Object determination method and device, equipment and storage medium

Abstract

The embodiment of the invention provides an object determination method and device, equipment and a storage medium. The method comprises the following steps: acquiring an object information set of candidate objects in a candidate object set; inputting the object information set into a pre-trained gain model, and performing event attribute prediction on a target event associated with the candidate object based on the object information set through the gain model to obtain a first event prediction attribute and a second event prediction attribute corresponding to the candidate object, the gain model is obtained by training based on a sample object and an event attribute correction value, and the event attribute correction value is generated by correcting a historical event attribute of a target event based on a first attribute prediction mean value of the target event associated with the sample object; and determining a target object associated with the target service in the candidate object set according to the first event prediction attribute and the second event prediction attribute corresponding to the candidate object. According to the embodiment of the invention, the influence of the individual difference of the sample objects on model prediction can be eliminated, and the prediction accuracy of the gain model is improved.

Description

Technical Field

[0001] This disclosure relates to computer technology, and more particularly to a method, apparatus, device, and storage medium for determining an object. Background Technology

[0002] With the development of computer technology, online shopping has become a convenient and widely adopted method of shopping. Users can shop online through e-commerce platforms. These platforms frequently launch various promotional activities.

[0003] Currently, gain models can be used to predict target objects that will benefit positively from promotional activities, and then implement the promotional activities on these target objects. Before training the gain model, sample objects need to be randomly divided into experimental and control groups. Training samples are constructed based on the object information of sample objects in the experimental group and the object information of sample objects in the control group. However, the characteristics of the sample objects themselves may significantly influence the event attributes of the target event, thus interfering with the training effect of the gain model. This could prevent the gain model from accurately learning the true causal relationship between the promotional activity and the target object, thereby affecting the accuracy of the gain model. Summary of the Invention

[0004] This disclosure provides a method, apparatus, device, and storage medium for determining an object, which can improve the prediction accuracy of the gain model.

[0005] In a first aspect, embodiments of this disclosure provide a method for determining an object, including:

[0006] Obtain an object information set of candidate objects from the candidate object set, wherein the object information set includes object information, and the object information includes object attributes and object interaction information;

[0007] The object information set is input into a pre-trained gain model. The gain model predicts the event attributes of the target events associated with the candidate objects based on the object information set, thereby obtaining the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects. The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the first attribute prediction mean of the target events associated with the sample objects. The first attribute prediction mean is determined based on the object information set of the sample objects.

[0008] Based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object, the target object associated with the target business in the candidate object set is determined.

[0009] Secondly, embodiments of this disclosure also provide an object determination apparatus, the apparatus comprising:

[0010] The information acquisition module is used to acquire the object information set of the candidate objects in the candidate object set, wherein the object information set includes object information, and the object information includes object attributes and object interaction information;

[0011] The attribute prediction module is used to input the object information set into a pre-trained gain model, and use the gain model to predict the event attributes of the target events associated with the candidate objects based on the object information set, so as to obtain the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects. The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the first attribute prediction mean of the target events associated with the sample objects. The first attribute prediction mean is determined based on the object information set of the sample objects.

[0012] The object determination module is used to determine the target object associated with the target business in the candidate object set based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object.

[0013] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:

[0014] One or more processors;

[0015] Storage device for storing one or more programs.

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the object determination method as described in any embodiment of this disclosure.

[0017] Fourthly, embodiments of this disclosure also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the object determination method as described in any embodiment of this disclosure.

[0018] This disclosure provides a method for determining objects. The method involves inputting a set of object information for candidate objects into a pre-trained gain model. The gain model then predicts the event attributes of target events associated with the candidate objects based on the object information set, obtaining a first event prediction attribute and a second event prediction attribute. Next, based on the first and second event prediction attributes corresponding to the candidate objects, the target objects associated with the target business are determined from the candidate object set. Since the gain model is trained based on sample objects and event attribute correction values, and the event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the mean of the first attribute predictions of the target events associated with the sample objects, the influence of individual differences among sample objects in the random grouping results on the model's predictions can be eliminated. This facilitates the gain model learning the true causal relationship between the event attributes of the target business and the target events, thereby improving the prediction accuracy of the gain model. Attached Figure Description

[0019] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0020] Figure 1 A flowchart illustrating a method for determining an object provided in an embodiment of this disclosure;

[0021] Figure 2 This is a structural block diagram of a modified model provided in an embodiment of the present disclosure;

[0022] Figure 3a This is a schematic diagram of the structure of a gain model provided in an embodiment of the present disclosure;

[0023] Figure 3b A flowchart illustrating a training method for a gain model provided in an embodiment of this disclosure;

[0024] Figure 4 A flowchart illustrating another method for determining an object provided in an embodiment of this disclosure;

[0025] Figure 5 A flowchart illustrating a training method for a classification model provided in an embodiment of this disclosure;

[0026] Figure 6 This is a schematic diagram of the structure of an object determination device provided in an embodiment of the present disclosure;

[0027] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0028] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0029] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0030] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0031] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0032] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0033] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0034] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0035] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0036] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0037] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0038] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0039] Figure 1 This is a flowchart illustrating a method for determining objects according to an embodiment of this disclosure. This embodiment is applicable to situations where target objects associated with a target business are determined, such as scenarios where target objects have a positive feedback relationship with a target business are determined. The target objects include objects such as goods or stores, and the target business may refer to business intervention information to be implemented. This method can be executed by an object determination device, which can be implemented in software and / or hardware, optionally through an electronic device, such as a mobile terminal, PC, or server.

[0040] like Figure 1 As shown, the method includes:

[0041] S110. Obtain the object information set of the candidate objects in the candidate object set.

[0042] The object information set includes object information, which includes object attributes and object interaction information.

[0043] In this embodiment of the disclosure, the candidate object set represents a collection of candidate objects. Candidate objects include products and stores, etc. The object information set is a collection of object information. Object information includes object attributes and object interaction information. If a candidate object is a product, then object attributes represent product attributes such as product category and product usage. Object interaction information represents product interaction information such as the number of times a product has been added to cart, the number of times a product has been shared, the number of times a product has been favorited, and the sales volume of the product. Optionally, object information may also include network status and client attributes associated with the candidate object, etc.

[0044] If it is necessary to identify products that can boost sales by implementing a certain target business, one can obtain product attributes and product interaction information of candidate products over a period of time to form an object information set.

[0045] S120. Input the object information set into a pre-trained gain model, and use the gain model to predict the event attributes of the target events associated with the candidate objects based on the object information set, so as to obtain the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects.

[0046] The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target event based on the first attribute prediction mean of the target event associated with the sample object. The first attribute prediction mean is determined based on the object information set of the sample object. This disclosure does not limit the structure of the gain model; an appropriate model structure can be selected according to the business scenario. Gain model structures include dual-model (T-Learner) and single-model (S-Learner), etc.

[0047] The sample object can refer to an object whose historical event attributes meet preset conditions within a certain period of time. For example, the sample object can be a product or store whose sales volume exceeds a set threshold in the past 30 days. Here, sales volume can be regarded as an event attribute of the product sales event. Different target businesses correspond to different target events, and the target event can be determined according to the target business to be implemented. This disclosure does not specifically limit this.

[0048] The event attribute correction value represents the event attribute value of the target event affected by the target business, after excluding the interference of individual differences of sample objects on the event attributes of the target event. The historical event attributes of the target event can be corrected based on the predicted mean of the first attribute of the target events associated with the sample objects to generate the event attribute correction value. The predicted mean of the first attribute represents the event attribute of the target event affected by the individual differences of the sample objects.

[0049] After dividing the sample objects into the control group and the experimental group, the target service is not applied to the sample objects in the control group, but is applied to the sample objects in the experimental group. The historical event attribute of the target event represents the event attribute value of the target event associated with the sample objects in the control group and the experimental group. For example, if product A is in the experimental group, for the week following the implementation of the promotional activity, the daily sales volume of product A is obtained as the historical event attribute of product A's sales event. If product B is in the control group, for the week following the implementation of the promotional activity in the experimental group, the daily sales volume of product B is obtained as the historical event attribute of product B's sales event.

[0050] In this embodiment of the disclosure, the sample objects are randomly assigned to a control group and an experimental group. Optionally, the number of sample objects in the control group and the experimental group is approximately the same.

[0051] In this embodiment of the disclosure, the training method of the gain model includes:

[0052] The sample objects are divided into a control group and an experimental group. The sample objects in the control group are not associated with the target service, while the sample objects in the experimental group are associated with the target service. An object information set of the sample objects is obtained. The predicted mean of the first attribute is determined based on the object information set. The historical event attributes of the target events associated with the sample objects are corrected based on the predicted mean of the first attribute, resulting in the corrected event attribute value. The gain model is trained based on the object information set of the sample objects and the corrected event attribute value. Optionally, the second time period represents a set time interval after the implementation of the target service. The object information set of the target events associated with the sample objects in the control group within the second time period is obtained, and the grouping identifier of the control group is added to the object information set. The object information set of the target events associated with the sample objects in the experimental group within the second time period is obtained, and the grouping identifier of the experimental group is added to the object information set.

[0053] Taking products as an example, the sample objects are divided into a control group or an experimental group. A targeted promotional activity is implemented for the sample objects in the experimental group, while the original promotional activity is maintained for the sample objects in the control group. In other words, the difference between the control and experimental groups is that the targeted promotional activity is not implemented. Object information for the control and experimental groups is obtained for a period of time after grouping. For example, based on product attributes such as product category and function, as well as interaction information such as the number of times items are added to cart, shared, favorited, and sold, and the mean of the above object information, the predicted mean of the first attribute is determined. The predicted mean of the first attribute represents the sales volume generated by the characteristics of the product itself. Since the historical event attribute of the target event is the sales volume of the product under the dual influence of promotional activities and the characteristics of the product itself, the event attribute correction value is obtained by subtracting the predicted mean of the first attribute from the historical event attribute. The object information corresponding to the sample objects is input into the gain model to be trained, and the training of the gain model is supervised by the event attribute correction value.

[0054] It should be noted that, due to the use of random grouping, the sample objects are divided into control and experimental groups. Taking commodities as an example, some sample objects are inherently best-selling products, while others are inherently slow-moving products. For best-selling products, sales are relatively high even without the target business. The individual differences among sample objects in the random grouping result may interfere with the prediction results of the gain model. For example, the experimental group may consist almost entirely of best-selling products, while the control group may consist almost entirely of slow-moving products. This grouping result will introduce interference information into the training of the gain model, meaning that the model prediction result is affected by the individual differences of the products themselves rather than by the target business. Here, individual differences refer to the differences in the event attributes of the target events associated with the sample objects due to the inherent attributes of the sample objects.

[0055] To eliminate the impact of individual differences among sample objects on the training of the gain model, it is necessary to use a correction model to modify the historical event attributes of the target events associated with the sample objects, thereby improving the accuracy of the gain model training.

[0056] The correction model can be a neural network model that predicts the mean of a first attribute based on the object information set of sample objects. This disclosure does not specifically limit the network structure of the correction model. For example, the correction model may include a first correction module, at least one second correction module, and a third correction module. The first correction module may employ a recurrent neural network. For example, the first correction module may be a long short-term memory artificial neural network, etc. The first correction module is used to process sample information in the object information set of sample objects. For example, the sample information may be sample object attributes and sample object interaction information obtained with a sampling period of days. The second correction module may employ a linear regression model and / or a boosting tree model, etc. The second correction module is used to process the mean of the object information of the sample objects. The third correction module is used to process the output results of the first and second correction modules. The third correction module may employ a boosting tree model, etc.

[0057] The modified model is trained by using the object information set of the sample objects before grouping, so that the modified model can learn the influence of individual differences of the sample objects on the event attributes of the target event. Then, the modified model is used to perform model inference based on the object information set of the sample objects after grouping, and predict the event attribute prediction mean of the target event associated with the sample objects, which is the first attribute prediction mean.

[0058] In some embodiments, the training method for the modified model includes:

[0059] For any day within the first time period, the object information of the sample object on that day is input into the first correction module to obtain the first sample prediction value of the target event associated with the sample object, output by the first correction module. The object information includes the historical event attributes of the target event associated with the sample object. The first time period represents a set time interval before the sample object is divided into the control group or the experimental group. The sample information of the sample object within the first time period can be obtained at a daily granularity as training samples for the first correction module. The sample information of the sample object within the first time period includes the historical event attributes of the target event. The first correction module is trained under supervised supervision using the historical event attributes of the target event.

[0060] Assume the object information of the sample objects is A, B, C, D, and E. The sample information for the first day is represented as (A1, B1, C1, D1, E1). Inputting (A1, B1, C1, D1, E1) into the first correction module, the first sample prediction value output by the first correction module is represented as Y1. The sample information for the second day is represented as (A2, B2, C2, D2, E2). Inputting (A2, B2, C2, D2, E2) into the first correction module, the first sample prediction value output by the first correction module is represented as Y2. And so on, the sample information for the 30th day is represented as (A30, B30, C30, D30, E30). Inputting (A30, B30, C30, D30, E30) into the first correction module, the first sample prediction value output by the first correction module is represented as Y30. Based on the first sample prediction values ​​for each day within the first time period, the average value of the first sample prediction value is determined.

[0061] The mean of the object information of the sample object is input into the at least one second correction module to obtain the second sample prediction value of the target event output by the at least one second correction module.

[0062] The mean of the sample information represents the average value of the object information within the first time period. The average value of object information A can be expressed as (A1 + A2 + ... + A30) / 30. The average value of object information B can be expressed as (B1 + B2 + ... + B30) / 30. The average value of object information C can be expressed as (C1 + C2 + ... + C30) / 30. The average value of object information D can be expressed as (D1 + D2 + ... + D30) / 30. The average value of object information E can be expressed as (E1 + E2 + ... + E30) / 30.

[0063] If the average value of the object information is input into a second correction module, the output of the second correction module will be... This is the predicted value for the second sample.

[0064] If the average value of the object information is input into two second correction modules, the output from the second correction modules will be... and This is represented as the predicted value of the second sample. Multiple second correction modules can be selected to improve the correction performance of the correction model. It should be noted that the more second correction modules used, the more computational resources are required. Therefore, the number of second correction modules should be selected based on model performance and resource consumption.

[0065] The average value of the first sample prediction value within the first time period is determined, and the average value and the second sample prediction value are input into the third correction module to obtain the third sample prediction value of the target event output by the third correction module.

[0066] The first loss function value is calculated by combining the average value of the event attributes of the target event included in the sample information within the first time period and the predicted value of the third sample, and the model parameters of the modified model are adjusted according to the first loss function value.

[0067] For example, the average value of the event attributes associated with the target event of sample object A over the past 30 days is calculated to obtain the true value of the predicted result of the target event associated with the sample object by the corrected model. Since the predicted value of the third sample represents the average value of the event attribute predicted by the corrected model over the past 30 days, it is the predicted value. The first loss function value is calculated based on the true value and the predicted value, and the model parameters of the corrected model are adjusted based on the first loss function value. The above process is iteratively executed until the first loss function value meets the preset training termination condition or the set number of iterations is reached.

[0068] In this embodiment of the disclosure, determining the predicted mean of the first attribute based on the object information set, and correcting the historical event attributes of the target event associated with the sample object based on the predicted mean of the first attribute to obtain the event attribute correction value includes:

[0069] Based on the object information set of the sample objects, determine the mean value of the object information corresponding to the sample objects;

[0070] The object information set and the mean of the object information are input into a pre-trained correction model. The correction model determines the third event prediction attribute of the target event associated with the sample object based on the object information of the sample object, determines the second attribute prediction mean of the target event associated with the sample object based on the mean of the object information, and determines the first attribute prediction mean of the target event associated with the sample object based on the third event prediction attribute and the second attribute prediction mean. The correction model is trained using the object information set of the sample objects before grouping. The third event prediction attribute represents the business attribute prediction result of the target event after the sample objects are divided into the control group or the experimental group.

[0071] Based on the predicted mean of the first attribute and the historical event attributes of the target event, an event attribute correction value is generated.

[0072] By using the modified model based on the object information set of sample objects in a set event segment after the implementation of the target business, the event attributes of the target events that are not affected by the target business can be predicted, that is, the event attributes of the target events that are affected by the individual differences of the sample objects.

[0073] Figure 2 This is a structural block diagram of a modified model provided in an embodiment of this disclosure. Figure 2 As shown, the modified model includes a Long Short-Term Memory (LSTM) artificial neural network 210, a linear regression model 220, a first boosting tree model 230, and a second boosting tree model 240. The input data for the LSTM artificial neural network 210 is the object information of the sample objects within a second time period, such as single-day object information 250. The output data for the LSTM artificial neural network 210 is the third event prediction attribute 260 of the target event. Based on the third event prediction attribute 260, the mean of the third event prediction attribute 280 is determined. The mean of the object information 2100 of the sample objects is determined, and this mean is input into the linear regression model 220 and the first boosting tree model 230 respectively, outputting two second attribute prediction means 270. The values ​​of the two second attribute prediction means 270 may be the same or different. The input data for the second boosting tree model 240 are the third event prediction attribute mean 280 and the two second attribute prediction means 270, and the output data for the second boosting tree model 240 is the first attribute prediction mean 290 of the target event.

[0074] Optionally, the prediction attributes 260 of each third event, the mean of the prediction attributes 280 of the third event, and the two prediction means of the second attributes 270 can be input into the second boosting tree model 240, and the first attribute prediction mean 290 of the target event can be output through the second boosting tree model 240.

[0075] For example, the event attribute correction value is obtained by subtracting the predicted mean of the first attribute from the historical event attributes of the target event associated with the sample object within the second time period. The event attribute correction value represents the event attribute of the target event after eliminating the influence of individual differences among sample objects. The object information set of the sample objects is used as training data, and the event attribute correction value is used as the sample label to train the gain model.

[0076] In this embodiment of the disclosure, training the gain model based on the object information set of the sample object and the event attribute correction value includes:

[0077] The object information set of the sample objects is input into the gain model to be trained. The gain model predicts the event attributes of the target event based on the object information set of the sample objects, obtaining a third event prediction attribute corresponding to the sample object. A prediction loss value is determined based on the historical event attributes of the target event associated with the third event prediction attribute and the sample object. The information distribution difference value of the sample objects in the control group and the experimental group is determined based on the group identifier of the sample objects, wherein the group identifier indicates whether the sample object belongs to the control group or the experimental group. A loss function value is determined based on the prediction loss value and the information distribution difference value, and the gain model is trained based on the loss function value.

[0078] Specifically, based on the grouping identifier in the object information set of the sample objects, the object information set is input into the prediction sub-network corresponding to the gain model to be trained. The prediction sub-network includes a first sub-network and a second sub-network. The first sub-network is used to predict the third event prediction attribute of the sample objects in the control group. The second sub-network is used to predict the third event prediction attribute of the sample objects in the experimental group. If the grouping identifier indicates that the sample object belongs to the control group, the object information set of the sample object is input into the first sub-network, and inference prediction is performed through the first sub-network to obtain the third event prediction attribute. Based on the third event prediction attribute and the historical event attributes of the target events associated with the sample object in the object information set, the prediction loss value of the first sub-network for that sample object is calculated. If the grouping identifier indicates that the sample object belongs to the experimental group, the object information set of the sample object is input into the second sub-network, and inference prediction is performed through the second sub-network to obtain the third event prediction attribute. Based on the third event prediction attribute and the historical event attributes of the target events associated with the sample object in the object information set, the prediction loss value of the second sub-network for that sample object is calculated.

[0079] A second loss function value is determined based on the predicted loss value and the information distribution difference value, and the gain model is trained based on the second loss function value. Specifically, the second loss function value can be obtained by combining the predicted loss value and the distribution difference value with preset coefficients. The model parameters of the gain model are adjusted with the goal of reducing the second loss function value, and the above model training process is iteratively executed until the training termination condition is met.

[0080] Figure 3a This is a schematic diagram of the structure of a gain model provided in an embodiment of this disclosure. Figure 3aAs shown, the gain model includes a feature representation learning network 310, a first sub-network 320, a second sub-network 330, and a distribution distance calculation module 340. The feature representation learning network 310 includes a discrete feature representation learning sub-network 311, a continuous feature representation learning sub-network 312, and a feature concatenation sub-network 313. The discrete feature representation learning sub-network 311 includes a first input layer 3110, a hash processing layer 3111, an embedding layer 3112, a first fully connected layer 3113, a first dropout layer 3114, and a flatten layer 3115. The first input layer 3110 receives discrete features from the object information set of the sample object, performs hash processing on the discrete features through the hash processing layer 3111, encodes the hash processing result through the embedding layer 3112, and then inputs the encoded result into the fully connected layer 3113 for representation learning. The continuous feature representation learning subnetwork 312 includes a second input layer 3120, a first batch normalization layer 3121, a second fully connected layer 3122, a second Dropout layer 3123, a third fully connected layer 3124, and a third Dropout layer 3125. The second input layer 3120 receives continuous features from the object information set of the sample objects and performs representation learning on these continuous features through the first batch normalization layer 3121, the second fully connected layer 3122, the second Dropout layer 3123, the third fully connected layer 3124, and the third Dropout layer 3125. Representation learning is used to map discrete features and continuous features into feature vectors in a high-dimensional space. The feature concatenation subnetwork 313 includes a feature concatenation layer 3130, a first batch normalization layer 3131, a third fully connected layer 3132, and a fourth fully connected layer 3133. The feature vectors corresponding to discrete features and the feature vectors corresponding to continuous features are input into the feature concatenation layer 3130 for feature concatenation to obtain concatenated features. The concatenated features are learned through a first batch normalization layer 3131, a third fully connected layer 3132, and a fourth fully connected layer 3133 to map them into a high-dimensional feature vector, phi, as shown in Figure 3. Both the first sub-network 320 and the second sub-network 330 include multiple sequentially connected fully connected layers and an output layer. The output layer of the first sub-network 320 outputs the third event prediction attribute. The output layer of the second sub-network 330 also outputs the third event prediction attribute. The distribution distance calculation module 340 determines the feature vector phi(t=0) corresponding to the control group and the feature vector phi(t=1) corresponding to the experimental group based on the group identifier, and calculates the distribution distance dis(phi(t=1), phi(t=0)) between the feature vectors corresponding to the control group and the experimental group, respectively, as the information distribution difference value between the sample objects in the control group and the experimental group. The prediction loss value is determined based on the difference between the third event prediction attribute and the historical event attribute of the target event associated with the sample object. The second loss function value is determined based on the prediction loss value and the distribution difference value. The gain model is trained based on the second loss function value.

[0081] Figure 3b This is a schematic flowchart illustrating a training method for a gain model provided in an embodiment of this disclosure. Figure 3b As shown, if the group identifier for the control group is t=0 and the group identifier for the experimental group is t=1, then the object information set corresponding to the sample objects in the control group is represented as (X+X_S, t=0, Y), and the object information set corresponding to the sample objects in the experimental group is represented as (X+X_S, t=1, Y). Here, X_S represents the historical event attributes of the target event associated with the sample object, X represents the object attributes and object interaction information in the object information other than the historical event attributes of the target event, etc. Y represents the sample label, which refers to the event attribute correction value. For example, X_S represents the product sales volume, X represents the number of times the product was added to the cart, the number of times the product was shared, the number of times the product was favorited, and the number of product reviews, etc. Y represents the product sales volume. (X+X_S, t=0, Y) and (X+X_S, t=1, Y) are input into the correction model respectively, and the predicted mean P of the first attribute is determined through the correction model. Z is used to represent the corrected event attribute correction value, where Z=YP. The corrected object information sets for the control and experimental groups are represented as (X+X_S, t=0, Z) and (X+X_S, t=1, Z), respectively. (X+X_S, t=1, Z) is input into the first sub-network of the gain model to obtain the third event prediction attribute. (X+X_S, t=0, Z) is input into the second sub-network of the gain model to obtain the third event prediction attribute. The prediction loss value is determined based on the third event prediction attribute and Z. The information distribution difference between (X+X_S, t=0) and (X+X_S, t=1) is calculated, and the second loss function value is determined based on the prediction loss value and the information distribution difference value. The gain model is then trained based on the second loss function value.

[0082] S130. Based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object, determine the target object associated with the target business in the candidate object set.

[0083] Wherein, the first event prediction attribute represents the business attribute prediction result when the target event is associated with the target business; the second event prediction attribute represents the business attribute prediction result when the target event is not associated with the target business. Specifically, the first event prediction attribute can be the event attribute prediction result of the target event associated with the candidate object output by the first sub-network of the gain model. The second event prediction attribute can be the event attribute prediction result of the target event associated with the candidate object output by the second sub-network of the gain model. The target object can refer to an object that can improve the event attribute of the target event due to the influence of the target business. For example, the target object can refer to a product that may increase sales due to a promotional activity. Since the target object is a promotional activity to be implemented, precise and real-time promotional activities can be carried out for such products.

[0084] For example, the difference between the first event prediction attribute and the second event prediction attribute of the candidate object is determined. If the difference is greater than a set threshold, the candidate object is determined as the first object; otherwise, the candidate object is determined as the second object. For any second object in the candidate object set, the object information set of the second object is input into a pre-trained classification model, and the similarity between the second object and the first object is determined by the classification model. Based on the second object and the first object whose similarity satisfies a preset similarity condition, the target object associated with the target business in the candidate object set is determined.

[0085] Here, the similarity between the second object and the first object represents the probability that the second object is similar to the first object. The preset similarity condition can be that the probability of the second object being similar to the first object is greater than a set similarity threshold.

[0086] It should be noted that the historical event attributes of the target events associated with the candidate objects are first removed from the object information set of the second object, and then the object information set after removing the event attributes is input into the classification model.

[0087] Taking a product as an example, if the difference is greater than 0, the product is determined to be affected by the marketing campaign and its sales will increase; such a product is the first object. If the difference is less than 0, the product is determined to be unaffected by the campaign and its sales will not increase; such a product is the second object. Based on the product attributes of the second object, such as its product category and function, as well as interaction information such as the number of times it was added to cart, shared, and favorited, the similarity between the second object and the first object is determined. Then, the target object is determined based on the second object, which is similar to the first object, and the first object itself.

[0088] The technical solution of this disclosure involves inputting a set of object information for candidate objects into a pre-trained gain model. This gain model then predicts the event attributes of target events associated with the candidate objects based on the object information set, obtaining a first event prediction attribute and a second event prediction attribute. Next, based on the first and second event prediction attributes corresponding to the candidate objects, the target objects associated with the target business within the candidate object set are determined. Since the gain model is trained based on sample objects and event attribute correction values, and the event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the mean of the first attribute predictions of the target events associated with the sample objects, the influence of individual differences among sample objects in the random grouping results on the model's predictions can be eliminated. This facilitates the gain model learning the true causal relationship between the event attributes of the target business and the target events, thereby improving the prediction accuracy of the gain model.

[0089] Figure 4This is a flowchart illustrating another method for determining objects provided in this embodiment of the present disclosure. Based on the above embodiments, this embodiment of the present disclosure further discloses that after the gain model training is completed, positive and negative samples are determined based on the output results of the gain model, and a classification model is trained based on the positive and negative samples.

[0090] like Figure 4 As shown, the method includes:

[0091] S410. Input the object information set of the sample object into the pre-trained gain model, and use the gain model to predict the event attributes of the target event based on the object information set of the sample object to obtain the first event prediction attribute and the second event prediction attribute corresponding to the sample object.

[0092] For example, the object information set of the sample objects is input into a pre-trained gain model. The discrete features in the object information set are vectorized using the gain model to obtain discrete vectors. The continuous features in the object information set are also vectorized using the gain model to obtain continuous vectors. The discrete and continuous vectors are concatenated to obtain a concatenated vector. Then, the concatenated vector is vectorized to obtain a high-dimensional feature vector. Based on the grouping identifiers in the object information set, the high-dimensional feature vector is input into the first and second sub-networks, respectively. The first event prediction attribute output by the first sub-network and the second event prediction attribute output by the second sub-network are obtained.

[0093] S420. Based on the difference between the first event prediction attribute and the second event prediction attribute, the sample object is marked as a positive sample or a negative sample.

[0094] In this context, positive samples represent objects whose event attributes are improved due to the influence of the target service. Negative samples represent objects whose event attributes are not improved by the target service. If the difference is greater than zero, it is determined that the target service improves the event attributes of the target event. If the difference is less than zero, it is determined that the target service does not improve the event attributes of the target event.

[0095] For example, sample objects whose difference between the predicted attribute of the first event and the predicted attribute of the second event is greater than zero are considered positive samples, and sample objects whose difference is less than zero are considered negative samples. This embodiment of the present disclosure can determine negative samples based on the inference results of the gain model, solving the problem in related technologies of the difficulty in determining the negative samples required for training a classification model. In related technologies, sample objects whose historical event attributes of a target event within a set time period meet set screening conditions are considered positive samples, and sample objects whose historical event attributes of a target event within a set time period do not meet the set screening conditions are considered negative samples. However, not meeting the set screening conditions in the past does not mean that they will not meet them in the future. Therefore, using sample objects that did not meet the set screening conditions in the past as negative samples to train the classification model will affect the accuracy of the classification model training.

[0096] S430. Determine the training dataset based on the object information set corresponding to the positive and negative samples, and train the classification model using the training dataset.

[0097] The classification model is used to determine candidate objects that are similar to the positive samples.

[0098] The training dataset includes object attributes and object interaction information, such as historical event attributes of target events from the object information sets corresponding to positive and negative samples. Positive samples are labeled 1, and negative samples are labeled 0. The positive sample dataset is constructed based on the object information sets and labels of the positive samples. The negative sample dataset is constructed based on the sample object information sets and labels of the negative samples. Historical event attributes of target events are removed from both the positive and negative sample datasets. The training dataset is then constructed based on the positive and negative sample datasets after removing the attribute values.

[0099] Optionally, by setting a sampling ratio, positive and negative samples can be sampled separately according to the sampling ratio, resulting in positive and negative samples of the same order of magnitude.

[0100] Optionally, the feature vectors in the high-dimensional space can be filtered using the MMoE (Multi-gate Mixture-of-Experts) model to remove the historical event attributes of the target events in the object information set corresponding to positive and negative samples.

[0101] Optionally, the historical event attributes of the target event in the feature vector of the high-dimensional space are determined by the feature identifier (index) corresponding to the feature vector of the high-dimensional space, and the historical event attributes of the target event in the feature vector of the high-dimensional space are deleted.

[0102] This disclosure does not limit the specific structure of the classification model; the classification model can be freely selected according to the business scenario. The object information set of positive or negative samples in the training dataset is input into the classification model to obtain the classification result of the sample object output by the model. The classification result represents the probability that a sample object is a positive sample. A loss value is calculated based on the classification result and the label of the sample object, and the classification model is trained based on the loss value.

[0103] Figure 5 This is a flowchart illustrating a training method for a classification model provided in an embodiment of this disclosure. Figure 5 As shown, X+X_S of any object in the sample object Q is input into the gain model 510. The gain model 510 performs inference based on X+X_S to obtain the first event prediction attribute and the second event prediction attribute of the sample object. The positive sample Q+ and the negative sample Q- in the sample object are determined according to the difference between the first event prediction attribute and the second event prediction attribute. The classification model 520 is trained based on the X features of the positive sample Q+ and the X features of the negative sample Q-.

[0104] The technical solution of this disclosure uses a gain model to determine positive and negative samples in the sample objects. It can accurately identify negative samples that do not improve the event attributes of the target event. The training dataset is determined based on the sample information set of positive and negative samples. The classification model is then trained based on the training dataset. This allows the classification model to find candidate objects similar to positive samples based on positive samples, avoiding the problem of affecting the training accuracy of the classification model by treating objects that do not meet the set screening conditions as negative samples.

[0105] Figure 6 This is a schematic diagram of an object determination device provided in an embodiment of the present disclosure. The device can be implemented in the form of software and / or hardware, and optionally, it can be implemented by an electronic device, such as a mobile terminal, a PC, or a server.

[0106] like Figure 6 As shown, the device includes: an information acquisition module 610, an attribute prediction module 620, and an object determination module 630.

[0107] The information acquisition module 610 is used to acquire the object information set of the candidate objects in the candidate object set, wherein the object information set includes object information, and the object information includes object attributes and object interaction information;

[0108] The attribute prediction module 620 is used to input the object information set into a pre-trained gain model, and use the gain model to predict the event attributes of the target events associated with the candidate objects based on the object information set, so as to obtain the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects. The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the first attribute prediction mean of the target events associated with the sample objects. The first attribute prediction mean is determined based on the object information set of the sample objects.

[0109] The object determination module 630 is used to determine the target object associated with the target business in the candidate object set based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object.

[0110] Optionally, the first event prediction attribute represents the business attribute prediction result when the target event is associated with the target service; the second event prediction attribute represents the business attribute prediction result when the target event is not associated with the target service.

[0111] The object determination module 630 is specifically used for:

[0112] Determine the difference between the first event prediction attribute and the second event prediction attribute of the candidate object;

[0113] If the difference is greater than a set threshold, the candidate object is determined as the first object; otherwise, the candidate object is determined as the second object.

[0114] For any second object in the candidate object set, the object information set of the second object is input into a pre-trained classification model, and the similarity between the second object and the first object is determined by the classification model.

[0115] Based on the second object and the first object that satisfy the preset similarity conditions, the target object associated with the target business is determined in the candidate object set.

[0116] Optionally, the training method of the gain model includes:

[0117] The sample objects are divided into a control group or an experimental group. The sample objects in the control group are not associated with the target business, while the sample objects in the experimental group are associated with the target business.

[0118] Obtain the object information set of the sample object, determine the predicted mean of the first attribute based on the object information set, correct the historical event attribute of the target event associated with the sample object based on the predicted mean of the first attribute, and obtain the event attribute correction value.

[0119] The gain model is trained based on the object information set of the sample objects and the event attribute correction values.

[0120] Further, the step of determining the predicted mean of the first attribute based on the object information set, and correcting the historical event attributes of the target event associated with the sample object based on the predicted mean of the first attribute to obtain the event attribute correction value includes:

[0121] Based on the object information set of the sample objects, determine the mean value of the object information corresponding to the sample objects;

[0122] The object information set and the mean of the object information are input into a pre-trained correction model. The correction model determines the third event prediction attribute of the target event associated with the sample object based on the object information of the sample object, determines the second attribute prediction mean of the target event associated with the sample object based on the mean of the object information, and determines the first attribute prediction mean of the target event associated with the sample object based on the third event prediction attribute and the second attribute prediction mean. The correction model is trained using the object information set of the sample objects before grouping. The third event prediction attribute represents the business attribute prediction result of the target event after the sample objects are divided into the control group or the experimental group.

[0123] Based on the predicted mean of the first attribute and the historical event attributes of the target event, an event attribute correction value is generated.

[0124] Further, training the gain model based on the object information set of the sample object and the event attribute correction value includes:

[0125] The object information set of the sample object is input into the gain model to be trained. The gain model predicts the event attribute of the target event based on the object information set of the sample object, and obtains the third event prediction attribute corresponding to the sample object.

[0126] The prediction loss value is determined based on the historical event attributes of the target event associated with the third event prediction attribute and the sample object.

[0127] Based on the grouping identifier of the sample objects, the information distribution difference value of the sample objects in the control group and the experimental group is determined, wherein the grouping identifier is used to indicate whether the sample object belongs to the control group or the experimental group;

[0128] The loss function value is determined based on the predicted loss value and the information distribution difference value, and the gain model is trained based on the loss function value.

[0129] Optionally, the training method of the classification model includes:

[0130] The object information set of the sample object is input into a pre-trained gain model. The gain model predicts the event attributes of the target event based on the object information set of the sample object, thereby obtaining the first event prediction attribute and the second event prediction attribute corresponding to the sample object.

[0131] Based on the difference between the first event prediction attribute and the second event prediction attribute, the sample object is marked as a positive sample or a negative sample.

[0132] The training dataset is determined based on the object information sets corresponding to the positive and negative samples, and the classification model is trained using the training dataset.

[0133] Further, determining the training dataset based on the object information sets corresponding to the positive and negative samples includes:

[0134] The training dataset is obtained by deleting the historical event attributes of the target event from the object information set corresponding to the positive and negative samples.

[0135] The object determination apparatus provided in this disclosure can execute the object determination method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the method execution.

[0136] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this disclosure.

[0137] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Reference is made below. Figure 7 It illustrates an electronic device suitable for implementing embodiments of the present disclosure (e.g., Figure 7 The diagram below shows the structure of the terminal device or server 700. The terminal device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0138] like Figure 7As shown, the electronic device 700 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device 700. The processing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An edit / output (I / O) interface 705 is also connected to the bus 704.

[0139] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic device 700 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 An electronic device 700 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0140] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 709, or installed from storage device 708, or installed from ROM 702. When the computer program is executed by processing device 701, it performs the functions defined in the methods of embodiments of this disclosure.

[0141] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0142] The electronic device provided in this embodiment belongs to the same inventive concept as the method provided in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0143] This disclosure provides a computer storage medium storing a computer program that, when executed by a processor, implements the object determination method provided in the above embodiments.

[0144] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0145] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0146] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0147] The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to:

[0148] Obtain an object information set of candidate objects from the candidate object set, wherein the object information set includes object information, and the object information includes object attributes and object interaction information;

[0149] The object information set is input into a pre-trained gain model. The gain model predicts the event attributes of the target events associated with the candidate objects based on the object information set, thereby obtaining the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects. The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the first attribute prediction mean of the target events associated with the sample objects. The first attribute prediction mean is determined based on the object information set of the sample objects.

[0150] Based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object, the target object associated with the target business in the candidate object set is determined.

[0151] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0152] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0153] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0154] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0155] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0156] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0157] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0158] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for determining an object, characterized in that, include: Obtain an object information set of candidate objects from the candidate object set, wherein the object information set includes object information, and the object information includes object attributes and object interaction information; The object information set is input into a pre-trained gain model. The gain model predicts the event attributes of the target events associated with the candidate objects based on the object information set, thereby obtaining the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects. The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the first attribute prediction mean of the target events associated with the sample objects. The first attribute prediction mean is determined based on the object information set of the sample objects. Based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object, the target object associated with the target business in the candidate object set is determined.

2. The method according to claim 1, characterized in that, The first event prediction attribute represents the business attribute prediction result when the target event is associated with the target service; the second event prediction attribute represents the business attribute prediction result when the target event is not associated with the target service. The step of determining the target object associated with the target business in the candidate object set based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object includes: Determine the difference between the first event prediction attribute and the second event prediction attribute of the candidate object; If the difference is greater than a set threshold, the candidate object is determined as the first object; otherwise, the candidate object is determined as the second object. For any second object in the candidate object set, the object information set of the second object is input into a pre-trained classification model, and the similarity between the second object and the first object is determined by the classification model. Based on the second object and the first object that satisfy the preset similarity conditions, the target object associated with the target business is determined in the candidate object set.

3. The method according to claim 1, characterized in that, The training methods for the gain model include: The sample objects are divided into a control group or an experimental group. The sample objects in the control group are not associated with the target business, while the sample objects in the experimental group are associated with the target business. Obtain the object information set of the sample object, determine the predicted mean of the first attribute based on the object information set, correct the historical event attribute of the target event associated with the sample object based on the predicted mean of the first attribute, and obtain the event attribute correction value. The gain model is trained based on the object information set of the sample objects and the event attribute correction values.

4. The method according to claim 3, characterized in that, The step of determining the predicted mean of the first attribute based on the object information set, and correcting the historical event attributes of the target event associated with the sample object based on the predicted mean of the first attribute to obtain the event attribute correction value includes: Based on the object information set of the sample objects, determine the mean value of the object information corresponding to the sample objects; The object information set and the mean of the object information are input into a pre-trained correction model. The correction model determines the third event prediction attribute of the target event associated with the sample object based on the object information of the sample object, determines the second attribute prediction mean of the target event associated with the sample object based on the mean of the object information, and determines the first attribute prediction mean of the target event associated with the sample object based on the third event prediction attribute and the second attribute prediction mean. The correction model is trained using the object information set of the sample objects before grouping. The third event prediction attribute represents the business attribute prediction result of the target event after the sample objects are divided into the control group or the experimental group. Based on the predicted mean of the first attribute and the historical event attributes of the target event, an event attribute correction value is generated.

5. The method according to claim 3, characterized in that, Training the gain model based on the object information set of the sample object and the event attribute correction value includes: The object information set of the sample object is input into the gain model to be trained. The gain model predicts the event attribute of the target event based on the object information set of the sample object, and obtains the third event prediction attribute corresponding to the sample object. The prediction loss value is determined based on the historical event attributes of the target event associated with the third event prediction attribute and the sample object. Based on the grouping identifier of the sample objects, the information distribution difference value of the sample objects in the control group and the experimental group is determined, wherein the grouping identifier is used to indicate whether the sample object belongs to the control group or the experimental group; The loss function value is determined based on the predicted loss value and the information distribution difference value, and the gain model is trained based on the loss function value.

6. The method according to claim 2, characterized in that, The training methods for the classification model include: The object information set of the sample object is input into a pre-trained gain model. The gain model predicts the event attributes of the target event based on the object information set of the sample object, thereby obtaining the first event prediction attribute and the second event prediction attribute corresponding to the sample object. Based on the difference between the first event prediction attribute and the second event prediction attribute, the sample object is marked as a positive sample or a negative sample; The training dataset is determined based on the object information set corresponding to the positive and negative samples, and the classification model is trained using the training dataset.

7. The method according to claim 6, characterized in that, The step of determining the training dataset based on the object information sets corresponding to the positive and negative samples includes: The training dataset is obtained by deleting the historical event attributes of the target event from the object information set corresponding to the positive and negative samples.

8. An object determining device, characterized in that, include: The information acquisition module is used to acquire the object information set of the candidate objects in the candidate object set, wherein the object information set includes object information, and the object information includes object attributes and object interaction information; The attribute prediction module is used to input the object information set into a pre-trained gain model, and use the gain model to predict the event attributes of the target events associated with the candidate objects based on the object information set, so as to obtain the first event prediction attribute and the second event prediction attribute corresponding to the candidate objects. The gain model is trained based on sample objects and event attribute correction values. The event attribute correction values ​​are generated by correcting the historical event attributes of the target events based on the first attribute prediction mean of the target events associated with the sample objects. The first attribute prediction mean is determined based on the object information set of the sample objects. The object determination module is used to determine the target object associated with the target business in the candidate object set based on the first event prediction attribute and the second event prediction attribute corresponding to the candidate object.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining the object as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the method for determining the object as described in any one of claims 1-7.

Images