Transformation prediction model training method and device

Abstract

The embodiment of the invention provides a transformation prediction model training method and device, and the method comprises the steps: carrying out the training of a transformation prediction model in a model training process of the transformation prediction model, after a sample set constructed by user features of each user in multiple resource purchase channels and user data of a target resource purchase channel is obtained, conversion prediction processing is performed on the sample set through a to-be-trained model of the target resource purchase channel, and a conversion index of the pushed resource amount of each user for the target resource purchase channel is obtained. Carrying out loss calculation by combining the conversion index, the sample number of the sample set and the sample label corresponding to each user sample in the sample set, and carrying out parameter adjustment on the to-be-trained model based on the calculated loss to obtain a conversion prediction model; the conversion prediction model is used for pushing the resource amount in the resource purchase process of the target resource purchase channel.

Description

[0001] This patent application is a divisional application of the Chinese patent application with application number CN202411000899.7, application date July 24, 2024, and invention name “Conversion Prediction Model Training Method and Device”. Technical Field

[0002] This document relates to the field of data processing technology, and in particular to a conversion prediction model training method and device. Background Art

[0003] With the continuous development of Internet technology and the continuous advancement of social information, in order to realize resource appreciation and management, more and more users have begun to purchase resources online, and achieve the purpose of resource management and resource appreciation through resource purchase. This method helps the flow of user resources. In the process of user resource purchase, different users may have different purchasing needs, and different users may have different resource purchasing preferences. All these make the resource purchasing service provider face greater pressure and challenges. Summary of the invention

[0004] One or more embodiments of the present specification provide a conversion prediction model training method, including: obtaining user features of each user in multiple resource purchasing channels and a sample set constructed by user data of a target resource purchasing channel. The user features are obtained by clustering the initial user features of each user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and feature screening is performed based on the clustering results. The sample set is input into the model to be trained of the target resource purchasing channel for conversion prediction processing to obtain a conversion indicator of the amount of resources pushed by each user for the target resource purchasing channel. Loss calculation is performed based on the conversion indicator, the number of samples in the sample set, and the sample labels corresponding to each user sample in the sample set. The parameters of the model to be trained are adjusted based on the calculated loss to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchasing channel.

[0005] One or more embodiments of the present specification provide a push processing method, including: obtaining user characteristics of a user in multiple resource purchase channels and user data of a target resource purchase channel. The user characteristics are clustered by a clustering network in a global conversion prediction model to the initial user characteristics of the user in the multiple resource purchase channels, and feature screening is performed based on the clustering results. The user characteristics and the user data are input into the conversion prediction model of the target resource purchase channel for conversion prediction processing to obtain a conversion index of the user's pushed resource amount for the target resource purchase channel. According to the conversion index, a target pushed resource amount is determined from the pushed resource amount, and based on the target pushed resource amount, the resource amount is pushed during the resource purchase process of the target resource purchase channel.

[0006] One or more embodiments of the present specification provide a conversion prediction model training device, including: a sample set acquisition module, configured to acquire user features of each user in multiple resource purchase channels and a sample set constructed by user data of a target resource purchase channel. The user features are obtained by clustering the initial user features of each user in the multiple resource purchase channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results. The conversion prediction module is configured to input the sample set into the model to be trained of the target resource purchase channel for conversion prediction processing, and obtain the conversion index of the amount of resources pushed by each user for the target resource purchase channel. The loss calculation module is configured to perform loss calculation based on the conversion index, the number of samples in the sample set, and the sample labels corresponding to each user sample in the sample set. The parameter adjustment module is configured to adjust the parameters of the model to be trained based on the calculated loss, so as to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchase channel.

[0007] One or more embodiments of the present specification provide a push processing device, including: a data acquisition module, configured to obtain user characteristics of users in multiple resource purchase channels and user data of a target resource purchase channel. The user characteristics are obtained by clustering the initial user characteristics of the user in the multiple resource purchase channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results. A conversion prediction processing module is configured to input the user characteristics and the user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing, and obtain a conversion indicator of the user's pushed resource amount for the target resource purchase channel. A resource amount push module is configured to determine a target pushed resource amount from the pushed resource amount according to the conversion indicator, and based on the target pushed resource amount, push the resource amount in the resource purchase process of the target resource purchase channel.

[0008] One or more embodiments of the present specification provide a conversion prediction model training device, including: a processor; and a memory configured to store computer executable instructions, wherein when the computer executable instructions are executed, the processor: obtains the user characteristics of each user in multiple resource purchase channels and the user data of the target resource purchase channel to construct a sample set. The user characteristics are obtained by clustering the initial user characteristics of each user in the multiple resource purchase channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results. The sample set is input into the model to be trained of the target resource purchase channel for conversion prediction processing to obtain the conversion index of the amount of pushed resources of each user for the target resource purchase channel. According to the conversion index, the number of samples in the sample set and the sample labels corresponding to each user sample in the sample set, a loss calculation is performed. Based on the calculated loss, the parameters of the model to be trained are adjusted to obtain a conversion prediction model for pushing resource amounts in the resource purchase process of the target resource purchase channel.

[0009] One or more embodiments of the present specification provide a push processing device, including: a processor; and a memory configured to store computer executable instructions, wherein the computer executable instructions, when executed, cause the processor to: obtain user characteristics of a user in multiple resource purchase channels and user data of a target resource purchase channel. The user characteristics are obtained by clustering the initial user characteristics of the user in the multiple resource purchase channels through a clustering network in a global conversion prediction model, and feature screening is performed based on the clustering results. The user characteristics and the user data are input into the conversion prediction model of the target resource purchase channel for conversion prediction processing, and a conversion index of the user's pushed resource amount for the target resource purchase channel is obtained. A target pushed resource amount is determined from the pushed resource amount according to the conversion index, and based on the target pushed resource amount, the resource amount is pushed during the resource purchase process of the target resource purchase channel.

[0010] One or more embodiments of the present specification provide a computer-readable storage medium for storing computer-executable instructions, which implement the following steps when executed: obtaining the user characteristics of each user in multiple resource purchasing channels and the user data of the target resource purchasing channel to construct a sample set. The user characteristics are obtained by clustering the initial user characteristics of each user in the multiple resource purchasing channels through the clustering network in the global conversion prediction model, and performing feature screening based on the clustering results. The sample set is input into the model to be trained of the target resource purchasing channel for conversion prediction processing to obtain the conversion index of the amount of resources pushed by each user for the target resource purchasing channel. Loss calculation is performed based on the conversion index, the number of samples in the sample set, and the sample labels corresponding to each user sample in the sample set. The parameters of the model to be trained are adjusted based on the calculated loss to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchasing channel.

[0011] One or more embodiments of the present specification provide another computer-readable storage medium for storing computer-executable instructions, which implement the following steps when executed: obtaining user characteristics of a user in multiple resource purchasing channels and user data of a target resource purchasing channel. The user characteristics are obtained by clustering the initial user characteristics of the user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results. The user characteristics and the user data are input into the conversion prediction model of the target resource purchasing channel for conversion prediction processing to obtain a conversion index of the user's pushed resource amount for the target resource purchasing channel. According to the conversion index, a target pushed resource amount is determined from the pushed resource amount, and based on the target pushed resource amount, the resource amount is pushed during the resource purchase process of the target resource purchasing channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate one or more embodiments of the present specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor. Figure 1 A schematic diagram of an implementation environment of a conversion prediction model training method provided in one or more embodiments of this specification; Figure 2 A processing flow chart of a conversion prediction model training method provided by one or more embodiments of this specification; Figure 3 A processing flow chart of a conversion prediction model training method applied to a model training scenario provided in one or more embodiments of this specification; Figure 4 A processing flow chart of a push processing method provided in one or more embodiments of this specification; Figure 5 A processing flow chart of a push processing method applied to a resource purchase scenario provided in one or more embodiments of this specification; Figure 6 A schematic diagram of an embodiment of a conversion prediction model training device provided by one or more embodiments of this specification; Figure 7 A schematic diagram of an embodiment of a push processing device provided in one or more embodiments of this specification; Figure 8 A schematic diagram of the structure of a conversion prediction model training device provided in one or more embodiments of this specification; Fig. 9 A schematic diagram of the structure of a push processing device provided in one or more embodiments of this specification. DETAILED DESCRIPTION

[0013] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0014] The conversion prediction model training method provided in one or more embodiments of this specification can be applied to the implementation environment of model training. Figure 1 , the implementation environment at least includes: A model to be trained 101 for performing conversion prediction processing, and a training system 102 for training the model to be trained 101; in addition, the implementation environment may also include a global conversion prediction model 103 for performing clustering processing and feature screening.

[0015] In this implementation environment, in the process of training the conversion prediction model, the training system 102 obtains a sample set, performs conversion prediction processing based on the sample set through the to-be-trained model 101 of the target resource purchase channel, obtains the conversion index of the amount of pushed resources of each user in the target resource purchase channel, and performs loss calculation based on the conversion index of the amount of pushed resources of each user in the target resource purchase channel, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set. Based on the calculated loss, the parameters of the to-be-trained model 101 are adjusted to obtain a conversion prediction model for pushing resource amounts in the resource purchase process of the target resource purchase channel; Among them, the sample set can be constructed based on the user characteristics of each user in multiple resource purchasing channels and the user data of each user in the target resource purchasing channel; the user characteristics of each user in multiple resource purchasing channels are clustered by the clustering network in the global conversion prediction model 103 to cluster the initial user characteristics of each user in multiple resource purchasing channels, and the characteristics are obtained based on the clustering results.

[0016] One or more embodiments of a conversion prediction model training method provided in this specification are as follows: Reference Figure 2 The conversion prediction model training method provided in this embodiment specifically includes steps S202 to S208.

[0017] Step S202 , obtaining user characteristics of each user in multiple resource purchasing channels and user data of the target resource purchasing channel to construct a sample set.

[0018] Each resource purchasing channel in the multiple resource purchasing channels described in this embodiment refers to each channel for purchasing resources. Each resource purchasing channel can be used to purchase the same resource, which can be a financial resource or a financial project. The financial resource or financial project can specifically be a fund, etc. This embodiment does not specifically limit the type of financial resource or financial project. For this purpose, the multiple resource purchasing channels can be multiple channels for purchasing the same financial resource or the same financial project. The multiple resource purchasing channels may include the target resource purchasing channel, or may not include the target resource purchasing channel.

[0019] The user data refers to the user data of each user in the target resource purchasing channel; optionally, the user data includes at least one of the following: user attribute data, user resource behavior data, user purchase behavior data, user resource data; wherein, the user attribute data may include age, occupation and / or male or female data, in addition, the user attribute data may also be other data; the user resource behavior data includes the behavior data of each user in the target resource purchasing channel, specifically, it may be the activity of each user in the target resource purchasing channel; the user purchase behavior data includes the purchase data of each user for resource purchases in the target resource purchasing channel, such as the purchase amount and purchase frequency of each user for resource purchases in the target resource purchasing channel; the user resource data includes the resource holding data of each user.

[0020] There may be multiple user samples in the sample set, and each user sample may contain two parts of data, specifically, the user features of each user in multiple resource purchase channels and the user data of each user in the target resource purchase channel. Optionally, the user features of each user in multiple resource purchase channels are obtained by clustering the initial user features of each user in the multiple resource purchase channels through the clustering network in the global conversion prediction model, and performing feature screening based on the clustering results; the clustering process here may be an adaptive clustering process; Among them, the global conversion prediction model refers to a global conversion prediction model obtained by training the second model to be trained based on a second sample set constructed based on user data of each user in multiple resource purchasing channels; the global conversion prediction model can be used to predict the conversion indicators of the amount of pushed resources for users for multiple resource purchasing channels.

[0021] In actual applications, in the process of training the conversion prediction model of the target resource purchase channel, since the user data of the target resource purchase channel may be relatively single and limited, if only the sample set constructed based on the user data of the target resource purchase channel is used for training, the accuracy of the conversion prediction model of the target resource purchase channel obtained through training may be low. In view of this, in order to improve the comprehensiveness of the sample set, and thus improve the accuracy of the conversion prediction model of the trained target resource purchase channel, and further enhance the effectiveness of pushing resource amounts, this embodiment introduces a global conversion prediction model of multiple resource purchase channels, and uses the global conversion prediction model to assist in the training of the conversion prediction model of a certain target resource purchase channel, thereby enhancing the richness of the sample set; in an optional implementation manner provided by this embodiment, in the process of clustering the initial user features of each user in the multiple resource purchase channels and performing feature screening based on the clustering results, the following operations are performed: Calculating, according to the initial user features, the category probability of each user for each user category and the intermediate user features of each user in each user category; Feature screening is performed in the intermediate user features according to the category probability to obtain the user features.

[0022] Wherein, each user category may be a plurality of user categories obtained by adaptive clustering; the initial user features of each user in multiple resource purchasing channels may be obtained by extracting features from the user data of each user in multiple resource purchasing channels, and the user data of each user in multiple resource purchasing channels may include user attribute data, user resource behavior data, user purchase behavior data and / or user resource data of each user in multiple resource purchasing channels. In the case where the user data of each user in multiple resource purchasing channels is multiple data, feature extraction may be performed on each user data of each user in multiple resource purchasing channels, and each user feature may be spliced ​​to obtain the initial user features of each user in multiple resource purchasing channels; similar to the user data of each user in the target resource purchasing channel mentioned above, the user data of each user in multiple resource purchasing channels here is only the user data of multiple resource purchasing channels. The user features may be target user features corresponding to the category probability whose sorting position of the category probability in the intermediate user features is before the preset position.

[0023] For example, the initial user characteristics of each user in multiple resource purchasing channels are , based on the initial user features, calculate the category probability vector of each user belonging to each user category in n categories And the user feature matrix composed of the intermediate user features of each user in each user category (n m-dimensional vectors) , the intermediate user features of each user in each user category are all m-dimensional vectors, and the user features are obtained by feature screening in the intermediate user features according to the category probability vector.

[0024] On this basis, in order to automatically select features that can better characterize the user from the intermediate user features; in an optional implementation manner provided by this embodiment, in the process of performing feature screening in the intermediate user features according to the category probability to obtain the user features, the target category probability in the category probability of each user for each user category can be determined, and the target user feature corresponding to the target category probability in the intermediate user features is used as the user feature, and the target category probability can be the category probability whose sorting position is before the preset position in the category probability; specifically, the following operations can be performed: Probability conversion is performed on the category probability of each user for each user category to obtain the converted category probability of each user for each user category; According to the conversion category probability, the target user features of the respective users in the target user category are screened out from the intermediate user features as the user features.

[0025] The conversion category probability may be a conversion category probability vector of each user for each user category, and the conversion category probability vector may be a onehot vector; optionally, the target user category includes a user category corresponding to a conversion category probability whose sorting position is before a preset position in the conversion category probability of each user.

[0026] Specifically, in the process of performing probability conversion on the category probability of each user for each user category to obtain the conversion category probability of each user for each user category, distribution features can be generated according to the number of user categories, and the category probability features composed of the distribution features and the category probability of each user for each user category can be normalized to obtain normalized probability features; in the process of screening out the target user features of each user in the target user category from the intermediate user features according to the conversion category probability, the product of the normalized probability feature and the intermediate user feature can be calculated as the user feature.

[0027] Among them, the number of user categories refers to the total number of each user category. Using the above example, each user is adaptively clustered into n categories, then the number of user categories is n; the distribution feature can be a uniform distribution feature, specifically, it can be an independent sample showing a uniform distribution U (0, 1); the normalized probability feature can include two normalized probabilities of 0 and 1, and the number of 1s can be one, and the number of 0s can be at least one.

[0028] Using the previous example, we can generate distribution features based on the number of user categories according to the principle of gumbel-softmax (a technique used to process discrete random variables in deep learning). , by calculating Convert the category probability features composed of the category probabilities of each user category into normalized probability features , i=1,2…n; calculate the product of the normalized probability feature and the intermediate user feature = ;in, represents the temperature parameter, which can be adjusted; for example, when the category probability of each user for each user category The probability of the category whose sorting position is before the preset position is , then the normalized probability feature Can approach , so as to improve the discrimination of the category probability of each user for each user category.

[0029] In actual applications, when the target resource purchase channel has just been released or has not been released for a long time, the user data of the target resource purchase channel may be relatively scarce, and it may not be possible to train based on the user data of the target resource purchase channel to obtain the conversion prediction model of the target resource purchase channel. However, the target resource purchase channel still has the need to push the resource amount. In view of this, in order to push the resource amount in the target resource purchase channel in a timely manner and improve the user retention rate of the target resource purchase channel, when it is detected that the release time of the target resource purchase channel is less than the preset time threshold or the conversion prediction model of the target resource purchase channel does not exist, the resource amount can be pushed in the resource purchase process of the target resource purchase channel through the global conversion prediction model; when the conversion prediction model of the target resource purchase channel exists, the resource amount can be pushed in the resource purchase process of the target resource purchase channel through the conversion prediction model; Specifically, in the process of pushing resource amounts in the resource purchase process of the target resource purchase channel through the global conversion prediction model, user data of the user in multiple resource purchase channels can be obtained, and the user data can be input into the global conversion prediction model for conversion prediction processing to obtain the conversion index of the user's pushed resource amounts for multiple resource purchase channels, and the target pushed resource amount is determined from the pushed resource amounts of multiple resource purchase channels according to the conversion index, and the resource amount is pushed in the resource purchase process of the target resource purchase channel based on the target pushed resource amount; that is, in the case where the conversion prediction model of the target resource purchase channel has not been trained, the resource amount can be pushed in the target resource purchase channel based on the conversion index of the user's pushed resource amount for multiple resource purchase channels predicted by the global conversion prediction model, and the resource amount can be pushed in time when the target resource purchase channel is just released, thereby enhancing the marketing breadth of the target resource purchase channel; In the process of pushing resource amounts in the target resource purchasing channel based on the conversion indicators of the user's pushed resource amounts for multiple resource purchasing channels predicted by the global conversion prediction model, a specific pushed resource amount can be determined in each pushed resource amount according to the conversion indicators of the user for each pushed resource amount for multiple resource purchasing channels, and the resource amount is pushed in the resource purchasing process of the target resource purchasing channel based on the specific pushed resource amount; in the process of determining the specific pushed resource amount in each pushed resource amount according to the conversion indicators of the user for each pushed resource amount for multiple resource purchasing channels, in order to improve the determination efficiency, the conversion indicators of the user for each pushed resource amount for multiple resource purchasing channels can also be input into the push amount identification determination algorithm to determine the push amount identification, obtain the push amount identification, and use the push resource amount corresponding to the push amount identification as the specific push resource amount; the push amount identification can be a push amount gear; For example, the push amount identification determination algorithm for multiple resource purchase channels includes:

[0030]

[0031] in, Represents the conversion indicators of the amount of pushed resources 1 to n for users from multiple resource purchasing channels; Represents the conversion indicator threshold corresponding to the first push resource amount; Represents the push amount identifier.

[0032] In an optional implementation manner provided in this embodiment, the global conversion prediction model is obtained by training in the following manner: Inputting the second sample set into the second to-be-trained model for conversion prediction processing to obtain conversion indicators of the amount of pushed resources of each user for the multiple resource purchase channels; Calculating the loss according to the conversion index of the amount of pushed resources by each user for the multiple resource purchasing channels, the number of samples in the second sample set, and the sample label corresponding to each user sample in the second sample set; The parameters of the second model to be trained are adjusted according to the calculated loss value to obtain the global conversion prediction model.

[0033] Among them, the pushed resource amounts of the multiple resource purchasing channels can be one or more, and the conversion indicators of the pushed resource amounts of each user for the multiple resource purchasing channels include the conversion rate of the pushed resource amounts of each user for the multiple resource purchasing channels, that is, after the pushed resource amounts of multiple resource purchasing channels are pushed to each user, the probability that each user purchases resources based on the pushed resource amounts; the number of samples in the second sample set refers to the number of samples of user samples in the second sample set. For example, the second sample set includes user data of a users in multiple resource purchasing channels, then the number of samples in the second sample set is a; the sample labels corresponding to each user sample in the second sample set represent whether the user corresponding to each user sample purchases resources in multiple resource purchasing channels.

[0034] Optionally, the second sample set is constructed based on the user data of each user in multiple resource purchase channels, that is, each user sample in the second sample set can be composed of the user data of each user in multiple resource purchase channels; optionally, the second model to be trained includes a feature extraction network, a clustering network and a conversion prediction network; the feature extraction network can be used to extract features from each user sample in the second sample set to obtain the initial user features of each user in multiple resource purchase channels; the clustering network can be used to cluster the initial user features of each user in multiple resource purchase channels, and perform feature screening based on the clustering results to obtain the user features of each user in multiple resource purchase channels, and the clustering network here can be an adaptive clustering network; the conversion prediction network can be used to perform conversion prediction processing based on the user features of each user in multiple resource purchase channels to obtain the conversion indicators of the pushed resource amounts of each user for multiple resource purchase channels; the conversion prediction network in the second model to be trained can be composed of multiple conversion prediction algorithms, the number of multiple conversion prediction algorithms can be the same as the number of pushed resource amounts of multiple resource purchase channels, and the multiple conversion prediction algorithms can be multiple activation functions, such as multiple sigmoid functions; It should be noted that the above-mentioned process of clustering the initial user features of each user in multiple resource purchasing channels and performing feature screening based on the clustering results is similar to the process of clustering the initial user features of each user in multiple resource purchasing channels by the clustering network in the global conversion prediction model and performing feature screening based on the clustering results, which will not be repeated here in this embodiment; in the process of the conversion prediction network performing conversion prediction processing based on the user features of each user in multiple resource purchasing channels to obtain the conversion indicators of the pushed resource amounts of each user for multiple resource purchasing channels, the user features of each user in multiple resource purchasing channels can be respectively input into the conversion prediction algorithms corresponding to each pushed resource amount of the multiple resource purchasing channels for conversion prediction, so as to obtain the conversion indicators of each user for each pushed resource amount of the multiple resource purchasing channels.

[0035] For example, the conversion prediction algorithms corresponding to the push resource amounts of multiple resource purchasing channels include:

[0036] in, Represents the conversion indicator of the amount of pushed resources i by user j for multiple resource purchase channels; Represents the user characteristics of user j in multiple resource purchasing channels; Represents the bias of the conversion prediction algorithm corresponding to the push resource amount i; Represents the weight of the conversion prediction algorithm corresponding to the push resource amount i. Both bias and weight are parameters of the conversion prediction algorithm.

[0037] In actual applications, in resource purchase scenarios, there are often characteristics of high pushed resource amounts and low conversion indicators. That is, in resource purchase scenarios, if push is based on a higher pushed resource amount, the user's conversion indicator is lower, and if push is based on a lower pushed resource amount, the user's conversion indicator is higher. In order to ensure that the global conversion prediction model of multiple resource purchase channels meets the characteristics of high pushed resource amounts and low conversion indicators when pushing resource amounts in resource purchase scenarios, this embodiment introduces soft ordered constraints for the construction of the loss function during the model training process for the second model to be trained; Specifically, in the process of calculating the loss based on the conversion index of the amount of pushed resources of each user for the multiple resource purchase channels, the number of samples in the second sample set, and the sample label corresponding to each user sample in the second sample set, the following operations may be performed: Calculate the comparison loss according to the conversion index of the amount of pushed resources of each user for multiple resource purchase channels, the number of samples in the second sample set, and the sample label corresponding to each user sample; The constraint loss is calculated according to the conversion index of the amount of pushed resources of each user for multiple resource purchasing channels and the number of samples in the second sample set.

[0038] Among them, the contrast training loss and the constraint loss can be used as the loss value obtained by loss calculation.

[0039] In the above process of calculating the constraint loss based on the conversion index of the amount of pushed resources of each user for multiple resource purchase channels and the number of samples in the second sample set, the indicator difference can be calculated based on the conversion index of the previous amount of pushed resources and the conversion index of the next amount of pushed resources of each user; the indicator difference is converted based on the difference attribute of the indicator difference to obtain the conversion result, and the constraint loss is calculated based on the conversion result and the number of samples in the second sample set, and the constraint loss is used to evaluate whether the model meets the characteristics of high amount of pushed resources and low conversion index; the previous amount of pushed resources here refers to the previous amount of pushed resources among the amounts of pushed resources of multiple resource purchase channels, and the next amount of pushed resources refers to the next amount of pushed resources of the previous amount of pushed resources of multiple resource purchase channels; the conversion of the indicator difference based on the difference attribute of the indicator difference to obtain the conversion result can also be replaced by converting the indicator difference through an activation function to obtain the conversion result.

[0040] Referring to the above-mentioned training method for training the second model to be trained based on the second sample set, the above-mentioned training process is repeated to train the second model to be trained until the second model to be trained converges to complete the training and obtain the global conversion prediction model.

[0041] For example, in the process of training the global conversion prediction model from the second model to be trained, the loss function constructed is:

[0042]

[0043]

[0044] in, Represents the weight of contrast loss; Represents the weight of the constraint loss; stands for contrast loss; represents the constraint loss; N represents the number of samples in the second sample set; Represents the conversion indicator of the amount of pushed resources i by user j for multiple resource purchase channels; Represents the sample label of user j for the pushed resource amount i; Represents the amount of pushed resources for user j for multiple resource purchasing channels ( ) conversion indicator; the push resource amount i can be higher than the push resource amount ( ); in this way, the constraint loss is used to ensure that the model learning of the second model to be trained conforms to the characteristics of high push resource amount and low conversion index, thereby ensuring that the global conversion prediction model obtained through training has the characteristics of high push resource amount and low conversion index.

[0045] The model structure of the global conversion prediction model is the same as the model structure of the second model to be trained, and the functions of the feature extraction network, clustering network and conversion prediction network are the same; the input of the global conversion prediction model can be the user data of the user in multiple resource purchasing channels, or the user data of the user in a specific resource purchasing channel.

[0046] Step S204: input the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing to obtain a conversion index of the amount of pushed resources of each user for the target resource purchase channel.

[0047] The sample set is constructed by obtaining the user characteristics of each user in multiple resource purchasing channels and the user data of the target resource purchasing channel. In this step, the sample set is input into the to-be-trained model of the target resource purchasing channel for conversion prediction processing to obtain the conversion index of the amount of pushed resources for each user for the target resource purchasing channel.

[0048] The pushed resource amount of the target resource purchase channel described in this embodiment includes one or more pushed resource amounts pre-set for the target resource purchase channel, and the pushed resource amount refers to the resource amount pushed to the user during the resource purchase process of the target resource purchase channel, such as the pushed resource amount is x1, x2, x3..., that is, the pushed resource level; the conversion index of each user for the pushed resource amount of the target resource purchase channel includes the conversion rate of each user for the pushed resource amount of the target resource purchase channel, that is, during the resource purchase process of each user on the target resource purchase channel, if the pushed resource amount of the target resource purchase channel is pushed to each user, the conversion probability of each user actually purchasing resources based on the pushed resource amount.

[0049] In the specific implementation, in order to improve the comprehensiveness of the training process of the to-be-trained model of the target resource purchase channel and prevent the lack of data of the target resource purchase channel, resulting in insufficient accuracy of the conversion prediction model of the target resource purchase channel obtained through training; in an optional implementation manner provided by this embodiment, the to-be-trained model performs conversion prediction processing in the following manner: Perform feature extraction on user data of the target resource purchasing channel of the users in the respective user samples to obtain initial user features of the respective users in the target resource purchasing channel; Performing attention calculation based on the user characteristics and the initial user characteristics of each user in the target resource purchase channel to obtain the attention characteristics of each user; The conversion index is obtained by performing conversion index prediction according to the user characteristics and the attention characteristics.

[0050] Among them, the user data of users in each user sample in the target resource purchase channel can be multiple types of data, such as any two or more of the four types of user attribute data, user resource behavior data, user purchase behavior data, and user resource data. In this case, in the process of performing feature extraction on the user data of users in each user sample in the target resource purchase channel to obtain the initial user features of each user in the target resource purchase channel, feature extraction can be performed on multiple user data of users in each user sample in the target resource purchase channel, and the obtained multiple extracted features can be spliced ​​to obtain the initial user features of each user in the target resource purchase channel; assuming that the initial user features of each user in the target resource purchase channel are .

[0051] It should be noted that the model to be trained may also include a feature extraction network, an attention calculation network and a conversion prediction network; the feature extraction network can be used to extract features from multiple user data of users in each user sample in the target resource purchase channel, and concatenate the multiple extracted features obtained to obtain the initial user features of each user in the target resource purchase channel; the attention calculation network can be used to perform attention calculation based on user features and the initial user features of each user in the target resource purchase channel to obtain the attention features of each user; the conversion prediction network can be used to predict conversion indicators based on user features and attention features to obtain conversion indicators; the feature extraction network can be an embedding encoding network.

[0052] On this basis, in order to enable the conversion prediction model of the target resource purchase channel to more accurately identify the user's resource push amount, the user characteristics of multiple resource purchase channels and the initial user characteristics of the target resource purchase channel can be integrated through the attention mechanism to achieve model training of the to-be-trained model of the target resource purchase channel assisted by multiple resource purchase channels; in an optional implementation provided by this embodiment, in the above-mentioned process of performing attention calculation based on the user characteristics and the initial user characteristics of each user in the target resource purchase channel to obtain the attention characteristics of each user, the following operations are performed: The initial user features of each user in the target resource purchasing channel are used as queries and the user features are used as keys and values ​​to perform multi-head attention calculations, and the obtained multi-head attention features are used as the attention features.

[0053] For example, let Q = ,K、V= , then multi-head attention calculation is performed to obtain the attention characteristics of each user , the specific process is as follows:

[0054]

[0055]

[0056] On the basis of obtaining the conversion indicator by predicting the conversion indicator according to the user characteristics and the attention characteristics, in order to improve the comprehensiveness and accuracy of the conversion indicator obtained by the conversion indicator prediction, the conversion indicator prediction can still be assisted by the user characteristics of multiple resource purchase channels to improve the effectiveness of the conversion indicator; in an optional implementation provided by this embodiment, in the process of obtaining the conversion indicator by predicting the conversion indicator according to the user characteristics and the attention characteristics, the conversion indicator of each user for each pushed resource amount of the target resource purchase channel can be calculated according to the user characteristics and attention characteristics of each user in multiple resource purchase channels, and the following operations can be specifically performed: Performing full connection processing on the user feature, and performing feature splicing on the full connection processing result and the attention feature to obtain a spliced ​​feature; Conversion index prediction is performed based on the splicing features to obtain conversion indexes of each pushed resource amount of each user for the target resource purchase channel.

[0057] Furthermore, in actual application scenarios, each user has different resource purchasing capabilities and different resource purchasing demands in the target resource purchasing channels, resulting in different pushed resource amounts being used as the basis for pushing resource amounts to different users. To this end, in order to push different pushed resource amounts to users with different resource purchasing capabilities, multiple pushed resource amounts can be set for the target resource purchasing channels to calculate the conversion index of each user for each pushed resource amount. In an optional implementation provided by this embodiment, in the above-mentioned process of predicting the conversion index based on the splicing feature and obtaining the conversion index of each user for each pushed resource amount of the target resource purchasing channel, the following operations are performed: The splicing features are input into the conversion prediction algorithm corresponding to each pushed resource amount of the target resource purchase channel to perform conversion index prediction, and the conversion index of each user for each pushed resource amount is obtained.

[0058] It should be noted that the conversion prediction network of the model to be trained can be used to predict the conversion index based on the splicing features to obtain the conversion index of each user for each pushed resource amount of the target resource purchase channel. The conversion prediction network of the model to be trained can be composed of multiple conversion prediction algorithms. The multiple conversion prediction algorithms can correspond one-to-one to the multiple pushed resource amounts of the target resource purchase channel. The multiple conversion prediction algorithms can be multiple activation functions, such as multiple sigmoid functions. In addition, in the process of predicting the conversion index according to the user characteristics and the attention characteristics to obtain the conversion index, the user characteristics and the attention characteristics of each user in the multiple resource purchase channels can also be input into the conversion prediction algorithms corresponding to the multiple pushed resource amounts of the target resource purchase channel to predict the conversion index, so as to obtain the conversion index of each user for each pushed resource amount of the target resource purchase channel; For example, conversion prediction algorithms include:

[0059] in, Represents the amount of resources pushed by user j for the target resource purchasing channel conversion indicators; Represents the user characteristics of user j in multiple resource purchasing channels; Represents the amount of pushed resources The corresponding bias of the conversion prediction algorithm; Represents the amount of pushed resources The corresponding conversion prediction algorithm weights, biases and weights are all parameters of the conversion prediction algorithm; Represents user characteristics Perform full connection processing; Representatives and The stitching features are obtained by performing feature stitching.

[0060] Step S206, performing loss calculation according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set.

[0061] The above-mentioned conversion prediction processing is performed on the sample set to obtain the conversion index of the amount of pushed resources for each user to purchase the target resource channel. In this step, the loss is calculated by combining the conversion index, the number of samples in the sample set and the sample labels corresponding to each user sample in the sample set.

[0062] In actual applications, in resource purchase scenarios, there are often characteristics of high pushed resource amounts and low conversion indicators, that is, in resource purchase scenarios, if push is based on a higher pushed resource amount, the user's conversion indicator is low, and if push is based on a lower pushed resource amount, the user's conversion indicator is high. In order to ensure that the conversion prediction model of the target resource purchase channel obtained through training meets the characteristics of high pushed resource amounts and low conversion indicators when pushing resource amounts in resource purchase scenarios, and in order to improve the comprehensiveness of loss calculation, soft ordered constraints are introduced in the construction of the loss function; in an optional implementation manner provided by this embodiment, in the process of calculating the loss according to the conversion indicator, the number of samples in the sample set, and the sample labels corresponding to each user sample in the sample set, the following operations are performed: Calculate a first training loss according to the conversion index, the number of samples and the sample labels; A second training loss is calculated according to the conversion index and the number of samples.

[0063] The first training loss may be a contrast loss; and the second training loss may be a constraint loss.

[0064] On this basis, in the second training loss calculation process, a soft ordered constraint is introduced, that is, the loss function obtained by constructing the characteristics of high pushed resource amount and low conversion index is used to perform constrained loss calculation, so that the conversion prediction model of the target resource purchase channel obtained by training can actually predict the conversion index of the user obtained, which can meet the characteristics of high pushed resource amount and low conversion index, so that the conversion prediction model of the target resource purchase channel is more suitable for the actual operation of the user, and the user's usage experience is improved by accurately pushing the resource amount, and the user's retention rate in the target resource purchase channel is improved; in an optional implementation manner provided by this embodiment, in the process of calculating the second training loss according to the conversion index and the number of samples, the following operations are performed: Calculate the index difference according to the conversion index of the previous pushed resource amount and the conversion index of the next pushed resource amount of each user; The indicator difference is converted based on a difference attribute of the indicator difference to obtain a conversion result, and the second training loss is calculated according to the conversion result and the number of samples.

[0065] Among them, the previous pushed resource amount refers to the previous pushed resource amount among multiple pushed resource amounts of the target resource purchasing channel, and the subsequent pushed resource amount refers to the subsequent pushed resource amount among multiple pushed resource amounts; the previous pushed resource amount and the subsequent pushed resource amount can be adjacent.

[0066] Specifically, in the process of converting the indicator difference based on the difference attribute of the indicator difference to obtain the conversion result, if the difference attribute of the indicator difference is a negative number, 0 is used as the conversion result; if the difference attribute of the indicator difference is a non-negative number, the indicator difference is used as the conversion result; in the process of calculating the second training loss according to the conversion result and the number of samples, the sum of multiple conversion results can be calculated, and the ratio of the sum to the number of samples is calculated as the second training loss.

[0067] In addition, in the process of calculating the second training loss based on the conversion index and the number of samples, the conversion index of each user's previous pushed resource amount and the conversion index of the next pushed resource amount can also be input into the activation function for conversion processing to obtain each conversion result, calculate the sum of each conversion result, and calculate the ratio of the sum to the number of samples as the second training loss; the activation function here can be a ReLU activation function.

[0068] For example, the loss functions used in the loss calculation process include:

[0069]

[0070]

[0071] in, represents the weight of the first training loss; represents the weight of the second training loss; represents the first training loss; represents the second training loss; N represents the number of samples in the sample set; Represents the amount of resources pushed by user j for the target resource purchasing channel conversion indicators; Represents the amount of pushed resources for user j Sample labels of Represents the amount of resources pushed by user j for the target resource purchasing channel ( ) conversion indicators; the amount of pushed resources Can be higher than the amount of pushed resources ( ); in this way, the second training loss is used to ensure that the model learning of the model to be trained meets the characteristics of high push resource amount and low conversion index, and then ensure that the conversion prediction model obtained by training has the characteristics of high push resource amount and low conversion index.

[0072] Step S208, adjusting the parameters of the model to be trained based on the calculated loss to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchase channel.

[0073] The above-mentioned loss calculation is performed based on the conversion index of each user's pushed resource amount for the target resource purchase channel, the number of samples in the sample set, and the sample labels corresponding to each user sample in the sample set. In this step, the parameters of the to-be-trained model are adjusted based on the calculated loss to obtain the conversion prediction model of the target resource purchase channel for pushing resource amounts during the resource purchase process of the target resource purchase channel.

[0074] In specific implementation, the parameters of the model to be trained can be adjusted based on the calculated losses, and the training process can be repeated with reference to the training method of training the model to be trained of the target resource purchase channel based on the sample set, that is, the conversion prediction processing through the model to be trained, the loss calculation and the parameter adjustment can be repeated until the model to be trained converges, the model training can be terminated, and the conversion prediction model of the target resource purchase channel can be obtained; the conversion prediction model can push the resource amount during the resource purchase process of the target resource purchase channel.

[0075] In the specific execution process, after the conversion prediction model of the target resource purchase channel is obtained through the above training, the conversion prediction model can be used to predict the conversion indicators of users of the target resource purchase channel for each pushed resource amount, and the target pushed resource amount is determined in each pushed resource amount according to the conversion indicator, and the resource amount is pushed based on the target pushed resource amount, so as to improve the user's perception of the resource purchase process of the target resource purchase channel, improve the interaction with the user in the resource purchase process, and thus improve the user's conversion rate by accurately pushing the prediction of the resource amount; in an optional implementation manner provided by this embodiment, in the process of pushing the resource amount in the resource purchase process of the target resource purchase channel, the following operations are performed: Inputting the user characteristics of the user in the multiple resource purchasing channels and the user data of the target resource purchasing channel into the conversion prediction model for conversion prediction processing, and obtaining the conversion index of each pushed resource amount of the user for the target resource purchasing channel; determining a target push resource amount from among the push resource amounts according to a conversion index of the user for each push resource amount; Based on the target push resource amount, the resource amount is pushed during the resource purchase process of the target resource purchase channel.

[0076] On this basis, in order to accurately push resource amounts to users, in an optional implementation manner provided by this embodiment, in the process of determining the target pushed resource amount from the pushed resource amounts according to the conversion index of the user for the pushed resource amounts, the following operations are performed: Comparing the conversion index of the user for each pushed resource amount with the conversion index threshold corresponding to each pushed resource amount; Based on the comparison result, the conversion index of the user for each push resource amount is converted to obtain a conversion value, and the target push resource amount is determined from the push resource amounts according to the conversion value.

[0077] Specifically, in the process of converting the conversion indicators of the user for each pushed resource amount based on the comparison result to obtain the conversion value, if the comparison result is that the conversion indicator is greater than the conversion indicator threshold, 1 can be used as the conversion value, if the comparison result is that the conversion indicator is less than or equal to the conversion indicator threshold, then 0 can be used as the conversion value, and a conversion value with a value of 1 can be determined among the conversion values ​​corresponding to each comparison result, and a conversion value with a value of 1 corresponding to the corresponding push resource amount as a preset push resource amount can be determined, and the push resource amount corresponding to the conversion value is used as the target push resource amount.

[0078] Further, in an optional implementation manner provided by this embodiment, in the process of determining the target push resource amount from the push resource amounts according to the conversion value, the following operations are performed: Determine a target conversion value whose sorting position in the conversion values ​​is before the preset position; The push resource amount whose sorting position is before the specific position among the push resource amounts corresponding to the target conversion value is used as the target push resource amount.

[0079] Among them, the preset position can be 2 and the specific position can also be 2.

[0080] For example, the conversion values ​​obtained by converting the user's conversion indicators for each push resource amount are [0, 1, 0, 0, 1, 1, 1, 0, 0] respectively, and the target conversion value whose sorting position in the conversion value is before the preset position 2 is determined to be 1, 1, 1, 1, and the push resource amount whose sorting position in the push resource amount corresponding to the target conversion value is before the preset position is taken as the target push resource amount, that is, the push resource amount corresponding to the 7th conversion value 1 in the conversion value is taken as the target push resource amount.

[0081] In addition, in the process of determining the target push resource amount from the push resource amounts according to the conversion index of the user for the push resource amounts, in order to improve the determination efficiency, the user's conversion index for the push resource amounts can also be input into the push amount identification determination algorithm to determine the push amount identification, obtain the push amount identification, and use the push resource amount corresponding to the push amount identification as the target push resource amount; the push amount identification can be a push amount gear; For example, the push amount identification determination algorithm includes:

[0082]

[0083] in, Represents the conversion indicators of the amount of each pushed resource 1 to n for the target resource purchase channel of the user; Represents the conversion indicator threshold corresponding to the first push resource amount; Represents the push amount identifier.

[0084] It should be supplemented that each optional implementation method in steps S202 to S208 provided in this embodiment can be executed independently as needed, or can be combined and referenced with each other. At the same time, each specific execution step in each optional implementation method can also be executed independently, and this embodiment does not make specific limitations.

[0085] In summary, the one or more conversion prediction model training methods provided in the present embodiment first obtain the user features of each user in multiple resource purchase channels and the user data of the target resource purchase channel to construct a sample set; the user features are clustered by the clustering network in the global conversion prediction model to the initial user features of each user in multiple resource purchase channels, and the features are screened based on the clustering results; secondly, the sample set is input into the to-be-trained model of the target resource purchase channel to extract the features of the user data of the user in the target resource purchase channel in each user sample in the sample set, to obtain the initial user features of each user in the target resource purchase channel, to calculate the attention based on the user features and the initial user features of each user in the target resource purchase channel, to obtain the attention features of each user, to predict the conversion index based on the user features and the attention features, to obtain the output conversion index of the amount of pushed resources of each user for the target resource purchase channel; Secondly, a first training loss is calculated according to the conversion index, the number of samples in the sample set and the sample labels corresponding to each user sample in the sample set, and a second training loss is calculated according to the conversion index and the number of samples. The parameters of the model to be trained are adjusted based on the first training loss and the second training loss to obtain a conversion prediction model for pushing resource amounts in the resource purchase process of the target resource purchase channel. In this way, the user characteristics of each user in multiple resource purchase channels obtained by processing the global conversion prediction model are introduced during the training process of the model to be trained in the target resource purchase channel, so that the sample set is richer, and not only the data of the target resource purchase channel but also the data of multiple resource purchase channels exists, thereby improving the richness of the sample set. The user characteristics of multiple resource purchase channels are used to assist in the training of the model to be trained, thereby improving the accuracy of model training, and improving the effectiveness and accuracy of the conversion prediction model. At the same time, the second training loss is introduced to make the conversion prediction model meet the characteristics of high pushed resource amounts and low conversion indicators in the actual processing process.

[0086] The following uses the application of a conversion prediction model training method provided by this embodiment in a model training scenario as an example to further illustrate the conversion prediction model training method provided by this embodiment. Figure 3 , a conversion prediction model training method applied to a model training scenario specifically includes the following steps.

[0087] Step S302: Acquire the user characteristics of each user in multiple resource purchasing channels and the user data of the target resource purchasing channel to construct a sample set.

[0088] Optionally, user features are obtained by clustering the initial user features of each user in multiple resource purchasing channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results.

[0089] Step S304: input the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing to obtain the conversion rate of the pushed resource amount of each user for the target resource purchase channel.

[0090] Among them, conversion prediction processing may include: extracting features from user data of users in the target resource purchasing channel in each user sample in the sample set, obtaining the initial user features of each user in the target resource purchasing channel, performing attention calculation based on user features and the initial user features of each user in the target resource purchasing channel, obtaining the attention features of each user, predicting conversion indicators based on user features and attention features, and obtaining conversion indicators.

[0091] Step S306, calculating the first training loss according to the conversion rate, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set.

[0092] Step S308, calculate the conversion rate difference according to the conversion rate of the previous pushed resource amount and the conversion rate of the next pushed resource amount of each user, convert the conversion rate difference based on the difference attribute of the conversion rate difference to obtain the conversion result, and calculate the second training loss according to the conversion result and the number of samples.

[0093] Step S310, adjusting parameters of the model to be trained based on the first training loss and the second training loss to obtain a conversion prediction model for pushing resource amounts during a resource purchase process of a target resource purchase channel.

[0094] Among them, pushing resource amounts during the resource purchase process of the target resource purchase channel may include: Input the user characteristics of the user in multiple resource purchasing channels and the user data of the target resource purchasing channel into the conversion prediction model for conversion prediction processing, and obtain the conversion rate of each pushed resource amount of the user for the target resource purchasing channel; Determining a target push resource amount among each push resource amount according to the user's conversion rate for each push resource amount; Based on the target push resource amount, the resource amount is pushed during the resource purchase process of the target resource purchase channel.

[0095] It should be noted that any one of steps S302 to S310 or any combination of multiple steps can be replaced by the corresponding technical means provided in the above steps S202 to S208 according to the needs of implementation and deployment, and will not be described one by one here.

[0096] It should be supplemented that the above-mentioned embodiment provides relevant content of conversion prediction model training, and the following embodiment provides relevant content of conversion prediction model processing in actual application. Some corresponding content in the two embodiments can be read for reference, such as the training process of the conversion prediction model in the above-mentioned embodiment, the actual prediction process of the conversion prediction model, and the training process of the global conversion prediction model, etc. For ease of understanding, they can be read for reference in the process of reading the following embodiments. Similarly, in the process of reading the above embodiments, the relevant content in the following embodiments can also be referred to.

[0097] One or more embodiments of a push processing method provided in this specification are as follows: Reference Figure 4 The push processing method provided in this embodiment specifically includes steps S402 to S406.

[0098] Step S402, obtaining user characteristics of the user in multiple resource purchasing channels and user data of the target resource purchasing channel.

[0099] Each resource purchasing channel in the multiple resource purchasing channels described in this embodiment refers to each channel for purchasing resources. Each resource purchasing channel can be used to purchase the same resource, which can be a financial resource or a financial project. The financial resource or financial project can specifically be a fund, etc. This embodiment does not specifically limit the type of financial resource or financial project. For this purpose, the multiple resource purchasing channels can be multiple channels for purchasing the same financial resource or the same financial project. The multiple resource purchasing channels may include the target resource purchasing channel, or may not include the target resource purchasing channel.

[0100] The user data refers to the user data of the user in the target resource purchasing channel; optionally, the user data includes at least one of the following: user attribute data, user resource behavior data, user purchase behavior data, user resource data; wherein, the user attribute data may include age, occupation and / or male or female data, in addition, the user attribute data may also be other data; the user resource behavior data includes the user's behavior data in the target resource purchasing channel, specifically, it may be the user's activity in the target resource purchasing channel; the user purchase behavior data includes the user's purchase data of resources purchased in the target resource purchasing channel, such as the purchase amount and purchase frequency of resources purchased by the user in the target resource purchasing channel; the user resource data includes the user's holding resource data.

[0101] Optionally, the user features are clustered by a clustering network in a global conversion prediction model to the initial user features of the user in the multiple resource purchase channels, and feature screening is performed based on the clustering results; the clustering process here can be an adaptive clustering process. The training process of the global conversion prediction model in this embodiment is involved in the relevant content provided in the above embodiment, which is referred to and will not be repeated here.

[0102] Among them, the global conversion prediction model refers to a global conversion prediction model obtained by training the second model to be trained based on a second sample set constructed based on user data of each user in multiple resource purchasing channels; the global conversion prediction model can be used to predict the conversion indicators of the amount of pushed resources for users for multiple resource purchasing channels.

[0103] In actual applications, the user data of the target resource purchase channel may be relatively single and limited. To this end, in order to improve the accuracy of conversion prediction, a clustering network of a global conversion prediction model may be introduced to perform clustering processing and feature screening to obtain user features; specifically, in the process of clustering the initial user features of the user in the multiple resource purchase channels and performing feature screening based on the clustering results, the following operations are performed: According to the initial user features, the category probability of the user for each user category and the intermediate user features of the user in each user category are calculated; Feature screening is performed in the intermediate user features according to the category probability to obtain user features.

[0104] Wherein, each user category may be a plurality of user categories obtained by adaptive clustering; the initial user features of the user in the plurality of resource purchasing channels may be obtained by extracting features from the user data of the user in the plurality of resource purchasing channels, and the user data of the user in the plurality of resource purchasing channels may include the user attribute data, user resource behavior data, user purchase behavior data and / or user resource data of the user in the plurality of resource purchasing channels. In the case that the user data of the user in the plurality of resource purchasing channels is a plurality of data, features may be extracted from each user data of the user in the plurality of resource purchasing channels, and each user feature may be spliced ​​to obtain the initial user features of the user in the plurality of resource purchasing channels. The user features may be target user features corresponding to the category probability whose sorting position of the category probability in the intermediate user features is before the preset position.

[0105] For example, the initial user characteristics of users in multiple resource purchasing channels are , based on the initial user features, calculate the category probability vector of the user belonging to each user category in n categories And the user feature matrix (n m-dimensional vectors) composed of the intermediate user features of the user in each user category , the intermediate user features of the user in each user category are all m-dimensional vectors, and the user features are obtained by feature screening in the intermediate user features according to the category probability vector.

[0106] On this basis, in order to automatically select features that better represent the user from the intermediate user features; in the process of performing feature selection in the intermediate user features according to the category probability to obtain the user features, the target category probability in the category probability of each user for each user category can be determined, and the target user features corresponding to the target category probability in the intermediate user features are used as user features. The target category probability can be the category probability whose sorting position is before the preset position in the category probability; specifically, the following operations can be performed: Probability conversion is performed on the category probability of the user for each user category to obtain the converted category probability of the user for each user category; According to the conversion category probability, the target user features of the user in the target user category are screened out from the intermediate user features as user features.

[0107] Among them, the conversion category probability can be a conversion category probability vector of the user for each user category, and the conversion category probability vector can be a onehot vector; optionally, the target user category includes the user category corresponding to the conversion category probability whose sorting position in the user's conversion category probability is before the preset position.

[0108] Specifically, in the process of performing probability conversion on the category probability of the user for each user category to obtain the conversion category probability of the user for each user category, distribution features can be generated according to the number of user categories, and the category probability features composed of the distribution features and the category probability of the user for each user category can be normalized to obtain normalized probability features; in the process of screening out the target user features of the user in the target user category from the intermediate user features according to the conversion category probability, the product of the normalized probability feature and the intermediate user feature can be calculated as the user feature.

[0109] Among them, the number of user categories refers to the total number of each user category. Using the above example, users are adaptively clustered into n categories, then the number of user categories is n; the distribution feature can be a uniform distribution feature, specifically, it can be an independent sample showing a uniform distribution U (0, 1); the normalized probability feature can include two normalized probabilities of 0 and 1, and the number of 1 is one, and the number of 0 is at least one.

[0110] Using the previous example, we can generate distribution features based on the number of user categories according to the principle of gumbel-softmax (a technique used to process discrete random variables in deep learning). , by calculating Convert the category probability features composed of the category probabilities of each user category into normalized probability features , i=1,2…n; calculate the product of the normalized probability feature and the intermediate user feature = ;in, Represents the temperature parameter, which can be adjusted; for example, when the user has a category probability for each user category The probability of the category whose sorting position is before the preset position is , then the normalized probability feature Can approach , so as to improve the discrimination of the category probability of each user category.

[0111] Step S404: Input the user characteristics and the user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing to obtain a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel.

[0112] The pushed resource amount of the target resource purchase channel described in this embodiment includes one or more pushed resource amounts pre-set for the target resource purchase channel, and the pushed resource amount refers to the resource amount pushed to the user during the resource purchase process of the target resource purchase channel, such as the pushed resource amount is x1, x2, x3..., that is, the pushed resource level; the user's conversion indicator for the pushed resource amount of the target resource purchase channel includes the user's conversion rate for the pushed resource amount of the target resource purchase channel, that is, during the user's resource purchase process on the target resource purchase channel, if the pushed resource amount of the target resource purchase channel is pushed to the user, the user actually purchases resources based on the pushed resource amount.

[0113] In the specific implementation, in order to improve the accuracy of the conversion index prediction of the target resource purchase channel and prevent the lack of data of the target resource purchase channel, resulting in inaccurate conversion indexes obtained by prediction; in an optional implementation provided by this embodiment, during the conversion prediction process, the following operations are performed: Performing feature extraction on the user data of the user in the target resource purchasing channel to obtain initial user features of the user in the target resource purchasing channel; Performing attention calculation based on the user characteristics and the initial user characteristics of the user in the target resource purchase channel to obtain the user's attention characteristics; The conversion index is obtained by performing conversion index prediction according to the user characteristics and the attention characteristics.

[0114] Among them, the user data of the user in the target resource purchasing channel can be multiple types of data, such as any two or more of the four types of user attribute data, user resource behavior data, user purchase behavior data, and user resource data. In this case, in the process of performing feature extraction on the user data of the user in the target resource purchasing channel to obtain the initial user features of the user in the target resource purchasing channel, feature extraction can be performed on multiple user data of the user in the target resource purchasing channel, and the multiple extracted features obtained can be spliced ​​to obtain the initial user features of the user in the target resource purchasing channel; assuming that the initial user features of the user in the target resource purchasing channel are .

[0115] It should be noted that the conversion prediction model of the target resource purchasing channel may include a feature extraction network, an attention calculation network and a conversion prediction network; the feature extraction network can be used to extract features from multiple user data of users in the target resource purchasing channel, and to concatenate the multiple extracted features obtained to obtain the initial user features of the user in the target resource purchasing channel; the attention calculation network can be used to perform attention calculation based on user features and the initial user features of the user in the target resource purchasing channel to obtain the user's attention features; the conversion prediction network can be used to predict conversion indicators based on user features and attention features to obtain conversion indicators; the feature extraction network can be an embedding encoding network.

[0116] On this basis, in order to enable the conversion prediction model of the target resource purchase channel to more accurately identify the user's resource push amount, the user characteristics of multiple resource purchase channels and the initial user characteristics of the target resource purchase channel can be integrated through the attention mechanism to achieve conversion prediction of the conversion prediction model of the target resource purchase channel assisted by multiple resource purchase channels; in an optional implementation provided by this embodiment, in the process of performing attention calculation based on the user characteristics and the initial user characteristics of the user in the target resource purchase channel to obtain the attention characteristics of the user, the following operations are performed: The initial user features of the user in the target resource purchasing channel are used as a query and the user features are used as keys and values ​​to perform multi-head attention calculations, and the obtained multi-head attention features are used as the attention features.

[0117] For example, let Q = ,K、V= , then multi-head attention calculation is performed to obtain the user's attention features , the specific process is as follows:

[0118]

[0119]

[0120] On the basis of obtaining the conversion indicator by predicting the conversion indicator according to the user characteristics and the attention characteristics, in order to improve the comprehensiveness and accuracy of the conversion indicator obtained by the conversion indicator prediction, the conversion indicator prediction can still be assisted by the user characteristics of multiple resource purchasing channels to improve the effectiveness of the conversion indicator; in an optional implementation manner provided by this embodiment, in the process of obtaining the conversion indicator by predicting the conversion indicator according to the user characteristics and the attention characteristics, the following operations are performed: Performing full connection processing on the user feature, and performing feature splicing on the full connection processing result and the attention feature to obtain a spliced ​​feature; A conversion index prediction is performed based on the splicing features to obtain a conversion index of each pushed resource amount of the user for the target resource purchase channel.

[0121] Furthermore, in actual application scenarios, each user has different resource purchasing capabilities and different resource purchasing demands in the target resource purchasing channels, which results in different pushed resource amounts being used as the basis for pushing resource amounts to different users. To this end, in order to push different pushed resource amounts to users with different resource purchasing capabilities, multiple pushed resource amounts can be set for the target resource purchasing channels to calculate the conversion index of each user for each pushed resource amount. In the above process of predicting the conversion index based on the splicing features and obtaining the conversion index of each pushed resource amount of the user for the target resource purchasing channel, the following operations can be performed: The splicing features are input into the conversion prediction algorithm corresponding to each pushed resource amount of the target resource purchase channel to perform conversion index prediction, and obtain the user's conversion index for each pushed resource amount.

[0122] It should be noted that the conversion prediction network of the conversion prediction model can be used to predict the conversion index based on the splicing features to obtain the user's conversion index for each pushed resource amount of the target resource purchase channel. The conversion prediction network of the conversion prediction model can be composed of multiple conversion prediction algorithms, and the multiple conversion prediction algorithms correspond one-to-one to the multiple pushed resource amounts of the target resource purchase channel. The multiple conversion prediction algorithms can be multiple activation functions, such as multiple sigmoid functions; in addition, in the process of predicting the conversion index according to the user characteristics and the attention characteristics to obtain the conversion index, the user characteristics and the attention characteristics of the user in the multiple resource purchase channels can also be input into the conversion prediction algorithms corresponding to the multiple pushed resource amounts of the target resource purchase channel to predict the conversion index, and obtain the user's conversion index for each pushed resource amount of the target resource purchase channel; Step S406, determining a target push resource amount from the push resource amounts according to the conversion index, and pushing resource amounts in the resource purchase process of the target resource purchase channel based on the target push resource amount.

[0123] In order to accurately push resource amounts to users, in an optional implementation manner provided by this embodiment, in the process of determining a target push resource amount from the push resource amount according to the conversion indicator, the following operations are performed: Comparing the conversion index of the user for the pushed resource amount with the conversion index threshold corresponding to the pushed resource amount; Based on the comparison result, the conversion index of the user for the pushed resource amount is converted to obtain a conversion value, and the target pushed resource amount is determined from the pushed resource amount according to the conversion value.

[0124] Specifically, in the process of converting the conversion indicators of the user for each pushed resource amount based on the comparison result to obtain the conversion value, if the comparison result is that the conversion indicator is greater than the conversion indicator threshold, 1 can be used as the conversion value, if the comparison result is that the conversion indicator is less than or equal to the conversion indicator threshold, then 0 can be used as the conversion value, and a conversion value with a value of 1 can be determined among the conversion values ​​corresponding to each comparison result, and a conversion value with a value of 1 corresponding to the corresponding push resource amount as a preset push resource amount can be determined, and the push resource amount corresponding to the conversion value is used as the target push resource amount.

[0125] On this basis, in an optional implementation manner provided by this embodiment, in the process of determining the target push resource amount from the push resource amount according to the conversion value, the following operations are performed: Determine a target conversion value whose sorting position in the conversion values ​​is before the preset position; The push resource amount whose sorting position is before the specific position among the push resource amounts corresponding to the target conversion value is used as the target push resource amount.

[0126] Among them, the preset position can be 2 and the specific position can also be 2.

[0127] For example, the conversion values ​​obtained by converting the user's conversion indicators for each push resource amount are [0, 1, 0, 0, 1, 1, 1, 0, 0] respectively, and the target conversion value whose sorting position in the conversion value is before the preset position 2 is determined to be 1, 1, 1, 1, and the push resource amount whose sorting position in the push resource amount corresponding to the target conversion value is before a specific position is taken as the target push resource amount, that is, the push resource amount corresponding to the 7th conversion value 1 in the conversion value is taken as the target push resource amount.

[0128] In addition, in the process of determining the target push resource amount from the push resource amounts according to the conversion index of the user for the push resource amounts, in order to improve the determination efficiency, the user's conversion index for the push resource amounts can also be input into the push amount identification determination algorithm to determine the push amount identification, obtain the push amount identification, and use the push resource amount corresponding to the push amount identification as the target push resource amount; the push amount identification can be a push amount gear; For example, the push amount identification determination algorithm includes:

[0129]

[0130] in, Represents the conversion indicators of the amount of each pushed resource 1 to n for the target resource purchase channel of the user; Represents the conversion indicator threshold corresponding to the first push resource amount; Represents the push amount identifier.

[0131] It should be supplemented that each optional implementation method in steps S402 to S406 provided in this embodiment can be executed independently as needed, or can be combined and referenced with each other. At the same time, each specific execution step in each optional implementation method can also be executed independently, and this embodiment does not make specific limitations.

[0132] The following takes the application of a push processing method provided by this embodiment in a resource purchase scenario as an example to further illustrate the push processing method provided by this embodiment. Figure 5 , a push processing method applied to resource purchase scenarios specifically includes the following steps.

[0133] Step S502: obtaining user characteristics of the user in multiple resource purchasing channels and user data of the target resource purchasing channel.

[0134] Optionally, user features are obtained by clustering the initial user features of users in multiple resource purchasing channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results.

[0135] Step S504: input the user characteristics and user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing to obtain the conversion rate of the user's pushed resource amount for the target resource purchase channel.

[0136] Step S506: Compare the user's conversion rate for the pushed resource amount with the conversion rate threshold corresponding to the pushed resource amount.

[0137] Step S508: convert the user's conversion rate for the amount of pushed resources based on the comparison result to obtain a conversion value.

[0138] Step S510, determining a target conversion value whose ranking position in the conversion value is before a preset position, and taking a push resource amount whose ranking position in the push resource amount corresponding to the target conversion value is before a specific position as a target push resource amount.

[0139] Step S512: Pushing the resource amount based on the target push resource amount during the resource purchase process of the target resource purchase channel.

[0140] It should be noted that any one of steps S502 to S512 or any combination of multiple steps can be replaced by the corresponding technical means provided in the above steps S402 to S406 according to the needs of implementation and deployment, and will not be described one by one here.

[0141] An embodiment of a conversion prediction model training device provided in this specification is as follows: In the above-mentioned embodiment, a conversion prediction model training method is provided, and correspondingly, a conversion prediction model training device is also provided, which is described below with reference to the accompanying drawings.

[0142] Reference Figure 6 , which shows a schematic diagram of an embodiment of a conversion prediction model training device provided in this embodiment.

[0143] Since the device embodiment corresponds to the method embodiment, the description is relatively simple, and the relevant parts can refer to the corresponding description of the method embodiment provided above. The device embodiment described below is only illustrative.

[0144] This embodiment provides a conversion prediction model training device, including: The sample set acquisition module 602 is configured to acquire the user features of each user in multiple resource purchase channels and the user data of the target resource purchase channel to construct a sample set; the user features are obtained by clustering the initial user features of each user in the multiple resource purchase channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results; The conversion prediction module 604 is configured to input the sample set into the to-be-trained model of the target resource purchase channel to perform conversion prediction processing, and obtain the conversion index of the amount of pushed resources of each user for the target resource purchase channel; The loss calculation module 606 is configured to perform loss calculation according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set; The parameter adjustment module 608 is configured to adjust the parameters of the model to be trained based on the calculated loss to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchase channel.

[0145] An embodiment of a push processing device provided in this specification is as follows: In the above-mentioned embodiment, a push processing method is provided, and correspondingly, a push processing device is also provided, which is described below with reference to the accompanying drawings.

[0146] Reference Figure 7, which shows a schematic diagram of an embodiment of a push processing device provided in this embodiment.

[0147] Since the device embodiment corresponds to the method embodiment, the description is relatively simple, and the relevant parts can refer to the corresponding description of the method embodiment provided above. The device embodiment described below is only illustrative.

[0148] This embodiment provides a push processing device, including: The data acquisition module 702 is configured to acquire user features of the user in multiple resource purchase channels and user data of the target resource purchase channel; the user features are obtained by clustering the initial user features of the user in the multiple resource purchase channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results; The conversion prediction processing module 704 is configured to input the user characteristics and the user data into the conversion prediction model of the target resource purchase channel to perform conversion prediction processing, and obtain a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel; The resource amount push module 706 is configured to determine a target push resource amount from the push resource amount according to the conversion index, and push the resource amount in the resource purchase process of the target resource purchase channel based on the target push resource amount.

[0149] An embodiment of a conversion prediction model training device provided in this specification is as follows: Corresponding to the above-described conversion prediction model training method, based on the same technical concept, one or more embodiments of this specification further provide a conversion prediction model training device, which is used to execute the above-described conversion prediction model training method. Figure 8 A schematic diagram of the structure of a conversion prediction model training device provided in one or more embodiments of this specification.

[0150] This embodiment provides a conversion prediction model training device, including: like Figure 8As shown, the conversion prediction model training device may have relatively large differences due to different configurations or performances, and may include one or more processors 801 and memory 802, and the memory 802 may store one or more storage applications or data. Among them, the memory 802 may be a short-term storage or a persistent storage. The application stored in the memory 802 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the conversion prediction model training device. Furthermore, the processor 801 may be configured to communicate with the memory 802 to execute a series of computer executable instructions in the memory 802 on the conversion prediction model training device. The conversion prediction model training device may also include one or more power supplies 803, one or more wired or wireless network interfaces 804, one or more input / output interfaces 805, one or more keyboards 806, etc.

[0151] In a specific embodiment, the conversion prediction model training device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the conversion prediction model training device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following: A sample set is constructed by obtaining user features of each user in multiple resource purchasing channels and user data of a target resource purchasing channel; the user features are obtained by clustering the initial user features of each user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and feature screening is performed based on the clustering results; Inputting the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion index of the amount of pushed resources of each user for the target resource purchase channel; Calculating the loss according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set; The parameters of the model to be trained are adjusted based on the calculated loss to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchase channel.

[0152] An embodiment of a push processing device provided in this specification is as follows: Corresponding to the push processing method described above, based on the same technical concept, one or more embodiments of this specification also provide a push processing device, which is used to execute the push processing method provided above. Fig. 9A schematic diagram of the structure of a push processing device provided in one or more embodiments of this specification.

[0153] This embodiment provides a push processing device, including: like Fig. 9 As shown, the push processing device may have relatively large differences due to different configurations or performances, and may include one or more processors 901 and memory 902, and the memory 902 may store one or more storage applications or data. Among them, the memory 902 may be a short-term storage or a persistent storage. The application stored in the memory 902 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the push processing device. Furthermore, the processor 901 may be configured to communicate with the memory 902, and execute a series of computer executable instructions in the memory 902 on the push processing device. The push processing device may also include one or more power supplies 903, one or more wired or wireless network interfaces 904, one or more input / output interfaces 905, one or more keyboards 906, etc.

[0154] In a specific embodiment, the push processing device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the push processing device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following: Obtaining user features of the user in multiple resource purchasing channels and user data of the target resource purchasing channel; the user features are obtained by clustering the initial user features of the user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results; Inputting the user characteristics and the user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel; A target pushed resource amount is determined from the pushed resource amount according to the conversion index, and based on the target pushed resource amount, resource amount is pushed during a resource purchase process in the target resource purchase channel.

[0155] An embodiment of a computer-readable storage medium provided in this specification is as follows: Corresponding to the above-described conversion prediction model training method, based on the same technical concept, one or more embodiments of this specification also provide a computer-readable storage medium.

[0156] The computer-readable storage medium provided in this embodiment is used to store computer-executable instructions, and the computer-executable instructions implement the following steps when executed: A sample set is constructed by obtaining user features of each user in multiple resource purchasing channels and user data of a target resource purchasing channel; the user features are obtained by clustering the initial user features of each user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and feature screening is performed based on the clustering results; Inputting the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion index of the amount of pushed resources of each user for the target resource purchase channel; Calculating the loss according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set; The parameters of the model to be trained are adjusted based on the calculated loss to obtain a conversion prediction model for pushing resource amounts during the resource purchase process of the target resource purchase channel.

[0157] It should be noted that the embodiment of a computer-readable storage medium in this specification and the embodiment of a conversion prediction model training method in this specification are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned corresponding method, and the repeated parts will not be repeated.

[0158] Another computer-readable storage medium embodiment provided in this specification is as follows: Corresponding to the push processing method described above, based on the same technical concept, one or more embodiments of this specification also provide another computer-readable storage medium.

[0159] The computer-readable storage medium provided in this embodiment is used to store computer-executable instructions, and the computer-executable instructions implement the following steps when executed: Obtaining user features of the user in multiple resource purchasing channels and user data of the target resource purchasing channel; the user features are obtained by clustering the initial user features of the user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results; Inputting the user characteristics and the user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel; A target pushed resource amount is determined from the pushed resource amount according to the conversion index, and based on the target pushed resource amount, resource amount is pushed during a resource purchase process in the target resource purchase channel.

[0160] It should be noted that the embodiment of another computer-readable storage medium in this specification and the embodiment of a push processing method in this specification are based on the same inventive concept, so the specific implementation of this embodiment can refer to the implementation of the aforementioned corresponding method, and the repeated parts will not be repeated.

[0161] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referenced to each other. Each embodiment focuses on the differences from other embodiments. For example, the device embodiment, equipment embodiment and computer-readable storage medium embodiment are similar to the method embodiment, so the description is relatively simple. To read the relevant contents in the device embodiment, equipment embodiment and computer-readable storage medium embodiment, please refer to the partial description of the method embodiment.

[0162] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0163] In the 1930s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many of today's method flow improvements can be regarded as direct improvements to hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using hardware entity modules. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask chip manufacturers to design and produce dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0164] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0165] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0166] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0167] It should be understood by those skilled in the art that one or more embodiments of the present specification may be provided as a method, system or computer program product. Therefore, one or more embodiments of the present specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0168] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable test processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable test processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0169] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable test processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0170] These computer program instructions may also be loaded onto a computer or other programmable test processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process described in the flow. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0171] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0172] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0173] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0174] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of features includes not only those features, but also other features not explicitly listed, or also includes features inherent to such process, method, commodity or device. In the absence of further restrictions, features defined by the phrase "including a ..." do not exclude the existence of other identical features in the process, method, commodity or device including the features.

[0175] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0176] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0177] The above description is only an embodiment of this document and is not intended to limit this document. For those skilled in the art, this document may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this document should be included in the scope of the claims of this document.

Claims

1. A conversion prediction model training method, comprising: Get a sample set; The sample set is constructed based on the user characteristics of each user in multiple resource purchasing channels and the user data of each user in the target resource purchasing channel; Inputting the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion index of the amount of pushed resources of each user for the target resource purchase channel; Calculating the loss according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set; The parameters of the model to be trained are adjusted based on the calculated loss to obtain a conversion prediction model.

2. The conversion prediction model training method according to claim 1, wherein the conversion prediction processing comprises: Perform feature extraction on user data of users in the target resource purchasing channel in each user sample to obtain initial user features of each user in the target resource purchasing channel; Performing attention calculation based on the user characteristics and the initial user characteristics of each user in the target resource purchase channel to obtain the attention characteristics of each user; The conversion index is obtained by performing conversion index prediction according to the user characteristics and the attention characteristics.

3. The conversion prediction model training method according to claim 2, wherein the step of performing attention calculation based on the user characteristics and the initial user characteristics of each user in the target resource purchase channel to obtain the attention characteristics of each user comprises: The initial user features of each user in the target resource purchasing channel are used as queries and the user features are used as keys and values ​​to perform multi-head attention calculations, and the obtained multi-head attention features are used as the attention features.

4. According to the conversion prediction model training method of claim 2, the step of performing conversion index prediction based on the user characteristics and the attention characteristics to obtain the conversion index comprises: Performing full connection processing on the user feature, and performing feature splicing on the full connection processing result and the attention feature to obtain a spliced ​​feature; Conversion index prediction is performed based on the splicing features to obtain conversion indexes of each pushed resource amount of each user for the target resource purchase channel.

5. According to the conversion prediction model training method of claim 4, the conversion index prediction based on the splicing features to obtain the conversion index of each pushed resource amount of each user for the target resource purchase channel includes: The splicing features are input into the conversion prediction algorithm corresponding to each pushed resource amount of the target resource purchase channel to perform conversion index prediction, and the conversion index of each user for each pushed resource amount is obtained.

6. The conversion prediction model training method according to claim 1, wherein the loss calculation is performed according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set, including: Calculate a first training loss according to the conversion index, the number of samples and the sample labels; A second training loss is calculated according to the conversion index and the number of samples.

7. The conversion prediction model training method according to claim 6, wherein the step of calculating the second training loss according to the conversion index and the number of samples comprises: Calculate the index difference according to the conversion index of the previous pushed resource amount and the conversion index of the next pushed resource amount of each user; The indicator difference is converted based on a difference attribute of the indicator difference to obtain a conversion result, and the second training loss is calculated according to the conversion result and the number of samples.

8. According to the conversion prediction model training method of claim 1, the user features are obtained by clustering the initial user features of each user in the multiple resource purchasing channels through a clustering network in a global conversion prediction model, and performing feature screening based on the clustering results.

9. The conversion prediction model training method according to claim 8, wherein the clustering of the initial user features of each user in the multiple resource purchase channels and feature screening based on the clustering results comprises: Calculating, according to the initial user features, the category probability of each user for each user category and the intermediate user features of each user in each user category; Feature screening is performed in the intermediate user features according to the category probability.

10. The conversion prediction model training method according to claim 9, wherein the step of screening features in the intermediate user features according to the category probability comprises: Probability conversion is performed on the category probability of each user for each user category to obtain the converted category probability of each user for each user category; According to the conversion category probability, the target user features of the users in the target user category are screened out from the intermediate user features as the user features.

11. The conversion prediction model training method according to claim 1, wherein the conversion prediction model is used to push resource amounts during the resource purchase process of the target resource purchase channel; The pushing of resource amount in the resource purchase process of the target resource purchase channel includes: Inputting the user characteristics of the user in the multiple resource purchasing channels and the user data of the target resource purchasing channel into the conversion prediction model for conversion prediction processing, and obtaining the conversion index of each pushed resource amount of the user for the target resource purchasing channel; determining a target push resource amount from among the push resource amounts according to a conversion index of the user for each push resource amount; Based on the target push resource amount, the resource amount is pushed during the resource purchase process of the target resource purchase channel.

12. The conversion prediction model training method according to claim 11, wherein determining a target push resource amount from each push resource amount according to the user's conversion index for each push resource amount comprises: Comparing the conversion index of the user for each pushed resource amount with the conversion index threshold corresponding to each pushed resource amount; Based on the comparison result, the conversion index of the user for each push resource amount is converted to obtain a conversion value, and the target push resource amount is determined from the push resource amounts according to the conversion value.

13. The conversion prediction model training method according to claim 12, wherein the step of determining the target push resource amount from among the push resource amounts according to the conversion value comprises: Determine a target conversion value whose sorting position in the conversion values ​​is before the preset position; The push resource amount whose sorting position is before the specific position among the push resource amounts corresponding to the target conversion value is used as the target push resource amount.

14. The conversion prediction model training method according to claim 8, wherein the global conversion prediction model is obtained by training in the following manner: Inputting the second sample set into the second to-be-trained model for conversion prediction processing to obtain conversion indicators of the amount of pushed resources of each user for the multiple resource purchase channels; Calculating the loss according to the conversion index of the amount of pushed resources by each user for the multiple resource purchasing channels, the number of samples in the second sample set, and the sample label corresponding to each user sample in the second sample set; The parameters of the second model to be trained are adjusted according to the calculated loss value to obtain the global conversion prediction model.

15. A push processing method, comprising: Obtain user characteristics of users in multiple resource purchasing channels and user data of target resource purchasing channels; Inputting the user characteristics and the user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel; A target pushed resource amount is determined from the pushed resource amount according to the conversion index, and based on the target pushed resource amount, resource amount is pushed during a resource purchase process in the target resource purchase channel.

16. The push processing method according to claim 15, wherein the conversion prediction processing comprises: Performing feature extraction on the user data of the user in the target resource purchasing channel to obtain initial user features of the user in the target resource purchasing channel; Performing attention calculation based on the user characteristics and the initial user characteristics of the user in the target resource purchase channel to obtain the user's attention characteristics; The conversion index is obtained by performing conversion index prediction according to the user characteristics and the attention characteristics.

17. The push processing method according to claim 16, wherein the step of predicting the conversion index according to the user characteristics and the attention characteristics to obtain the conversion index comprises: Performing full connection processing on the user feature, and performing feature splicing on the full connection processing result and the attention feature to obtain a spliced ​​feature; A conversion index prediction is performed based on the splicing features to obtain a conversion index of each pushed resource amount of the user for the target resource purchase channel.

18. The push processing method according to claim 15, wherein determining a target push resource amount from the push resource amount according to the conversion index comprises: Comparing the conversion index of the user for the pushed resource amount with the conversion index threshold corresponding to the pushed resource amount; Based on the comparison result, the conversion index of the user for the pushed resource amount is converted to obtain a conversion value, and the target pushed resource amount is determined from the pushed resource amount according to the conversion value.

19. The push processing method according to claim 18, wherein determining the target push resource amount from the push resource amount according to the conversion value comprises: Determine a target conversion value whose sorting position in the conversion values ​​is before the preset position; The push resource amount whose sorting position is before the specific position among the push resource amounts corresponding to the target conversion value is used as the target push resource amount.

20. A conversion prediction model training device, comprising: A sample set acquisition module is configured to acquire a sample set; The sample set is constructed based on the user characteristics of each user in multiple resource purchasing channels and the user data of each user in the target resource purchasing channel; A conversion prediction module is configured to input the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing, and obtain a conversion indicator of the amount of pushed resources of each user for the target resource purchase channel; A loss calculation module is configured to perform loss calculation according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set; The parameter adjustment module is configured to adjust the parameters of the model to be trained based on the calculated loss to obtain a conversion prediction model.

21. A push processing device, comprising: A data acquisition module is configured to acquire user characteristics of users in multiple resource purchasing channels and user data of a target resource purchasing channel; A conversion prediction processing module is configured to input the user characteristics and the user data into a conversion prediction model of the target resource purchase channel to perform conversion prediction processing, and obtain a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel; The resource amount push module is configured to determine a target push resource amount from the push resource amount according to the conversion index, and push the resource amount in the resource purchase process of the target resource purchase channel based on the target push resource amount.

22. A conversion prediction model training device, comprising: processor; and a memory configured to store computer executable instructions that, when executed, cause the processor to: Obtaining a sample set; the sample set is constructed based on user characteristics of each user in multiple resource purchasing channels and user data of each user in the target resource purchasing channel; Inputting the sample set into the to-be-trained model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion index of the amount of pushed resources of each user for the target resource purchase channel; Calculating the loss according to the conversion index, the number of samples in the sample set, and the sample label corresponding to each user sample in the sample set; The parameters of the model to be trained are adjusted based on the calculated loss to obtain a conversion prediction model.

23. A push processing device, comprising: processor; and a memory configured to store computer executable instructions that, when executed, cause the processor to: Obtain user characteristics of users in multiple resource purchasing channels and user data of target resource purchasing channels; Inputting the user characteristics and the user data into the conversion prediction model of the target resource purchase channel for conversion prediction processing, and obtaining a conversion indicator of the amount of pushed resources of the user for the target resource purchase channel; A target pushed resource amount is determined from the pushed resource amount according to the conversion index, and based on the target pushed resource amount, resource amount is pushed during a resource purchase process in the target resource purchase channel.

24. A computer-readable storage medium for storing computer-executable instructions, wherein the computer-executable instructions implement the steps of the method of claim 1 when executed.

25. A computer-readable storage medium for storing computer-executable instructions, wherein the computer-executable instructions implement the steps of the method of claim 15 when executed.

Images