction for at least one object to be recommended for the target account, bringing the predicted amount of virtual resources to at least one target object provider; the multiple preset indicator intervals are multiple adjacent indicator intervals.

[0237] In step S605, the predicted resource interaction index is determined based on the target interaction index label;

[0238] In an optional embodiment, determining the predicted resource interaction metric based on the target interaction metric label may include:

[0239] Determine the mean of the indicators corresponding to multiple preset indicator intervals and the difference of the target indicator corresponding to each preset indicator interval; generate predictive resource interaction indicators based on the mean of the indicators and the difference of the target indicators.

[0240] In a specific embodiment, the target indicator difference corresponding to each preset indicator interval is the difference between the probability corresponding to each preset indicator interval (the probability that the predicted resource interaction indicator is greater than the lower limit of the preset indicator interval) and the probability corresponding to the previous preset indicator interval. The previous preset indicator interval is the indicator interval whose upper limit value is adjacent to the lower limit value of each preset indicator interval. The predicted indicator data corresponding to the previous preset indicator region of the first preset indicator interval is zero. The first preset indicator interval is the interval with the smallest upper limit value among multiple preset indicator intervals. Since, ideally, the target indicator difference between the preset indicator interval where the predicted resource interaction indicator is located and the previous preset indicator interval is 1, and the difference between the other intervals and the previous interval is 0, the above-mentioned combination of target indicator difference and indicator mean to determine the preset resource interaction indicator ensures that even if the actual predicted resource interaction indicator is greater than the lower limit value of the preset indicator interval deviates to a certain extent from the actual probability, the preset resource interaction indicator is mainly determined by the indicator mean value of the preset indicator interval where the predicted resource interaction indicator is located, effectively ensuring the accuracy of the preset resource interaction indicator identification.

[0241] In an optional embodiment, generating the predicted resource interaction index based on the index mean and the target index difference may include summing the products of the index mean corresponding to each preset index interval and the target index difference corresponding to each preset index interval to obtain the predicted resource interaction index.

[0242] In the above embodiments, the predicted resource interaction index is determined by combining the mean index value corresponding to multiple preset index intervals and the target index difference between the probability corresponding to each preset index interval and the probability corresponding to the previous preset index interval. This ensures that the preset resource interaction index is determined mainly by combining the mean index value of the preset index interval in which the predicted resource interaction index is located, thus effectively guaranteeing the accuracy of the identification of the preset resource interaction index.

[0243] In step S607, based on the predicted resource interaction index, at least one target object from the list of objects to be recommended is recommended to the target account.

[0244] In one specific embodiment, the object to be recommended with the highest corresponding predicted resource interaction index can be taken as the target object, and the target object can be sent to the terminal corresponding to the target account, thereby realizing the recommendation of the target object to the target account.

[0245] In an optional embodiment, the above method may further include:

[0246] Obtain the target-related task characteristics and / or the target identifier characteristics corresponding to the target account;

[0247] The target indicator association features, target association task features, and / or target identification features are input into the feature fusion model for fusion processing to obtain the target fusion features;

[0248] Accordingly, the above-mentioned input of the target indicator association features into the target interaction indicator recognition model for interaction indicator recognition processing yields target interaction indicator labels including:

[0249] The target fusion features are input into the target interaction indicator recognition model for interaction indicator recognition processing to obtain the target interaction indicator label.

[0250] In a specific embodiment, the aforementioned target-related task features can be features required to identify task metrics associated with predicted resource interaction metrics; specifically, the task metrics associated with predicted resource interaction metrics can be metrics required to be identified by the task model associated with the target interaction metric identification model, such as click-through rate, conversion rate, and browsing time. The aforementioned feature fusion model is obtained through joint training with the target interaction metric identification model; specifically, the target-related task features can be extracted from target-related task information; specifically, the target-related task information can be task data required by at least one task model in the process of identifying the corresponding metrics. Optionally, the target-related task information can include account feature information corresponding to the target account (e.g., the gender of the user corresponding to the account, historical interaction operations, etc.), historical operation information of the target account for the corresponding object to be recommended, object feature information corresponding to the object to be recommended, and interaction features of the object to be recommended within a preset current time period, wherein the preset current time period includes the current time point.

[0251] In an optional embodiment, the aforementioned target identifier features can be obtained by extracting features from target identifier information based on an identifier feature extraction model. Optionally, the identifier feature extraction model can be jointly trained with a target interaction indicator recognition model, or it can be trained independently. Optionally, the target identifier information can include the account identifier information of the target account, the object identifier information of the historical recommended objects corresponding to the target account, and the account identifier information of the publishing account corresponding to the historical recommended objects. Correspondingly, the target identifier features can include the account identifier features of the target account, the object identifier features of the historical recommended objects corresponding to the target account, and the account identifier features of the publishing account corresponding to the historical recommended objects.

[0252] In the above embodiments, introducing target-related task features and / or target identifier features during the interaction indicator recognition process can enhance the model's memory by enriching the data used to learn the interaction indicators, thereby improving the accuracy of the interaction indicator recognition model.

[0253] As can be seen from the technical solutions provided in the embodiments of this specification above, in the object recommendation process, by inputting the target indicator association features corresponding to the target account into the target interaction indicator recognition model for interaction indicator recognition processing, a target interaction indicator label is obtained. This target interaction indicator label can characterize the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals. This ensures that the target interaction indicator label, while indicating the indicator interval where the predicted resource interaction indicator corresponding to the target account is located, also indicates the size relationship of multiple preset indicator intervals, thus guaranteeing the balanced distribution of the learned resource interaction indicators. Then, combined with the target interaction indicator label, a predicted resource interaction indicator reflecting the target account triggering object acquisition interaction for at least one object to be recommended, bringing virtual resources to at least one target object provider, can be determined. This can effectively guarantee the accuracy of indicator recognition, thereby effectively assisting object recommendation and greatly improving the user experience.

[0254] Figure 7 This is a block diagram illustrating an interactive indicator recognition model training device according to an exemplary embodiment. (Refer to...) Figure 7 The device includes:

[0255] The sample data acquisition module 710 is configured to acquire sample indicator association features corresponding to multiple historical exposure accounts, sample interaction tags corresponding to multiple historical exposure accounts, and historical resource interaction indicators corresponding to positive sample accounts among multiple historical exposure accounts. The historical resource interaction indicators are the amount of virtual resources brought by the corresponding object provider for the object acquisition interaction operation triggered by the positive sample account. The sample indicator association features are features that characterize the virtual resource consumption of multiple historical exposure accounts and the virtual resource acquisition of the corresponding object providers of multiple historical exposure accounts. The sample interaction tags characterize the probability of multiple historical exposure accounts triggering object acquisition interaction operations.

[0256] The label configuration module 720 is configured to perform label configuration on multiple preset indicator intervals based on historical resource interaction indicators to obtain sample interaction indicator labels. The sample interaction indicator labels represent the probability that the historical resource interaction indicator is greater than or equal to the lower limit of multiple preset indicator intervals. The multiple preset indicator intervals are multiple adjacent indicator intervals.

[0257] The joint training module 730 is configured to perform joint training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on sample indicator association features, sample interaction indicator labels and sample interaction labels, so as to obtain the target interaction indicator recognition model.

[0258] In an optional embodiment, the multiple historical exposure accounts also include negative sample accounts; the joint training module 730 includes:

[0259] The random sampling unit is configured to perform random sampling on the first indicator association feature corresponding to the negative sample account in the sample indicator association feature to obtain the second indicator association feature; the difference between the data volume corresponding to the second indicator association feature and the data volume corresponding to the third indicator association feature is less than a preset threshold, and the third indicator association feature is the indicator association feature corresponding to the positive sample account in the sample indicator association feature.

[0260] The interaction indicator recognition and processing unit is configured to input the third indicator association features into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and obtain the predicted interaction indicator label corresponding to the positive sample account.

[0261] The interaction recognition processing unit is configured to input the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label.

[0262] The joint training unit is configured to perform joint training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, so as to obtain the target interaction indicator recognition model.

[0263] In an optional embodiment, the above-described apparatus further includes:

[0264] The associated task training feature acquisition module is configured to acquire associated task training features corresponding to multiple historical exposure accounts; the associated task training features are the extracted task training features in the process of training at least one task model associated with the interaction index recognition model to be trained.

[0265] The first interactive indicator recognition and processing module is specifically configured to input the first training feature corresponding to the positive sample account in the third indicator association feature and the association task training feature into the interactive indicator recognition model to be trained for interactive indicator recognition processing, and obtain the predicted interactive indicator label.

[0266] The interaction recognition processing module is specifically configured to perform interaction recognition processing by inputting the second indicator association features and the third indicator association features, the first training features and the second training features into the interaction recognition model to be trained, and obtain the predicted interaction label; the second training features are the training features of the target negative sample account in the association task training features; the target negative sample account is the negative sample account corresponding to the second indicator association features.

[0267] In an optional embodiment, the above-described apparatus further includes:

[0268] The sample identifier feature acquisition module is configured to acquire sample identifier features corresponding to multiple historical exposure accounts;

[0269] The first interactive indicator recognition and processing module is specifically configured to input the first identifier feature corresponding to the positive sample account in the third indicator association feature and sample identifier feature into the interactive indicator recognition model to be trained for interactive indicator recognition processing, and obtain the predicted interactive indicator label.

[0270] The interaction recognition processing module is specifically configured to input the second indicator association feature and the third indicator association feature, the first identifier feature and the second identifier feature into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second identifier feature is the identifier feature of the target negative sample account in the sample identifier feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

[0271] In an optional embodiment, the sample identifier feature acquisition module includes:

[0272] The sample identification information acquisition unit is configured to acquire sample identification information corresponding to multiple historical exposure accounts;

[0273] The first feature extraction processing unit is configured to perform feature extraction processing by inputting sample identification information into the identification feature extraction model to be trained, and to obtain sample identification features.

[0274] The interaction recognition processing module is specifically configured to perform joint training on the target interaction indicator recognition model, the target identifier feature extraction model, the target interaction recognition model, and the target interaction indicator recognition model based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, so as to obtain the identifier feature extraction model corresponding to the target interaction indicator recognition model and the target identifier feature extraction model.

[0275] In an optional embodiment, the sample data acquisition module 710 includes:

[0276] The sample indicator association information acquisition unit is configured to acquire sample indicator association information corresponding to multiple historical exposure accounts. The sample indicator association information includes information on virtual resource consumption of multiple historical exposure accounts and information on virtual resource acquisition of object providers corresponding to multiple historical exposure accounts.

[0277] The second feature extraction processing unit is configured to input the sample indicator association information into the indicator feature extraction model to be trained for feature extraction processing to obtain the sample indicator association features.

[0278] The joint training module 730 is specifically configured to perform joint training on the indicator feature extraction model, the interaction recognition model, and the interaction indicator recognition model to be trained, based on the sample indicator association features, sample interaction indicator labels, and sample interaction labels, to obtain the indicator feature extraction model corresponding to the target interaction indicator recognition model and the indicator feature extraction model to be trained.

[0279] In an optional embodiment, the plurality of historical exposure accounts further includes negative sample accounts; the above apparatus also includes:

[0280] The feature acquisition module is configured to acquire the associated task training features and / or the sample identification features corresponding to multiple historical exposure accounts; the associated task training features are the task training features extracted during the training of at least one task model associated with the interaction index recognition model to be trained.

[0281] The first fusion processing module is configured to input the sample index association features, association task training features and / or sample identification features into the feature fusion model to be trained for fusion processing to obtain sample fusion features;

[0282] The joint training module 730 is specifically configured to perform joint training on the feature fusion model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained based on sample fusion features, sample interaction indicator labels, and sample interaction labels, so as to obtain the feature fusion model corresponding to the target interaction indicator recognition model and the feature fusion model to be trained.

[0283] In an optional embodiment, the tag configuration module 720 includes:

[0284] The target indicator interval determination unit is configured to determine the target indicator interval where the historical resource interaction indicator is located from multiple preset indicator intervals;

[0285] The first label configuration unit is configured to perform label configuration on the first indicator interval based on the first preset label to obtain the first interactive indicator label; the first indicator interval is the interval among multiple preset indicator intervals whose upper limit value is less than the target indicator and whose lower limit value is greater than or equal to the multiple preset indicator intervals.

[0286] The second label configuration unit is configured to perform label configuration on the target indicator range and the second indicator range based on the second preset label to obtain the second interactive indicator label; the second indicator range is the range among multiple preset indicator ranges whose lower limit is greater than the upper limit of the target indicator range.

[0287] The sample interaction indicator label generation unit is configured to generate sample interaction indicator labels based on the first interaction indicator label and the interaction indicator label.

[0288] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0289] Figure 8 This is a block diagram illustrating an object recommendation device according to an exemplary embodiment. (Refer to...) Figure 8 The device includes:

[0290] The target indicator association feature acquisition module 810 is configured to acquire target indicator association features; the target indicator association features are features that characterize the virtual resource consumption of the target account and the virtual resource acquisition of at least one target object provider, and at least one target object provider is the provider of at least one object to be recommended;

[0291] The second interaction indicator recognition and processing module 820 is configured to perform interaction indicator recognition processing by inputting the target indicator association features into the target interaction indicator recognition model obtained based on the interaction indicator recognition model training method provided in the first aspect, and obtain the target interaction indicator label. The target interaction indicator label represents the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals. The predicted resource interaction indicator is the predicted amount of virtual resources brought to at least one target object provider by the target account triggering object acquisition interaction for at least one object to be recommended. The multiple preset indicator intervals are multiple adjacent indicator intervals.

[0292] The predictive resource interaction indicator determination module 830 is configured to determine the predictive resource interaction indicator based on the target interaction indicator label.

[0293] The object push module 840 is configured to recommend a target object from at least one of the objects to be recommended to a target account based on predicted resource interaction metrics.

[0294] In an optional embodiment, the above-described apparatus further includes:

[0295] The feature acquisition module is configured to acquire the target-related task features and / or the target identifier features corresponding to the target account; the target-related task features are the features required to identify the task indicators associated with the predicted resource interaction indicators.

[0296] The second fusion processing module is configured to input target indicator association features, target association task features, and / or target identification features into a feature fusion model for fusion processing to obtain target fusion features; the feature fusion model is jointly trained with the target interaction indicator recognition model.

[0297] The second interactive indicator recognition and processing module 820 is specifically configured to input the target fusion features into the target interactive indicator recognition model for interactive indicator recognition processing to obtain the target interactive indicator label.

[0298] In an optional embodiment, the predictive resource interaction index determination module 830 includes:

[0299] The calculation unit is configured to determine the mean of indicators corresponding to multiple preset indicator intervals and the target indicator difference corresponding to each preset indicator interval. The target indicator difference corresponding to each preset indicator interval is the difference between the probability corresponding to each preset indicator interval and the probability corresponding to the previous preset indicator interval. The previous preset indicator interval is the indicator interval whose upper limit value is adjacent to the lower limit value of each preset indicator interval. The predicted indicator data corresponding to the previous preset indicator region of the first preset indicator interval is zero. The first preset indicator interval is the interval with the smallest upper limit value among multiple preset indicator intervals.

[0300] The predictive resource interaction index generation unit is configured to generate predictive resource interaction indices based on the difference between the index mean and the target index.

[0301] In an optional embodiment, the target indicator association feature acquisition module 810 includes:

[0302] The target indicator association information acquisition unit is configured to acquire the target indicator association information corresponding to the target account. The target indicator association information is information representing the virtual resource consumption of the target account and the virtual resource acquisition of at least one object to be recommended.

[0303] The third feature extraction processing unit is configured to input the target indicator association information into the indicator feature extraction model for feature extraction processing to obtain the target indicator association features. The indicator feature extraction model is jointly trained with the target interaction indicator recognition model.

[0304] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0305] Figure 9 This is a block diagram illustrating an electronic device for training an interactive metric recognition model according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 9As shown, the electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an interactive index recognition model training method.

[0306] Figure 10 This is a block diagram illustrating an electronic device for object recommendation according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an object recommendation method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0307] Those skilled in the art will understand that Figure 9 or Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0308] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the interactive metric recognition model training method or object recommendation method as described in the embodiments of this disclosure.

[0309] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the interactive index recognition model training method or object recommendation method of the present disclosure embodiments.

[0310] In an exemplary embodiment, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute the interactive index recognition model training method or object recommendation method of the present disclosure embodiments.

[0311] 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. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0312] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0313] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for training an interactive indicator recognition model, characterized in that, include: Acquire sample indicator association features corresponding to multiple historical exposure accounts, sample interaction tags corresponding to multiple historical exposure accounts, and historical resource interaction indicators corresponding to positive sample accounts among the multiple historical exposure accounts. The historical resource interaction indicators are the virtual resource quantity brought to the corresponding object provider by the object acquisition interaction operation triggered by the positive sample account. The sample index association features characterize the virtual resource consumption of multiple historical exposure accounts and the virtual resource acquisition of the object providers corresponding to multiple historical exposure accounts; the sample interaction tags characterize the probability of multiple historical exposure accounts triggering object acquisition interaction operations. Based on the historical resource interaction indicators, labels are configured for multiple preset indicator intervals to obtain sample interaction indicator labels. The sample interaction indicator labels represent the probability that the historical resource interaction indicator is greater than or equal to the lower limit of multiple preset indicator intervals. The multiple preset indicator intervals are multiple adjacent indicator intervals. Based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model.

2. The interaction indicator recognition model training method according to claim 1, characterized in that, The aforementioned historical exposure accounts also include negative sample accounts; the joint training of the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator tags, and the sample interaction tags to obtain the target interaction indicator recognition model includes: Randomly sample the first indicator association feature corresponding to the negative sample account in the sample indicator association features to obtain the second indicator association feature; the difference between the data volume corresponding to the second indicator association feature and the data volume corresponding to the third indicator association feature is less than a preset threshold, and the third indicator association feature is the indicator association feature corresponding to the positive sample account in the sample indicator association features. The third indicator association feature is input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label corresponding to the positive sample account. The second indicator association feature and the third indicator association feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label. Based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, the interaction indicator recognition model to be trained and the interaction recognition model to be trained are jointly trained to obtain the target interaction indicator recognition model.

3. The interaction indicator recognition model training method according to claim 2, characterized in that, The method further comprises: Obtain associated task training features corresponding to multiple historical exposure accounts; the associated task training features are extracted task training features during the process of training at least one task model associated with the interaction index recognition model to be trained. The step of inputting the third indicator association feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and obtaining the predicted interaction indicator label corresponding to the positive sample account, includes: The first training feature corresponding to the positive sample account in the third indicator association feature and the association task training feature is input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label. The step of inputting the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label includes: The second indicator association feature, the third indicator association feature, the first training feature, and the second training feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second training feature is the training feature of the target negative sample account in the association task training feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

4. The interaction indicator recognition model training method according to claim 2, characterized in that, The method further includes: Obtain sample identifier features corresponding to multiple historical exposure accounts; The step of inputting the third indicator association feature into the interaction indicator recognition model to be trained for interaction indicator recognition processing, and obtaining the predicted interaction indicator label corresponding to the positive sample account, includes: The first identifier feature corresponding to the positive sample account in the third indicator association feature and the sample identifier feature are input into the interaction indicator recognition model to be trained for interaction indicator recognition processing to obtain the predicted interaction indicator label. The step of inputting the second indicator association features and the third indicator association features into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label includes: The second indicator association feature, the third indicator association feature, the first identifier feature, and the second identifier feature are input into the interaction recognition model to be trained for interaction recognition processing to obtain the predicted interaction label; the second identifier feature is the identifier feature of the target negative sample account in the sample identifier feature; the target negative sample account is the negative sample account corresponding to the second indicator association feature.

5. The interaction indicator recognition model training method according to claim 4, characterized in that, The step of obtaining sample identifier features corresponding to multiple historical exposure accounts includes: Obtain sample identification information corresponding to multiple historical exposure accounts; The sample identification information is input into the identification feature extraction model to be trained for feature extraction processing to obtain the sample identification features; The step of jointly training the interaction recognition model and the interaction recognition model to be trained in the interaction recognition model to be trained, based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, to obtain the target interaction indicator recognition model includes: Based on the predicted interaction indicator label, the sample interaction indicator label, the predicted interaction label, and the interaction label corresponding to the predicted interaction label in the sample interaction label, the target interaction indicator recognition model, the target identifier feature extraction model, the target interaction recognition model, and the target interaction indicator recognition model in the target interaction indicator recognition model are jointly trained to obtain the identifier feature extraction model corresponding to the target interaction indicator recognition model and the target identifier feature extraction model.

6. The method for training an interaction indicator recognition model according to any one of claims 1 to 5, characterized in that, The method for obtaining the sample metric correlation features corresponding to multiple historical exposure accounts includes: Obtain sample indicator association information corresponding to multiple historical exposure accounts. The sample indicator association information includes information on virtual resource consumption corresponding to multiple historical exposure accounts and information on virtual resource acquisition by the object providers corresponding to multiple historical exposure accounts. The sample indicator association information is input into the indicator feature extraction model to be trained for feature extraction processing to obtain the sample indicator association features. The step of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels to obtain the target interaction indicator recognition model includes: Based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels, the indicator feature extraction model to be trained, the interaction recognition model to be trained, and the interaction indicator recognition model to be trained are jointly trained to obtain the indicator feature extraction model corresponding to the target interaction indicator recognition model and the indicator feature extraction model to be trained.

7. The method for training an interaction indicator recognition model according to any one of claims 1 to 5, characterized in that, The aforementioned historical exposure accounts also include negative sample accounts; the method further includes: Obtain associated task training features and / or sample identification features corresponding to multiple historical exposure accounts; the associated task training features are task training features extracted during the training of at least one task model associated with the interaction index recognition model to be trained. The sample indicator association features, the association task training features, and / or the sample identification features are input into the feature fusion model to be trained for fusion processing to obtain sample fusion features; The step of jointly training the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels, and the sample interaction labels to obtain the target interaction indicator recognition model includes: Based on the sample fusion features, the sample interaction index labels, and the sample interaction labels, the feature fusion model to be trained, the interaction recognition model to be trained, and the interaction index recognition model to be trained are jointly trained to obtain the feature fusion model corresponding to the target interaction index recognition model and the feature fusion model to be trained.

8. The method for training an interaction indicator recognition model according to any one of claims 1 to 5, characterized in that, The step of configuring labels for multiple preset indicator ranges based on the historical resource interaction indicators to obtain sample interaction indicator labels includes: Determine the target indicator interval where the historical resource interaction indicator is located from multiple preset indicator intervals; Based on the first preset label, the first indicator interval is labeled to obtain the first interactive indicator label; the first indicator interval is the interval in which the upper limit value of the multiple preset indicator intervals is less than the target indicator and greater than or equal to the lower limit value of the multiple preset indicator intervals. Based on the second preset label, the target indicator interval and the second indicator interval are labeled to obtain the second interactive indicator label; the second indicator interval is the interval among the plurality of preset indicator intervals whose lower limit is greater than the upper limit of the target indicator interval; The sample interaction indicator label is generated based on the first interaction indicator label and the interaction indicator label.

9. An object recommendation method, characterized in that, include: Obtain the correlation features of the target indicator; The target indicator association feature is a feature that characterizes the virtual resource consumption of the target account and the virtual resource acquisition of at least one target object provider, wherein the at least one target object provider is the provider of at least one object to be recommended; The target indicator association features are input into the target interaction indicator recognition model obtained based on the interaction indicator recognition model training method according to any one of claims 1 to 8 for interaction indicator recognition processing to obtain target interaction indicator labels. The target interaction indicator labels represent the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals. The predicted resource interaction indicator is the predicted amount of virtual resources brought to at least one target object provider by the target account triggering object acquisition interaction for at least one of the recommended objects. The multiple preset indicator intervals are multiple adjacent indicator intervals. The predicted resource interaction index is determined based on the target interaction index label; Based on the predicted resource interaction metrics, at least one target object from the list of objects to be recommended is recommended to the target account.

10. The object recommendation method according to claim 9, characterized in that, The method further comprises: Obtain the target-related task features and / or the target identifier features corresponding to the target account; the target-related task features are the features required to identify the task indicators associated with the predicted resource interaction indicators. The target indicator association features, the target association task features, and / or target identification features are input into a feature fusion model for fusion processing to obtain target fusion features; the feature fusion model is jointly trained with the target interaction indicator recognition model. The step of inputting the target indicator association features into the target interactive indicator recognition model obtained based on the interactive indicator recognition model training method according to any one of claims 1 to 8 for interactive indicator recognition processing to obtain the target interactive indicator label includes: The target fusion features are input into the target interaction indicator recognition model for interaction indicator recognition processing to obtain the target interaction indicator label.

11. The object recommendation method according to any one of claims 9 or 10, characterized in that, The step of determining the predicted resource interaction index based on the target interaction index label includes: The mean value of the indicators corresponding to multiple preset indicator intervals and the target indicator difference corresponding to each preset indicator interval are determined. The target indicator difference corresponding to each preset indicator interval is the difference between the probability corresponding to each preset indicator interval and the probability corresponding to the previous preset indicator interval. The previous preset indicator interval is the indicator interval whose upper limit value is adjacent to the lower limit value of each preset indicator interval. The predicted indicator data corresponding to the preset indicator region preceding the first preset indicator interval is zero. The first preset indicator interval is the interval with the smallest upper limit value among multiple preset indicator intervals. The predicted resource interaction index is generated based on the difference between the mean of the index and the target index.

12. The object recommendation method according to any one of claims 9 or 10, characterized in that, The features associated with the target indicator include: Obtain the target indicator association information corresponding to the target account. The target indicator association information is information that characterizes the virtual resource consumption of the target account and the virtual resource acquisition of at least one object to be recommended. The target indicator association information is input into the indicator feature extraction model for feature extraction processing to obtain the target indicator association features. The indicator feature extraction model is jointly trained with the target interaction indicator recognition model.

13. A training device for an interactive indicator recognition model, characterized in that, include: The sample data acquisition module is configured to acquire sample indicator association features corresponding to multiple historical exposure accounts, sample interaction tags corresponding to multiple historical exposure accounts, and historical resource interaction indicators corresponding to positive sample accounts among the multiple historical exposure accounts. The historical resource interaction indicators are the virtual resource quantity brought by the corresponding object provider for the object acquisition interaction operation triggered by the positive sample account. The sample index association features characterize the virtual resource consumption of multiple historical exposure accounts and the virtual resource acquisition of the object providers corresponding to multiple historical exposure accounts; the sample interaction tags characterize the probability of multiple historical exposure accounts triggering object acquisition interaction operations. The tag configuration module is configured to perform tag configuration on multiple preset indicator intervals based on the historical resource interaction indicators to obtain sample interaction indicator tags. The sample interaction indicator tags represent the probability that the historical resource interaction indicators are greater than or equal to the lower limit values ​​of multiple preset indicator intervals. The multiple preset indicator intervals are multiple adjacent indicator intervals. The joint training module is configured to perform joint training on the interaction indicator recognition model to be trained and the interaction recognition model to be trained based on the sample indicator association features, the sample interaction indicator labels and the sample interaction labels, so as to obtain the target interaction indicator recognition model.

14. An object recommendation device, characterized in that, include: The target indicator association feature acquisition module is configured to acquire the target indicator association features; The target indicator association feature is a feature that characterizes the virtual resource consumption of the target account and the virtual resource acquisition of at least one target object provider, wherein the at least one target object provider is the provider of at least one object to be recommended; The second interaction indicator recognition and processing module is configured to perform interaction indicator recognition processing by inputting the target indicator association features into a target interaction indicator recognition model obtained based on the interaction indicator recognition model training method according to any one of claims 1 to 8, to obtain a target interaction indicator label. The target interaction indicator label represents the probability that the predicted resource interaction indicator corresponding to the target account is greater than or equal to the lower limit of multiple preset indicator intervals. The predicted resource interaction indicator is the predicted amount of virtual resources brought to at least one target object provider by the target account triggering an object acquisition interaction for at least one of the recommended objects. The multiple preset indicator intervals are multiple adjacent indicator intervals. The predictive resource interaction index determination module is configured to determine the predictive resource interaction index based on the target interaction index label. The object push module is configured to recommend at least one target object from the objects to be recommended to the target account based on the predicted resource interaction metrics.

15. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the interactive indicator recognition model training method as described in any one of claims 1 to 8 or the object recommendation method as described in any one of claims 9 to 12.

16. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the interactive indicator recognition model training method as described in any one of claims 1 to 8 or the object recommendation method as described in any one of claims 9 to 12.

Images