Gift package generation model training method, gift package pushing method, device and electronic equipment

Abstract

The application discloses a gift package generation model training method, a gift package pushing method and device, electronic equipment and a computer readable storage medium. The training method comprises the following steps: obtaining a training set, wherein the training set comprises sample user portrait information corresponding to each sample user; determining first input information of a to-be-trained gift package generation model according to the sample user portrait information and 0th to (i-1)th output commodity information, inputting the first input information into the to-be-trained gift package generation model, and obtaining ith output commodity information in a sample gift package generated for the sample user, wherein i is traversed from 1 to N; generating a sample gift package for the sample user according to the 1st to Nth output commodity information; and adjusting model parameters of the to-be-trained gift package generation model according to a parameter adjustment principle of improving a purchase sample gift package intention parameter, so as to obtain a trained gift package generation model. By using the scheme provided in the application, the probability that a user likes the commodity in the pushed gift package can be improved, and the user can buy the favorite gift package with a higher probability.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a method for training a gift package generation model, a method for pushing gift packages, an apparatus, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In online games, game developers often offer gift packs to players to enhance their gaming experience. A game gift pack typically consists of a combination of multiple items in any quantity; when a player purchases a gift pack, all items within the pack are simultaneously purchased by the player.

[0003] When pushing gift packs to players, the relevant technology usually involves game designers with prior knowledge designing a batch of popular gift packs first, using these popular gift packs as a gift pack pool, and then selecting the gift packs that best match the player's preferences from the gift pack pool based on the player's browsing and purchase history, player attribute information, etc., and pushing them to the player.

[0004] However, the above-mentioned gift pack push method can only select gift packs from a fixed gift pack pool to push to players. Since the gift pack combinations in the gift pack pool are limited, it is easy to cause the items in the gift packs pushed to players to be not what players like. As a result, players may buy gift packs they do not like, which reduces players' game satisfaction and player experience, and thus reduces player retention rate. Summary of the Invention

[0005] This application provides a method for training a gift pack generation model, a method for pushing gift packs, an apparatus, an electronic device, and a computer-readable storage medium, which can increase the probability that users like the items in the pushed gift packs, enabling users to buy the gift packs they like more likely, improving user game satisfaction and user experience, thereby increasing user retention. The specific method is as follows.

[0006] In a first aspect, embodiments of this application provide a method for training a gift package generation model, the training method comprising:

[0007] Obtain a training set, which includes sample user profile information corresponding to each sample user;

[0008] Based on the sample user profile information and the 0th to i-1th output product information, the first input information of the gift pack generation model to be trained is determined, and the first input information is input into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty;

[0009] Based on the first to Nth output product information, a sample gift package is generated for the sample user;

[0010] Determine the sample users' intention parameters to purchase the sample gift package;

[0011] The model parameters of the gift pack generation model to be trained are adjusted according to the principle of improving the intention parameters until the intention parameters meet the preset conditions, thus obtaining a trained gift pack generation model.

[0012] Secondly, embodiments of this application provide a method for pushing gift packages, the method comprising:

[0013] Obtain the user profile information of the user to be pushed to;

[0014] The third input information of the trained gift pack generation model is determined based on the user profile information to be pushed and the information of the 0th to j-1th product to be pushed. The third input information is then input into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed. Here, j is an integer, j traverses from 1 to T, and T is the number of preset product types included in the push gift pack. The 0th product information to be pushed is empty. The trained gift pack generation model is trained using the gift pack generation model training method provided in the first aspect embodiment.

[0015] Based on the first to T items of information to be pushed, a push gift pack is generated for the user to be pushed;

[0016] The push notification package will be sent to the user to be pushed to.

[0017] Thirdly, this application also provides a gift pack generation model training device, the training device comprising:

[0018] The first acquisition unit is used to acquire a training set, which includes sample user profile information corresponding to each sample user.

[0019] The first calculation unit is used to determine the first input information of the gift pack generation model to be trained based on the sample user profile information and the 0th to i-1th output product information, and input the first input information into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty;

[0020] The first generation unit is used to generate sample gift packs for the sample users based on the first to Nth output product information.

[0021] A determining unit is used to determine the intention parameters of sample users to purchase the sample gift package;

[0022] The adjustment unit is used to adjust the model parameters of the gift pack generation model to be trained, based on the principle of improving the intention parameters, until the intention parameters meet the preset conditions, and thus obtain the trained gift pack generation model.

[0023] Fourthly, embodiments of this application also provide a gift package delivery device, the delivery device comprising:

[0024] The second acquisition unit is used to acquire the user profile information of the user to be pushed to;

[0025] The second calculation unit is used to determine the third input information of the trained gift pack generation model based on the user profile information to be pushed and the 0th to j-1th product information to be pushed, and input the third input information into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed, where j is an integer, j traverses 1 to T, T is the number of preset product types contained in the push gift pack, and the 0th product information to be pushed is empty. The trained gift pack generation model is trained by the gift pack generation model training method provided in the first aspect embodiment.

[0026] The second generation unit is used to generate a push gift pack for the user to be pushed based on the first to T pieces of information about the products to be pushed.

[0027] The push unit is used to push the push gift package to the user to be pushed.

[0028] Fifthly, embodiments of this application also provide an electronic device, including:

[0029] Processor; and

[0030] A memory for storing a data processing program, which, when the electronic device is powered on and runs by the processor, performs the method as described in either the first or second aspect.

[0031] Sixthly, embodiments of this application also provide a computer-readable storage medium storing a data processing program that is executed by a processor to perform the method described in either the first or second aspect.

[0032] Compared with the prior art, this application has the following advantages:

[0033] The training method for the gift package generation model provided in this application embodiment first obtains a training set, which includes sample user profile information corresponding to each sample user. Then, based on the sample user profile information and the 0th to i-1th output product information, the first input information of the gift package generation model to be trained is determined, and the first input information is input into the gift package generation model to be trained to obtain the i-th output product information in the sample gift package generated for the sample user. Since i traverses from 1 to N, where N is the number of preset product types contained in the sample gift package, and the 0th output product information is empty, this application embodiment can sequentially obtain the 1st, 2nd, ... Nth output product information in the sample gift package generated for the sample user. In this application embodiment, when determining the i-th output product information, the sample user profile information and the 0th to i-1th output product information are input into the training model. The gift package generation model, in other words, considers both the user profile information and the previously generated output items in the sample gift package when generating the i-th output item. Since the gift package is a bundle of individual items pushed to the user, the i-th output item can be understood as the user liking the i-th item in addition to liking the 0-i-1 items. Therefore, during model training, the training is based on ensuring that each item in the sample gift package generated for the user based on the 1-N output item information better meets the user's needs. After determining the user's intention to purchase the generated sample gift package, the model parameters of the gift package generation model can be adjusted by increasing this intention parameter until the intention parameter meets the preset conditions, resulting in a well-trained gift package generation model. The gift package generation model trained using the training method provided in this application can generate personalized gift packages tailored to the user's profile information, making the generated personalized gift packages more in line with user preferences.

[0034] When the gift pack generation model trained using the training method provided in this application pushes gift packs to users, the j-th item to be pushed in the gift pack is obtained by inputting the user profile information and the (j-1)-th item information into the gift pack generation model. j iterates through 1 to T, where T is the number of preset item types in the push gift pack. Thus, the j-th item to be pushed in the push gift pack can be understood as the user liking the j-th item in addition to liking the 0 to (j-1)-th items. Therefore, the items in the push gift pack generated for the user based on the 1 to T-th item information are more in line with the user's preferences. After the push gift pack is pushed to the user, the user will have a higher satisfaction with each item in the pushed gift pack, increasing the probability that the user likes the items in the pushed gift pack, enabling the user to buy the gift pack they like more likely, improving the user's game satisfaction and user experience, thereby improving the user retention rate.

[0035] In addition, the embodiments of this application can generate personalized gift packs that match the user's preferences based on the user's profile information, instead of selecting gift packs to push to the user from a fixed gift pack pool. This can further ensure that the gift packs pushed to the user match the user's preferences, thereby further improving the user's game satisfaction and user experience. Attached Figure Description

[0036] Figure 1 This is an example flowchart of the gift package generation model training method provided in the embodiments of this application;

[0037] Figure 2 This is a flowchart of an example of the gift pack generation process provided in this application embodiment;

[0038] Figure 3 This is a schematic diagram of the structure of the gift package push system provided in the embodiments of this application;

[0039] Figure 4 This is a schematic diagram of the encoding process provided in the embodiments of this application;

[0040] Figure 5 This is a flowchart illustrating the process of determining the output product or the product to be pushed, provided in an embodiment of this application.

[0041] Figure 6 This is a schematic diagram of the process for selecting a push notification package provided in an embodiment of this application;

[0042] Figure 7 This is a flowchart of an example of the gift package push method provided in the embodiments of this application;

[0043] Figure 8 This is a block diagram of the gift package generation model training device provided in the embodiments of this application;

[0044] Figure 9 This is a schematic diagram of the logical structure of an electronic device for implementing gift pack generation model training, provided in an embodiment of this application. Detailed Implementation

[0045] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0046] Game gift pack generation and delivery refers to the process by which the game system generates and delivers bundled items to players. In video games, game gift packs are an important feature that directly impacts players' game content and experience, thereby affecting user satisfaction and user retention.

[0047] In related technologies, game designers with game knowledge and experience can design a batch of popular gift packs and create gift packs, and then distribute popular gift packs to logged-in players based on users' historical purchase behavior; alternatively, machine learning models can be used to determine each player's product preferences based on their historical browsing and purchase records, and then select the gift packs that best match the player's preferences from a pool of popular gift packs designed by experts for push.

[0048] However, the above-mentioned gift pack push method can only select gift packs from a fixed gift pack pool to push to players. Since the gift pack combinations in the gift pack pool are limited, it is easy to cause the items in the gift packs pushed to players to be not what players like. As a result, players may buy gift packs they do not like, which reduces players' game satisfaction and player experience, and thus reduces player retention rate.

[0049] To increase the probability that users will like the items in the recommended gift packs, enabling them to purchase their preferred packs more likely, thereby improving user satisfaction and user experience, and ultimately increasing user retention, the first embodiment of this application provides a gift pack generation model training method. This method can be applied to generate gift packs for players in video game systems, as well as for generating gift packs for users in online shopping systems, social software systems, and other systems, thus enhancing the user experience of the application.

[0050] The execution subject of the gift package generation model training method provided in this application can be an electronic device, such as a desktop computer, laptop computer, tablet computer, mobile phone, server, etc., or other electronic devices with data processing functions. This application does not specifically limit the subject.

[0051] like Figure 1 As shown, the gift package generation model training method provided in this application includes the following steps S110 to S150.

[0052] Step S110: Obtain the training set.

[0053] The training set includes sample user profile information corresponding to each sample user.

[0054] Electronic devices can obtain sample user profile information from various clients or from storage devices, which store the user profile information for each user. Taking this solution as an example in a video game scenario, electronic devices can obtain sample user profile information from the game logs of sample users.

[0055] In this embodiment, the sample users can be users who meet preset conditions. These preset conditions may include at least one of the following: users who have made purchases within a preset time period (i.e., the time when model training started), users whose user rating ranks among the top preset positions, etc., but are not limited to these. The preset time period can be one month, 20 days, one week, etc., and the preset ranking can be, for example, 5000th, 10000th, 20000th, etc., but are not limited to these. When a sample user has recently made purchases or has a relatively high user rating, it indicates that the sample user is typically a normal player, making the user profile information corresponding to the sample user more effective, more relevant to model training, and resulting in a more accurate trained model.

[0056] The aforementioned sample user profile information may include at least one of the following: sample user's historical purchase information and sample user attribute description information. Sample user's historical purchase information includes at least one of the following: sample user's historical purchase of gift packs and sample user's historical purchase of goods, wherein the gift pack includes at least one item.

[0057] The attribute description information of the sample users is used to describe the attributes of the sample users themselves. For example, the attribute description information may include at least one of the following: the sample user's level, the sample user's average spending in the game, the sample user's equipment value, the sample user's online time information, the sample user's gender information, the sample user's age information, and the sample user's geographical location, but is not limited to these.

[0058] A user's purchase history can include information on purchases of equipment, clothing, accessories, etc., or information on purchases of gift packs. Gift packs can include different items. For example, the first gift pack includes three items: item 1, item 2, and item 3, while the second gift pack includes two items: item 3 and item 4.

[0059] During gameplay, users frequently update their attributes such as age and location. As users continue to participate in the game, their attributes such as level, spending, equipment, and online time are also updated synchronously. In this embodiment, the attribute description information of the sample users can be the latest attribute description information of the sample users, so as to make the model training more real-time and more accurate.

[0060] The sample user's historical purchase information can be used to record the purchase information of the sample within a time period of a preset time from the current time, so as to make the data for model training more real-time and the trained model more accurate.

[0061] Step S120: Determine the first input information of the gift pack generation model to be trained based on the sample user profile information and the 0th to i-1th output product information, and input the first input information into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user.

[0062] The above i is an integer, i iterates from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty.

[0063] The i-th output product information is the product information of the i-th product in the sample gift pack. This product information can be a product identifier, such as the product name, product ID, or other uniquely identifying information about the product. Based on the output product identifier, it is easy to determine which specific product it is.

[0064] In step S120, the sample user profile information and the 0th to i-1th output product information can be directly determined as the first input information; or the sample user profile information and the 0th to i-1th output product information can be preprocessed to be processed into a data form that is easy for model training (such as vector form, hash value form, etc.), and the preprocessed information can be determined as the first input information.

[0065] Since the 0th output product information is empty, such as Figure 2 As shown, an exemplary execution process for step S120 can be: [The process involves] processing the sample user profile information (i.e.,...] Figure 2 The first input information, obtained from all features of the players to be recommended, is input into the gift pack generation model to be trained (i.e., Figure 2 In the gift pack content generator, the first output item ID (i.e., ...) of the sample gift pack is obtained. Figure 2 The first input information, obtained from the sample user profile information and the first output product ID, is input again into the gift pack generation model to be trained to obtain the second output product ID in the sample gift pack; the first input information, obtained from the sample user profile information and the first to second output product IDs, is input again into the gift pack generation model to be trained to obtain the third output product ID in the sample gift pack; ...; until the Nth output product ID in the sample gift pack is obtained.

[0066] The number of preset product types included in the sample gift package can be preset. N can be 2, 3, 4, 5, or other quantities. Those skilled in the art can set the value of N according to the actual needs of the scenario.

[0067] Step S130: Generate sample gift packs for sample users based on the output product information from the 1st to the Nth time.

[0068] When the output product information is a product identifier, the N products corresponding to the 1st to Nth output product identifiers can be identified as the N types of products included in the sample gift pack, and these N types of products can be identified as the products in the sample gift pack generated for the sample user. Each type of product in the sample gift pack can include one, two, or more. The number of different types of products can be the same or different. Those skilled in the art can design according to actual needs, and this application does not specifically limit this.

[0069] Step S140: Determine the sample users' intention parameters to purchase the above sample gift pack.

[0070] In step S140, the sample user profile information and the sample gift package can be input into the trained probability calculation model to obtain the probability of the sample user purchasing the sample gift package. This probability is used to represent the sample user's intention parameter to purchase the aforementioned gift package.

[0071] The aforementioned probability calculation model can be trained as follows: First, user profile information and gift packs for each first sample user are obtained. Each gift pack includes historical gift packs already purchased and those not yet purchased by the first sample user. The user profile information for the first sample user includes, but is not limited to, the user's age, gender, level, online time, etc. The user profile information and historical gift packs already purchased by the first sample user are input into the probability calculation model to be trained to obtain a first probability of the first sample user purchasing a previously purchased historical gift pack. The model parameters in the probability calculation model are adjusted to make the first probability close to 1. Then, the user profile information and historical gift packs not yet purchased by the first sample user are input into the probability calculation model to be trained to obtain a second probability of the first sample user purchasing a previously unpurchased historical gift pack. The model parameters in the probability calculation model are adjusted to make the second probability close to 0 until the difference between the first probability and 1 is less than a preset value, and the probability between the second probability and 0 is less than a preset value, thus obtaining a trained probability calculation model. The probability calculation model can also be trained in other ways, which are not specifically limited in this application.

[0072] In step S140, the intention parameters for sample users to purchase the sample gift pack can also be determined in other ways. For example, based on the items in gift packs previously purchased by the sample user and the items previously purchased, the items already purchased by the sample user can be determined. The intention parameters for sample users to purchase the sample gift pack can then be determined based on the similarity between each item in the sample gift pack and the items already purchased by the sample user. This application does not specifically limit how the intention parameters for sample users to purchase the sample gift pack are determined.

[0073] Step S150: Adjust the model parameters of the gift pack generation model to be trained, based on the principle of improving the above intention parameters, until the above intention parameters meet the preset conditions, and obtain the trained gift pack generation model.

[0074] In step S150, since sample user profile information corresponding to multiple sample users was obtained in step S110, step S150 can specifically be as follows: when the above intention parameters do not meet the preset conditions, the model parameters of the gift package generation model to be trained are adjusted based on the principle of improving the above intention parameters. Then, steps S120 to S150 are re-executed for another sample user until the sample user's intention parameters for purchasing the sample gift package meet the preset conditions. The gift package generation model to be trained when the preset conditions are met is determined as the trained gift package generation model.

[0075] The aforementioned preset conditions can be that the intention parameter reaches a preset threshold. For example, when the intention parameter is the probability that a sample user will purchase a sample gift package, the preset condition can be that the preset probability is reached. The preset probability can be any value between 0.8 and 1, or other relatively large probability values. When the intention parameter is the intention parameter score of a sample user to purchase a sample gift package, the preset condition can be that the preset intention parameter value is reached. The preset intention parameter value can be any value between 8 and 10, or other relatively large intention parameter scores. 10 points means completely possible, and 0 points means impossible.

[0076] The preset conditions can also be at least one of the following: the intention parameters of a consecutive preset number of sample users reach a preset threshold; the proportion of the first sample users whose intention parameters reach the preset threshold is greater than a preset proportion. The first sample users include: the current sample user in the current input to the training model and a consecutive preset number of sample users input before the current sample user. The preset number can be any number from 50 to 200, and the preset proportion can be any proportion from 80% to 90%, with no specific value limited. For example, the preset condition can be that the intention parameters of 300 consecutive sample users reach a preset threshold, or the preset condition can be that the proportion of the current sample user and the 300 consecutive sample users input before the current sample user whose intention parameters reach the preset threshold is 80%, 85%, or 90%.

[0077] Those skilled in the art can determine preset conditions according to actual needs in order to obtain a well-trained gift package generation model that meets the requirements.

[0078] The training method for the gift package generation model provided in this application embodiment first obtains a training set, which includes sample user profile information corresponding to each sample user. Then, based on the sample user profile information and the 0th to i-1th output product information, the first input information of the gift package generation model to be trained is determined, and the first input information is input into the gift package generation model to be trained to obtain the i-th output product information in the sample gift package generated for the sample user. Since i traverses from 1 to N, where N is the number of preset product types contained in the sample gift package, and the 0th output product information is empty, this application embodiment can sequentially obtain the 1st, 2nd, ... Nth output product information in the sample gift package generated for the sample user. In this application embodiment, when determining the i-th output product information, the sample user profile information and the 0th to i-1th output product information are input into the training model. The gift package generation model, in other words, considers both the sample user's profile information and the information of the previous output products already generated in the sample gift package. Since the gift package is a bundle of individual products pushed to the user, the i-th output product generated in this way can be understood as the user liking the i-th product in addition to liking the 0 to i-1 products. Therefore, during model training, the training is based on making each product in the sample gift package generated for the sample user based on the 1 to N output product information more in line with the user's needs. After determining the sample user's intention parameter to purchase the generated sample gift package, the model parameters of the gift package generation model to be trained can be adjusted by increasing this intention parameter until the intention parameter meets the preset conditions, resulting in a well-trained gift package generation model. The gift package generation model trained using the training method provided in this application can generate product sequences in the gift package for users in a targeted manner, thereby generating personalized gift packages that better meet user preferences.

[0079] In one implementation, step S120 may involve determining the first input information according to step S121.

[0080] Step S121: Numerically encode the sample user profile information and the output product information from 0 to i-1 to obtain the first numerical feature, and determine the first numerical feature as the first input information of the gift pack generation model to be trained.

[0081] like Figure 2 As shown, in this embodiment, before inputting the sample user profile information into the gift package content generator, an encoder is used to encode the sample user profile information to obtain the first numerical feature, and then the first numerical feature is input into the gift package content generator.

[0082] In step S121, the user profile information and the 0th to i-1th output product information can be encoded as a whole, and the resulting first numerical feature can represent the user profile information and the 0th to i-1th output product information. This simplifies the first input information, thereby making the training of the model more efficient. Alternatively, in step S121, the sample user profile information and the 0th to i-1th output product information can be encoded separately to obtain the first numerical feature corresponding to the sample user profile information and the first numerical feature corresponding to the 0th to i-1th output product information. This method can also achieve the purpose of numerically representing user features.

[0083] The above-mentioned numerical encoding method can be one-hot encoding, label encoding, or dummy encoding, or other encoding methods. This application does not specifically limit it.

[0084] The aforementioned numerical encoding can be vector encoding, where the first numerical feature can be the first vector feature, and the numerical representation of the i-th output product information can be a product vector. In other words, user profile information and product information can be represented in vector form. Numerical encoding can also be non-vector encoding; for example, the first numerical feature can be a string of numbers. This application does not specifically limit this.

[0085] In step S120, the i-th output product information in the sample gift package generated for the sample user can be obtained by following the steps S122 to S123.

[0086] Step S122: Input the first input information into the gift pack generation model to be trained, and obtain the numerical representation information of the i-th output item in the sample gift pack generated for the sample user.

[0087] The first input information in step S122 is the first numerical feature. Therefore, after being input into the gift package generation model to be trained, the numerical representation information of the i-th output product in numerical form can be easily obtained.

[0088] Step S123: Determine the information of the i-th output product in the sample gift pack generated for the sample user based on the numerical representation information of the i-th output product.

[0089] In step S123, the numerical representation information of the i-th output product can be decoded by the numerical decoding method corresponding to the numerical encoding method, thereby obtaining the information of the i-th output product.

[0090] Alternatively, step S123 can also be implemented as follows: step S123a.

[0091] Step S123a: Select the target product from the pre-stored candidate products as the i-th output product information in the sample gift package generated for the sample user.

[0092] Among them, the numerical representation information corresponding to the target product in the sample has a similarity greater than the numerical representation information of the i-th output product information, and the candidate products in the sample are provided with corresponding numerical representation information. For example Figure 5 As shown, this step can also be understood as processing the numerical representation information corresponding to each candidate product in each sample (i.e., ... Figure 5 The product code in the output and the numerical representation information of the i-th output product information (i.e., the product code in the output) ... Figure 5 The similarity of the output code (including the content) is calculated, and the product with the first preset similarity is determined as the target product of each sample.

[0093] In this embodiment, each sample candidate product can be pre-encoded using a numerical encoding method to obtain the numerical representation information corresponding to each sample candidate product. The numerical encoding method of the candidate product can be the same as the numerical encoding method of the sample user profile information and the 0th to i-1th output product information in step S121, so that the numerical representation information of the candidate product is consistent with the numerical representation information of the product output by the gift pack generation model to be trained.

[0094] The first preset similarity can be any value between 80% and 95%, or it can be any other relatively large similarity.

[0095] Optionally, step S123a can be implemented as follows: selecting at least one sample target product from the pre-stored sample candidate products, and selecting the i-th output product information from the sample target products generated for the sample user. Since there may be more than one sample target product, meaning there may be multiple sample candidate products whose numerical representation information is more similar to the numerical representation information of the i-th output product than the first preset similarity, in this case, one product can be selected from the sample target products as the i-th output product in the sample gift package generated for the sample user. Specifically, the sample target product with the largest numerical representation information can be selected as the i-th output product, such as... Figure 5 As shown, a probability sampling algorithm can also be used to select one of the target products from each sample as the i-th output product (i.e., Figure 5 (The output results in the algorithm). Probability sampling algorithms can also be understood as probability sampling algorithms, which are a way of sampling samples according to a specific probability and a random principle.

[0096] This implementation method uses numerical encoding of user profile information and product information, and uses numerical information such as numerical features and numerical representation information for model training. This makes the training data used in the model training process simpler and the computation more efficient and faster, thereby making the model training more efficient.

[0097] In one specific embodiment, the first numerical feature can be obtained by numerical encoding in step S121 as follows: the sample user profile information and the 0th to i-1th output product information are input into the first encoding model for numerical encoding to obtain the first numerical feature.

[0098] The aforementioned first encoding model can be a pre-trained encoding model, which simplifies the training process of this application. Alternatively, the first encoding model can be an untrained encoding model. Since the sample user profile information and the output product information are complex, the pre-trained first encoding model may not achieve the required encoding accuracy. Therefore, both pre-trained and untrained first encoding models can be further trained during the training process of the gift package generation model provided in this application.

[0099] Optionally, the above training method may further include the following step S160.

[0100] Step S160: Adjust the model parameters in the first encoding model according to the principle of increasing the intention parameter of sample users to purchase sample gift packages, until the intention parameter meets the above preset conditions, and obtain the optimized first encoding model.

[0101] The parameter training of the first encoding model in step S160 is similar to the parameter training process of the gift package generation model to be trained in step S150, and will not be described again here.

[0102] This embodiment utilizes a first encoding model for convenient and efficient numerical encoding, making the training method for the gift package generation model more convenient and efficient. Furthermore, this embodiment trains the first encoding model based on the sample users' intention to purchase sample gift packages, resulting in more accurate numerical features output by the first encoding model. This allows the gift package generation model trained in this application and the first encoding model to work together to generate gift packages during subsequent gift package recommendations, improving the accuracy of gift package recommendations.

[0103] In one embodiment, prior to step S130, the training method may further include the following step S120a.

[0104] Step S120a: Based on the sample user profile information, the information of the first to i output products, and the quantity of the output products from the 0th to i-1th output products, determine the second input information of the gift pack generation model to be trained, and input the second input information into the gift pack generation model to be trained to obtain the quantity of the i output product.

[0105] It is understandable that step S120a is executed after obtaining the i-th output product information in step S120. Similar to step S120, in step S120a, the sample user profile information and the 0th to 1st output product information can be directly determined as the second input information; alternatively, the sample user profile information and the 0th to 1st output product information can be preprocessed, and the preprocessed information can be determined as the second input information.

[0106] Since i iterates through 1 to N, the quantity of the first to Nth output items can be obtained through step S120a.

[0107] Since the information for the 0th output product is empty, the quantity of the 0th output product is also empty. Figure 2 As shown, an exemplary execution process for step S120a can be: [The process involves] processing the sample user profile information (i.e.,...] Figure 2 (All characteristics of the player to be recommended) and the first output item information (i.e. Figure 2 The second input information, obtained from the product ID of product 1, is input into the gift pack generation model to be trained (i.e., Figure 2 In the gift pack content generator, the quantity of the first output item in the sample gift pack is obtained (i.e., Figure 2 The quantity of item 1 in the sample gift pack is calculated as follows: The second input information, obtained from the sample user profile information, the information of the first to second output items, and the quantity of the first output item, is then input into the gift pack generation model to be trained to obtain the quantity of the second output item in the sample gift pack. The second input information, obtained from the sample user profile information, the information of the first to third output items, and the quantity of the first to second output items, is then input into the gift pack generation model to be trained to obtain the quantity of the third output item in the sample gift pack. This process continues until the quantity of the Nth output item in the sample gift pack is obtained.

[0108] When i is greater than 1, the first input information in step S120 can be determined according to the following step S124.

[0109] Step S124: Determine the first input information of the gift pack generation model to be trained based on the sample user profile information, the information of the 0th to i-1th output products, and the quantity of the 0th to i-1th output products.

[0110] When i is greater than 1, step S120 is to determine the 2nd, 3rd, ..., Nth i-th output product information.

[0111] The method for determining the first input information in step S124 is similar to that in step S120, and will not be repeated here.

[0112] In this embodiment, the process of determining the information of the first to Nth output products and the quantity of the first to Nth output products can be as follows: If i is 1, execute step S120 to determine the information of the first output product, and then execute step S120a to obtain the quantity of the i-th output product; if i is 2, execute step S120 to determine the information of the second output product, and then execute step S120a to obtain the quantity of the second output product; ...; if i is N, execute step S120 to determine the information of the Nth output product, and then execute step S120a to obtain the quantity of the N-th output product. At this time, i has traversed from 1 to N, and the information of the first to Nth output products and the quantity of the first to Nth output products are obtained.

[0113] Step S130 can be implemented by following step S131.

[0114] Step S131: Generate sample gift packs for sample users based on the information of the first to Nth output products and the quantity of the first to Nth output products.

[0115] In step S131, if N is 3, the quantity of the first output item is 2, the quantity of the second output item is 1, and the quantity of the third output item is 4, then the sample gift pack may include 2 first output items, 1 second output item, and 4 third output items.

[0116] This implementation generates the quantity of each output item through a gift pack generation model. In other words, during model training, the quantity of each item in the gift pack is considered. When inputting the model, not only is the specific item considered, but also the quantity of each item. This allows the trained gift pack generation model to accurately predict the quantity of each item in the push gift packs sent to users. Through multi-dimensional prediction, the generated push gift packs are more in line with user preferences, further improving user game satisfaction and user experience.

[0117] In one implementation, in step S120a, the second input information can be determined according to the following step S120b.

[0118] Step S120b: Numerically encode the sample user profile information, the information of the first to i output products, and the quantity of the output products from the 0th to i-1th output products to obtain the second numerical feature, and determine the second numerical feature as the second input information of the gift pack generation model to be trained.

[0119] The method of numerical encoding in step S120b is similar to that in step S121, and will not be repeated here.

[0120] In step S120a, the quantity of the i-th output item can be obtained by following steps S120c to S120d.

[0121] Step S120c: Input the second input information into the gift pack generation model to be trained to obtain numerical representation information for representing the quantity of the i-th output item.

[0122] The second input information in step S120b is the second numerical feature. Therefore, after being input into the gift pack generation model to be trained, it is also very convenient to obtain the numerical representation information of the quantity of the i-th output item in numerical form.

[0123] Step S120d: Determine the quantity of the i-th output item based on the numerical representation information of the quantity of the i-th output item.

[0124] In step S120d, the numerical representation information of the quantity of the i-th output product can be decoded by the numerical decoding method corresponding to the numerical encoding method, thereby obtaining the quantity of the i-th output product.

[0125] Alternatively, a target sample quantity can be selected from the preset sample quantities as the quantity of the i-th output item in the sample gift pack generated for the sample user. The similarity between the numerical representation information corresponding to the target sample quantity and the numerical representation information of the quantity of the i-th output item is greater than a second preset similarity. Each preset sample quantity has corresponding numerical representation information. The second preset similarity and the first preset similarity can be the same or different.

[0126] This implementation method uses numerical encoding of user profile information, product information, and product quantity, and uses numerical information such as numerical features and numerical representation information for model training. This makes the training data used in the model training process simpler and the computation more efficient and faster.

[0127] In one implementation, the aforementioned sample user's historical purchase information may specifically include: a sequence of historical purchase items ordered by purchase time.

[0128] Specifically, the historical purchase item sequence can be a sequence of items sorted in chronological order from front to back or back to front. For example, the historical purchase item sequence could be [x:1, y:2, z:1], indicating that the user purchased 1 item x, 2 items y, and 1 item z in chronological order.

[0129] By inputting the historical purchase information of sample users into the gift pack generation model to be trained in a sequential manner, the model can learn which items were recently purchased by users and learn the changing trends of users' purchasing habits over time. This allows the gift pack generation model to be based on more dimensions when generating gift packs, resulting in higher accuracy of the generated gift packs.

[0130] Optionally, if the sample user's historical purchase information includes the sample user's historical purchase of gift packs and the sample user's historical purchase of goods, the aforementioned historical purchase item sequence (i.e., the sample user's historical purchase item information) may specifically include: each item included in the sample user's historical purchase of gift packs and the sample product sequence after the sample user's historical purchase of goods is sorted according to the purchase time. For example, the historical purchase item sequence is [x:1,y:2,z:1], and the historical purchase gift pack sequence is [x:3;y:1,y:1;z:1]. [x:3;y:1,y:1;z:1] indicates that the user bought two gift packs in chronological order. The first gift pack contains 3 x items and 1 y item, and the second gift pack contains 1 y gift pack and 1 z gift pack. If the sample user purchased each sample gift pack later than the sample item, then the sequence of items included in the historical purchase gift packs and the sample items after sorting the historical purchase items by purchase time is [x:1,y:2,z:1,x:3,y:1,y:1,z:1].

[0131] In the embodiments of this application, such as Figure 4 As shown, when encoding the sample user profile information, we can first encode the various information included in the sample user's historical purchase information and the product information of the output products already generated in the sample gift pack (i.e., Figure 4 The generated product sequence (i.e., the previously generated product sequence) is input into the self-attention network layer for encoding. Then, the encoding result of the self-attention network, along with the attribute description information of the sample user, is input into the fully connected network encoding layer for encoding, thus obtaining the encoded first numerical feature. Multi-layer encoding makes the encoding result of the optimized and trained encoding model more accurate, thereby making the push of gift packages more accurate. The information included in the sample user's historical purchase information can include, for example, the aforementioned sample product sequence (i.e., the previously generated product sequence). Figure 4 The new product sequence in the sample gift pack sequence mentioned above (i.e.) Figure 4 The hashed sequence of gift packs), and the sorted sequence of items in historically purchased gift packs (i.e., Figure 4 (The user's historical purchase sequence).

[0132] This embodiment sorts the products included in the historical purchase gift packs and all the products purchased in the past in a unified manner. The resulting sample product sequence can more comprehensively reflect the sample user's historical product purchases, thereby enabling the gift pack generation model to learn the user's purchase behavior for each product and making the trained gift pack generation model more accurate.

[0133] Optionally, the aforementioned historical purchase item sequence may further include: a sample gift package sequence after the sample user's historical purchase gift packages are sorted by purchase time. In this embodiment, the sample gift package sequence may include the sorting result of the identification information of each historical purchase gift package in order. The identification information may be a hash value, gift package ID, gift package name, etc., or other forms of identification information. For example, the sequence formed by hashing the historical purchase gift package sequence [x:3; y:1, y:1; z:1] is [00001, 00002].

[0134] In this embodiment, the entire gift package is used as input to train the gift package generation model, which allows the model to learn the user's purchase behavior of gift packages. This multi-dimensional and multi-angle training further improves the accuracy of the trained gift package generation model.

[0135] In one embodiment, the above training method may further include the following steps S170 to S180.

[0136] Step S170: When the preset timing is reached, obtain the latest sample user profile information corresponding to each sample user.

[0137] The time elapsed since the latest sample user profile information was generated is less than a preset time interval.

[0138] The aforementioned preset timing can be a specific time every two days, for example, 12:00 noon every two days. The preset timing can also be when an update training instruction is received, which can be triggered by staff. The preset timing can also be other times, which are not specifically limited in this application.

[0139] The aforementioned preset time interval can be any interval between 10 hours and 3 days, or other time intervals, and this application does not specifically limit it.

[0140] Step S180: Use the latest sample user profile information to update and train the trained gift pack generation model to obtain the updated gift pack generation model.

[0141] The specific process of updating and training in step S180 is the same as the process of training the model in steps S120 to S150, and will not be repeated here.

[0142] This implementation method can update and train the gift pack generation model at regular or irregular intervals, making the real-time performance of the gift pack generation model better and thus improving its accuracy.

[0143] The second embodiment of this application provides a method for pushing gift packages. The subject executing this method can be an electronic device, such as a desktop computer, laptop computer, tablet computer, mobile phone, server, or other electronic devices with data processing capabilities. This application does not specifically limit the type of device. In application scenarios such as video games and e-commerce, since gift package pushes are usually sent from the server to the client, the following embodiment uses the server as the subject of execution to illustrate the gift package push method.

[0144] like Figure 7 As shown, the gift package push method provided in the second embodiment of this application includes the following steps S210 to S240.

[0145] Step S210: Obtain the user profile information of the user to be pushed to.

[0146] In step S210, the user profile information of the user to be pushed to can be obtained when a preset timing is met. This preset timing can be at least one of the following: receiving a gift package push request from the client corresponding to the user to be pushed to, the user's continuous online time reaching a preset duration, or the user's level changing. Those skilled in the art can also design other preset timings according to actual scenario requirements; this application does not specifically limit this.

[0147] The aforementioned user profile information to be pushed may include at least one of the following: information on the user's historical purchases, and the user's attribute description information; information on the user's historical purchases may include at least one of the following: information on the user's historical purchases of gift packs, and information on the user's historical purchases of goods.

[0148] When the sample user profile information during the training of the gift pack generation model in the first embodiment includes the items included in the sample user's historical purchase gift packs and the sample item sequence of the sample user's historical purchase items sorted by purchase time, the historical purchase item information of the user to be pushed may specifically include: the items included in the historical purchase gift packs of the user to be pushed and the sample item sequence of the user's historical purchase items sorted by purchase time.

[0149] The aforementioned information on the historical purchases of users to be pushed to may also include a sequence of sample gift packs ordered by purchase time based on the historical purchases of the users to be pushed to.

[0150] The attribute description information of the user to be pushed may include at least one of the following: user level, average game spending amount, equipment value, online time, gender, age, and geographical location, but is not limited to these.

[0151] The specific content and related effects of the user profile information to be pushed are similar to the content and related effects of the sample user profile information in the first embodiment, and will not be repeated here.

[0152] Step S220: Determine the third input information of the trained gift pack generation model based on the user profile information to be pushed and the information of the 0th to j-1th products to be pushed, and input the third input information into the trained gift pack generation model to obtain the information of the jth product to be pushed in the push gift pack generated for the user to be pushed.

[0153] Where j is an integer, j iterates from 1 to T, T is the number of preset product types contained in the push gift pack, the information of the 0th product to be pushed is empty, and the trained gift pack generation model is trained by any of the gift pack generation model training methods provided in the first embodiment.

[0154] Step S230: Generate push gift packs for the users to be pushed based on the information of the first to T products to be pushed.

[0155] The specific execution process of steps S220 and S230 is similar to that of steps S120 and S130 in the first embodiment, and will not be repeated here. In step S220, the value of T can be the same as or different from the value of N in the first embodiment. Those skilled in the art can flexibly set the value of T according to the number of product types that need to be included in the push package, and this application does not specifically limit it.

[0156] Step S240: Push the push gift pack to the users to be pushed.

[0157] In step S240, the push notification package can be sent to the client corresponding to the user to be pushed to. The client can display the push notification package on the game interface to prompt the user to purchase the push notification package.

[0158] When the gift pack generation model trained using the training method provided in this application pushes gift packs to users, the j-th item to be pushed in the gift pack is obtained by inputting the user profile information and the (j-1)-th item information into the gift pack generation model. j iterates through 1 to T, where T is the number of preset item types in the push gift pack. Thus, the j-th item to be pushed in the push gift pack can be understood as the user liking the j-th item in addition to liking the 0 to (j-1)-th items. Therefore, the items in the push gift pack generated for the user based on the 1 to T-th item information are more in line with the user's preferences. After the push gift pack is pushed to the user, the user will have a higher satisfaction with each item in the pushed gift pack, increasing the probability that the user likes the items in the pushed gift pack, enabling the user to buy the gift pack they like more likely, improving the user's game satisfaction and user experience, thereby improving the user retention rate.

[0159] In addition, the embodiments of this application can generate personalized gift packs that match the user's preferences based on the user's profile information, instead of selecting gift packs to push to the user from a fixed gift pack pool. This can further ensure that the gift packs pushed to the user match the user's preferences, thereby further improving the user's game satisfaction and user experience.

[0160] Optionally, in step S220, the third input information can be determined according to the following step S221.

[0161] Step S221: Numerically encode the user profile information to be pushed and the information of the 0th to j-1th product to be pushed to obtain the third numerical feature, and determine the third numerical feature as the third input information of the trained gift pack generation model.

[0162] In step S221, the user profile information to be pushed and the information of the 0th to ijth products to be pushed can be input into the second encoding model for numerical encoding to obtain the third numerical feature. The second encoding model used in step S221 can refer to the first encoding model in the first embodiment. Alternatively, the optimized first encoding model obtained in the first embodiment can be used as the second encoding model in step S221. Since the optimized first encoding model has undergone the optimization training in the first embodiment, the pushed gift package can be more in line with user preferences.

[0163] In step S220, the information of the j-th product to be pushed can be obtained by following the steps S222 to S223.

[0164] Step 222: Input the third input information into the trained gift pack generation model to obtain the numerical representation information of the j-th item to be pushed in the push gift pack generated for the user to be pushed.

[0165] The first input information for the trained gift package generation model during training is the aforementioned first numerical feature.

[0166] Step 223: Determine the information of the j-th product to be pushed in the push gift package generated by the user to be pushed, based on the numerical representation information of the j-th product to be pushed.

[0167] The execution process of steps S221 to S223 is similar to that of steps S121 to S123 in the first embodiment, and will not be described again here.

[0168] Optionally, step S223 can be implemented as follows: step S223a.

[0169] Step S223a: Select the target product from the pre-stored candidate products as the j-th product information to be pushed in the push gift package generated for the user to be pushed.

[0170] Wherein, the similarity between the numerical representation information corresponding to the target product and the numerical representation information of the j-th product to be pushed is greater than a third preset similarity, and the candidate product is provided with corresponding numerical representation information.

[0171] The aforementioned third preset similarity can be the same as or different from the first preset similarity in the first embodiment; this application does not specifically limit this. Each candidate product can be a virtual product designed in a game, or a virtual product designed in an e-commerce system; this application does not specifically limit this.

[0172] Optionally, step S223a can be implemented in the following steps: selecting at least one target product from the pre-stored candidate products; selecting the j-th product information to be pushed from the push gift package generated for the user to be pushed from each of the target products based on a probability sampling algorithm.

[0173] The specific execution process of step S223a can be referred to step S123a in the first embodiment, and will not be repeated here.

[0174] In one specific embodiment, step S230 can be implemented according to the following steps S232 to S235.

[0175] Step S232: Perform the step of inputting the third input information into the trained gift pack generation model in step S220 S times to obtain the first to T items to be pushed in group S.

[0176] Step S233: Generate S candidate pushable gift packages for the user to be pushed to based on the output product information of the first to T items in group S.

[0177] In this embodiment, since step S223a selects the j-th product information to be pushed from each target product through a probability sampling algorithm, and the probability sampling algorithm randomly selects one product from each target product as the product to be pushed, if step S220, which inputs the third input information into the trained gift pack generation model, is executed S times, S j-th product information to be pushed will usually be obtained. These S j-th product information to be pushed are usually different, or at least partially different. Therefore, when j traverses 1 to T, the S groups of 1 to T product information to be pushed will usually also be different, and the S candidate gift packs to be pushed generated based on the S groups of 1 to T output product information will also be different.

[0178] Step S234: Determine the intention parameters of the users to be pushed to purchase each of the candidate push packages, and obtain S intention parameters.

[0179] Specifically, the user profile information to be pushed to can be input into a trained probability calculation model or gift package scoring model along with the candidate gift packages to be pushed to obtain the probability of the user to be pushed to purchase the candidate gift package, or the rating of the user to be pushed to purchase the candidate gift package. The method of determining the intention parameter in step S234 is similar to the method of determining the intention parameter in step S140 of the first embodiment, and will not be described again here.

[0180] Step S235: Generate a push notification package for the user to be pushed to based on the S intention parameters.

[0181] Specifically, the push gift package with the highest intention parameter can be selected from the S candidate push gift packages as the push gift package generated for the user to be pushed to. This allows for a simple and convenient determination of the push gift package, and the determined push gift package is more in line with the user's expectations.

[0182] For example, such as Figure 6 As shown, when n candidate gift packs are obtained (namely, candidate gift pack 1, ..., candidate gift pack n), which are the above S candidate gift packs to be pushed, a gift pack scoring model can be used to score each candidate gift pack, and the one with the highest score is determined as the push gift pack.

[0183] In one specific embodiment, the expected revenue corresponding to each of the S candidate pushable gift packages can be determined based on the gift package price corresponding to the S candidate pushable gift packages and the S intention parameters. The gift package with the highest expected revenue can then be selected from the S candidate pushable gift packages as the pushable gift package generated for the user to be pushed.

[0184] The price of a candidate push notification package can be determined by the sum of the prices of each item in the package, and the price of each item is usually pre-set.

[0185] The expected return for a candidate gift package can be the product of the intended parameter and the price of the candidate gift package.

[0186] In this embodiment, the gift package with the highest expected benefit is used as the push gift package generated for the user to be pushed, which can make the game or other software system more profitable.

[0187] Optionally, before step S230, the pushing method may further include the following step S220a.

[0188] Step S220a: Based on the user profile information to be pushed, the information of the first to j items to be pushed, and the quantity of the items of the 0 to j-1 items to be pushed, determine the fourth input information of the trained gift pack generation model, and input the fourth input information into the trained gift pack generation model to obtain the quantity of the j items to be pushed.

[0189] When j is greater than 1, the third input information in step S220 can be implemented according to the following step S224.

[0190] Step S224: Determine the third input information of the trained gift pack generation model based on the user profile information to be pushed, the information of the 0th to j-1th products to be pushed, and the quantity of the 0th to j-1th products to be pushed.

[0191] Step S230 can be implemented by following step S231.

[0192] Step S231: Generate a push gift pack for the user to be pushed based on the first to T pieces of information about the products to be pushed and the quantity of the first to T pieces of information about the products to be pushed.

[0193] The execution process of steps S220a, S224 and S231 is similar to that of steps S120a, S124 and S131 in the first embodiment, and will not be described again here.

[0194] Optionally, in step S220a, the fourth input information can be determined according to the following step S220b.

[0195] Step S220b: The user profile information to be pushed, the information of the first to j items to be pushed, and the quantity of the first to j-1 items to be pushed are numerically encoded to obtain a fourth numerical feature, and the fourth numerical feature is determined as the fourth input information of the trained gift pack generation model.

[0196] In step S220a, the quantity of the j-th product to be pushed can be obtained by following the steps S220c to S220d.

[0197] Step S220c: Input the fourth input information into the trained gift package generation model to obtain numerical representation information for representing the quantity of the j-th product to be pushed.

[0198] Step S220d: Determine the quantity of the j-th product to be pushed based on the numerical representation information of the quantity of the j-th product to be pushed.

[0199] The execution process of steps S220b to S220d can be referred to steps S120b to S120d in the first embodiment, and will not be repeated here.

[0200] The third embodiment of this application also provides a gift package push system, such as Figure 3 As shown, the system includes a data module, a training module, and a service module.

[0201] The data module is used to retrieve user data from user logs (such as user game logs) and preprocess the user data to obtain a training set (i.e., ...). Figure 3 The training module is used to train the gift pack generation model and encoding model based on the training set obtained from the data module. Specifically, the training module executes the steps in the first embodiment. After training the gift pack generation model, the training module can convert it into a .pb file format or other file formats suitable for storage and save it to solidify the trained gift pack generation model. The service module uses the gift pack generation model trained by the training module to provide a gift pack push service. Specifically, the service module can execute the steps in the second embodiment. For example, the service module can obtain user data from the game client, extract user profile information and purchase sequences from the user data, determine the push gift pack by calling the gift pack push service interface (API), and then push the obtained push gift pack to the game client.

[0202] Corresponding to the gift pack generation model training method provided in the first embodiment of this application, the fourth embodiment of this application provides a gift pack generation model training device, such as... Figure 8 As shown, the training device includes:

[0203] The first acquisition unit 310 is used to acquire a training set, which includes sample user profile information corresponding to each sample user.

[0204] The first calculation unit 320 is used to determine the first input information of the gift pack generation model to be trained based on the sample user profile information and the 0th to i-1th output product information, and input the first input information into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty;

[0205] The first generation unit 330 is used to generate sample gift packs for the sample users based on the first to Nth output product information.

[0206] The determining unit 340 is used to determine the sample user's intention parameters to purchase the sample gift package;

[0207] The adjustment unit 350 is used to adjust the model parameters of the gift pack generation model to be trained, based on the principle of improving the intention parameters, until the intention parameters meet the preset conditions, and thus obtain the trained gift pack generation model.

[0208] Optionally, the first computing unit is specifically used to: numerically encode the sample user profile information and the 0th to i-1th output product information to obtain a first numerical feature, and determine the first numerical feature as the first input information of the gift package generation model to be trained; input the first input information into the gift package generation model to be trained to obtain the numerical representation information of the i-th output product in the sample gift package generated for the sample user; and determine the i-th output product information in the sample gift package generated for the sample user based on the numerical representation information of the i-th output product.

[0209] Optionally, the first calculation unit is specifically used to: select a sample target product from each pre-stored sample candidate product as the i-th output product information in the sample gift package generated for the sample user, wherein the similarity between the numerical representation information corresponding to the sample target product and the numerical representation information of the i-th output product information is greater than a first preset similarity, and the sample candidate product is provided with corresponding numerical representation information.

[0210] Optionally, the first calculation unit is further configured to: determine the second input information of the gift pack generation model to be trained based on the sample user profile information, the first to i output product information and the product quantity of the 0 to i-1 output products, and input the second input information into the gift pack generation model to be trained to obtain the product quantity of the i output product;

[0211] The first calculation unit is specifically used to: when i is greater than 1, determine the first input information of the gift pack generation model to be trained based on the sample user profile information, the information of the 0th to i-1th output products and the quantity of the 0th to i-1th output products;

[0212] The first generation unit is specifically used to generate sample gift packs for the sample users based on the first to Nth output product information and the quantity of the first to Nth output products.

[0213] Optionally, the first calculation unit is specifically used to: numerically encode the sample user profile information, the information of the first to i output products, and the quantity of the 0th to i-1th output products to obtain a second numerical feature, and determine the second numerical feature as the second input information of the gift pack generation model to be trained; input the second input information into the gift pack generation model to be trained to obtain numerical representation information for representing the quantity of the i-th output product; and determine the quantity of the i-th output product based on the numerical representation information of the quantity of the i-th output product.

[0214] Optionally, the first calculation unit is specifically used to: input the sample user profile information and the 0th to i-1th output product information into the first encoding model for numerical encoding to obtain the first numerical feature.

[0215] Optionally, the adjustment unit is further configured to: adjust the model parameters in the first encoding model based on the principle of improving the intention parameter, until the intention parameter meets the preset condition, thereby obtaining the first encoding model after optimized training.

[0216] Optionally, the determining unit is specifically used to: input the sample user profile information and the sample gift package into a trained probability calculation model to obtain the probability that the sample user will purchase the sample gift package.

[0217] Optionally, the sample user profile information includes at least one of the following: sample user's historical purchase information, and sample user's attribute description information;

[0218] The sample user's historical purchase information includes at least one of the following: sample user's historical purchase information of gift packs and sample user's historical purchase information of goods, wherein the gift pack includes at least one good product.

[0219] Optionally, when the sample user's historical purchase information includes the sample user's historical purchase of gift packs and the sample user's historical purchase of goods, the sample user's historical purchase information specifically includes: each item included in the sample user's historical purchase of gift packs and the sample product sequence after the sample user's historical purchase of goods is sorted by purchase time.

[0220] Optionally, the sample user's historical purchase information may also include: a sequence of sample gift packages ordered by purchase time based on the sample user's historical purchases.

[0221] Optionally, the attribute description information of the sample users includes at least one of the following: sample user level, sample user's average game spending, sample user's equipment value, sample user's online time information, sample user's gender information, sample user's age information, and sample user's geographical location.

[0222] Optionally, the first acquisition unit is further configured to: when a preset time is reached, acquire the latest sample user profile information corresponding to each sample user, wherein the generation time of the latest sample user profile information is less than a preset time interval from the current time;

[0223] The training device also includes:

[0224] The update unit is used to update the trained gift pack generation model using the latest sample user profile information to obtain the updated gift pack generation model.

[0225] Corresponding to the gift pack push method provided in the second embodiment of this application, the fifth embodiment of this application provides a gift pack push device, such as... Figure 4 As shown, the pushing device includes:

[0226] The second acquisition unit is used to acquire the user profile information of the user to be pushed to;

[0227] The second calculation unit is used to determine the third input information of the trained gift pack generation model based on the user profile information to be pushed and the 0th to j-1th product information to be pushed, and input the third input information into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed, where j is an integer, j traverses 1 to T, T is the number of preset product types contained in the push gift pack, and the 0th product information to be pushed is empty. The trained gift pack generation model is trained by any of the gift pack generation model training methods provided in the first embodiment.

[0228] The second generation unit is used to generate a push gift pack for the user to be pushed based on the first to T pieces of information about the products to be pushed.

[0229] The push unit is used to push the push gift package to the user to be pushed.

[0230] Optionally, the second calculation unit is specifically used to: numerically encode the user profile information to be pushed and the information of the 0th to j-1th product to be pushed to obtain a third numerical feature, and determine the third numerical feature as the third input information of the trained gift pack generation model; input the third input information into the trained gift pack generation model to obtain the numerical representation information of the jth product to be pushed in the push gift pack generated for the user to be pushed;

[0231] Based on the numerical representation information of the j-th product to be pushed, it is determined to be the j-th product information in the push gift package generated by the user to be pushed.

[0232] Optionally, the second calculation unit is specifically used to: select a target product from the pre-stored candidate products as the j-th product information to be pushed in the push package generated for the user to be pushed, wherein the similarity between the numerical representation information corresponding to the target product and the numerical representation information of the j-th product to be pushed is greater than a second preset similarity, and the candidate products are provided with corresponding numerical representation information.

[0233] Optionally, the second calculation unit is further configured to: determine the fourth input information of the trained gift pack generation model based on the user profile information to be pushed, the information of the 1st to jth items to be pushed, and the quantity of the 0th to j-1th items to be pushed, and input the fourth input information into the trained gift pack generation model to obtain the quantity of the jth item to be pushed; when j is greater than 1, determine the third input information of the trained gift pack generation model based on the user profile information to be pushed, the information of the 0th to j-1th items to be pushed, and the quantity of the 0th to j-1th items to be pushed;

[0234] The second generation unit is specifically used to generate a push gift pack for the user to be pushed based on the first to T pieces of information about the products to be pushed and the quantity of the first to T pieces of information about the products to be pushed.

[0235] Optionally, the second calculation unit is specifically used to: numerically encode the user profile information to be pushed, the information of the first to j items to be pushed, and the quantity of the first to j-1 items to be pushed, to obtain a fourth numerical feature, and determine the fourth numerical feature as the fourth input information of the trained gift pack generation model; input the fourth input information into the trained gift pack generation model to obtain numerical representation information for representing the quantity of the j-th item to be pushed; and determine the quantity of the j-th item to be pushed based on the numerical representation information of the quantity of the j-th item to be pushed.

[0236] Optionally, the second calculation unit is specifically used to: select at least one target product from the pre-stored candidate products; and select the j-th product information to be pushed from the push gift package generated for the user to be pushed, based on a probability sampling algorithm.

[0237] Optionally, the second generation unit is specifically used to perform step S times of inputting the third input information into the trained gift package generation model to obtain the first to T items to be pushed in S groups; generate S candidate gift packages to be pushed for the user based on the first to T items in S groups; determine the intention parameters of the user to purchase each of the candidate gift packages to be pushed, to obtain S intention parameters; and generate a push gift package for the user based on the S intention parameters.

[0238] Optionally, the second generation unit is specifically used to select the gift package with the largest intention parameter from the S candidate gift packages to be pushed as the push gift package generated for the user to be pushed; or, based on the gift package price corresponding to the S candidate gift packages to be pushed and the S intention parameters, determine the expected benefits corresponding to the S candidate gift packages to be pushed respectively, and select the gift package with the largest expected benefit from the S candidate gift packages to be pushed as the push gift package generated for the user to be pushed.

[0239] Optionally, the second calculation unit is specifically used to: input the user profile information to be pushed and the information of the 0th to ijth products to be pushed into the second encoding model for numerical encoding to obtain the third numerical feature, wherein the second encoding model is the first encoding model after optimization training in the first embodiment.

[0240] Corresponding to the gift pack generation model training method provided in the first embodiment of this application, the sixth embodiment of this application also provides an electronic device for training a gift pack generation model. For example... Figure 9 As shown, the electronic device includes: a processor 401; and a memory 402 for storing a program for training a gift package generation model. After the device is powered on and the program for training the gift package generation model is run by the processor, the following steps are performed:

[0241] Obtain a training set, which includes sample user profile information corresponding to each sample user;

[0242] Based on the sample user profile information and the 0th to i-1th output product information, the first input information of the gift pack generation model to be trained is determined, and the first input information is input into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty;

[0243] Based on the first to Nth output product information, a sample gift package is generated for the sample user;

[0244] Determine the sample users' intention parameters to purchase the sample gift package;

[0245] The model parameters of the gift pack generation model to be trained are adjusted according to the principle of improving the intention parameters until the intention parameters meet the preset conditions, thus obtaining a trained gift pack generation model.

[0246] Corresponding to the gift package push method provided in the second embodiment of this application, the seventh embodiment of this application also provides an electronic device for pushing gift packages. The electronic device includes: a processor; and a memory for storing a program for the gift package push method. After the device is powered on and the program for the gift package push method is run through the processor, the following steps are performed:

[0247] Obtain the user profile information of the user to be pushed to;

[0248] The third input information of the trained gift pack generation model is determined based on the user profile information to be pushed and the information of the 0th to j-1th product to be pushed. The third input information is then input into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed. Here, j is an integer, j iterates from 1 to T, and T is the number of preset product types included in the push gift pack. The 0th product information to be pushed is empty. The trained gift pack generation model is trained by the gift pack generation model training method described in any one of the first embodiments.

[0249] Based on the first to T items of information to be pushed, a push gift pack is generated for the user to be pushed;

[0250] The push notification package will be sent to the user to be pushed to.

[0251] Corresponding to the gift package generation model training method provided in the first embodiment of this application, the eighth embodiment of this application provides a computer-readable storage medium storing a program for the gift package generation model training method. This program is executed by a processor to perform the following steps:

[0252] Obtain a training set, which includes sample user profile information corresponding to each sample user;

[0253] Based on the sample user profile information and the 0th to i-1th output product information, the first input information of the gift pack generation model to be trained is determined, and the first input information is input into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty;

[0254] Based on the first to Nth output product information, a sample gift package is generated for the sample user;

[0255] Determine the sample users' intention parameters to purchase the sample gift package;

[0256] The model parameters of the gift pack generation model to be trained are adjusted according to the principle of improving the intention parameters until the intention parameters meet the preset conditions, thus obtaining a trained gift pack generation model.

[0257] Corresponding to the gift package push method provided in the second embodiment of this application, the ninth embodiment of this application provides a computer-readable storage medium storing a program for the gift package push method. This program is executed by a processor to perform the following steps:

[0258] Obtain the user profile information of the user to be pushed to;

[0259] The third input information of the trained gift pack generation model is determined based on the user profile information to be pushed and the information of the 0th to j-1th product to be pushed. The third input information is then input into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed. Here, j is an integer, j iterates from 1 to T, and T is the number of preset product types included in the push gift pack. The 0th product information to be pushed is empty. The trained gift pack generation model is trained by the gift pack generation model training method described in any one of the first embodiments.

[0260] Based on the first to T items of information to be pushed, a push gift pack is generated for the user to be pushed;

[0261] The push notification package will be sent to the user to be pushed to.

[0262] It should be noted that for a detailed description of the gift pack push method, gift pack generation model training device, gift pack push device, electronic device and computer-readable storage medium provided in the embodiments of this application, please refer to the relevant description of the gift pack generation model training method in the first embodiment of this application, which will not be repeated here.

[0263] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

[0264] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0265] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0266] 1. Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, 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 technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage media, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include non-transitory computer-readable media, such as modulated data signals and carrier waves.

[0267] 2. Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0268] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A method for training a gift package generation model, characterized in that, The training method includes: Obtain a training set, which includes sample user profile information corresponding to each sample user; Based on the sample user profile information and the 0th to i-1th output product information, the first input information of the gift pack generation model to be trained is determined, and the first input information is input into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty; Based on the first to Nth output product information, a sample gift package is generated for the sample user; Determine the intention parameters of sample users to purchase the sample gift package; the intention parameters are represented by the probability of the sample users purchasing the sample gift package obtained by the sample user profile information and the sample gift package input probability calculation model; The model parameters of the gift pack generation model to be trained are adjusted according to the principle of improving the intention parameters until the intention parameters meet the preset conditions, so as to obtain the trained gift pack generation model. The step of determining the first input information for the gift pack generation model to be trained based on the sample user profile information and the information of the 0th to i-1th output products includes: The sample user profile information and the 0th to i-1th output product information are numerically encoded to obtain the first numerical feature, and the first numerical feature is determined as the first input information of the gift pack generation model to be trained. The step of inputting the first input information into the training model to obtain the i-th output product information in the sample gift package generated for the sample user includes: The first input information is input into the training gift pack generation model to obtain the numerical representation information of the i-th output product in the sample gift pack generated for the sample user; The information of the i-th output product is determined based on the numerical representation information of the i-th output product, which is the information of the i-th output product in the sample gift package generated by the sample user.

2. The training method according to claim 1, characterized in that, The step of determining the information of the i-th output product in the sample gift package generated by the sample user based on the numerical representation information of the i-th output product includes: Select a target product from the pre-stored candidate products as the i-th output product information in the sample gift package generated for the sample user, wherein the similarity between the numerical representation information corresponding to the target product and the numerical representation information of the i-th output product information is greater than a first preset similarity, and the candidate products are provided with corresponding numerical representation information.

3. The training method according to claim 1, characterized in that, Before generating sample gift packs for the sample users based on the first to Nth output product information, the training method further includes: Based on the sample user profile information, the information of the first to i output products, and the quantity of the output products from the 0th to the (i-1)th output products, the second input information of the gift pack generation model to be trained is determined, and the second input information is input into the gift pack generation model to be trained to obtain the quantity of the i-th output product. When i is greater than 1, determining the first input information of the gift package generation model to be trained based on the sample user profile information and the information of the 0th to i-1th output products includes: Based on the sample user profile information, the information of the 0th to i-1th output products, and the quantity of the 0th to i-1th output products, the first input information of the gift pack generation model to be trained is determined; The step of generating sample gift packs for the sample users based on the first to Nth output product information includes: Based on the first to Nth output product information and the quantity of the first to Nth output products, a sample gift pack is generated for the sample user.

4. The training method according to claim 3, characterized in that, The step of determining the second input information of the gift pack generation model to be trained based on the sample user profile information, the information of the first to i output products, and the quantity of the 0th to i-1th output products includes: The sample user profile information, the first to i output product information and the product quantity of the 0th to i-1th output products are numerically encoded to obtain the second numerical feature, and the second numerical feature is determined as the second input information of the gift pack generation model to be trained; The step of inputting the second input information into the training model to obtain the quantity of the i-th output item includes: The second input information is input into the training gift pack generation model to obtain numerical representation information for the quantity of the i-th output item. The quantity of the i-th output product is determined based on the numerical representation information of the quantity of the i-th output product.

5. The training method according to claim 1, characterized in that, The step of numerically encoding the sample user profile information and the 0th to (i-1)th output product information to obtain the first numerical feature includes: The sample user profile information and the output product information from 0 to i-1 are input into the first encoding model for numerical encoding to obtain the first numerical feature.

6. The training method according to claim 5, characterized in that, The training method also includes: The model parameters in the first encoding model are adjusted according to the principle of improving the intended parameters until the intended parameters meet the preset conditions, thereby obtaining the first encoding model after optimized training.

7. The training method according to claim 1, characterized in that, The parameters for determining the sample user's intention to purchase the sample gift package include: The sample user profile information and the sample gift package are input into a trained probability calculation model to obtain the probability that the sample user will purchase the sample gift package.

8. The training method according to any one of claims 1 to 7, characterized in that, The sample user profile information includes at least one of the following: sample user's historical purchase information, sample user's attribute description information; The sample user's historical purchase information includes at least one of the following: sample user's historical purchase information of gift packs and sample user's historical purchase information of goods, wherein the gift pack includes at least one good product.

9. The training method according to claim 7, characterized in that, When the sample user's historical purchase information includes the sample user's historical purchase of gift packs and the sample user's historical purchase of goods, the sample user's historical purchase information specifically includes: each item included in the sample user's historical purchase of gift packs and the sample product sequence after the sample user's historical purchase of goods is sorted by purchase time.

10. The training method according to claim 9, characterized in that, The sample user's historical purchase information also includes: a sequence of sample gift packs ordered by purchase time based on the sample user's historical purchases.

11. The training method according to claim 8, characterized in that, The attribute description information of the sample users includes at least one of the following: sample user level, sample user's average game spending, sample user's equipment value, sample user's online time information, sample user's gender information, sample user's age information, and sample user's geographical location.

12. The training method according to any one of claims 1 to 7, characterized in that, The training method also includes: When the preset time is reached, the latest sample user profile information corresponding to each sample user is obtained. The generation time of the latest sample user profile information is less than the current time interval. The trained gift pack generation model is updated and trained using the latest sample user profile information to obtain the updated gift pack generation model.

13. A method for pushing gift packages, characterized in that, The push method includes: Obtain the user profile information of the user to be pushed to; The third input information of the trained gift pack generation model is determined based on the user profile information to be pushed and the information of the 0th to j-1th product to be pushed. The third input information is then input into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed. Here, j is an integer, j traverses from 1 to T, and T is the number of preset product types included in the push gift pack. The 0th product information to be pushed is empty. The trained gift pack generation model is trained by the gift pack generation model training method according to any one of claims 1 to 12. A push notification package is generated for the user to be pushed the product information based on the first to T items. The push notification package will be sent to the user to be pushed to; The third input information for determining the trained gift pack generation model using the user profile information to be pushed and the information of the 0th to j-1th products to be pushed includes: The user profile information to be pushed and the information of the 0th to j-1th product to be pushed are numerically encoded to obtain the third numerical feature, and the third numerical feature is determined as the third input information of the trained gift pack generation model. The step of inputting the third input information into the trained gift package generation model to obtain the j-th item information to be pushed in the push gift package generated for the user to be pushed includes: The third input information is input into the trained gift pack generation model to obtain the numerical representation information of the j-th item to be pushed in the push gift pack generated for the user to be pushed; the trained gift pack generation model is trained by the gift pack generation model training method of any one of claims 1 to 6; Based on the numerical representation information of the j-th product to be pushed, it is determined to be the j-th product information in the push gift package generated by the user to be pushed.

14. The push method according to claim 13, characterized in that, The step of determining the information of the j-th product to be pushed in the push gift package generated for the user to be pushed, based on the numerical representation information of the j-th product to be pushed, includes: Select the target product from the pre-stored candidate products as the j-th product information to be pushed in the push gift package generated for the user to be pushed, wherein the similarity between the numerical representation information corresponding to the target product and the numerical representation information of the j-th product to be pushed is greater than a second preset similarity, and the candidate products are set with corresponding numerical representation information.

15. The push method according to claim 14, characterized in that, Before generating a push gift package for the user to be pushed based on the first to T pieces of information about the products to be pushed, the push method further includes: Based on the user profile information to be pushed, the information of the first to j products to be pushed, and the quantity of the products of the 0 to j-1 products to be pushed, the fourth input information of the trained gift pack generation model is determined, and the fourth input information is input into the trained gift pack generation model to obtain the quantity of the j-th product to be pushed. When j is greater than 1, the step of determining the third input information of the trained gift pack generation model based on the user profile information to be pushed and the information of the 0th to j-1th products to be pushed includes: Based on the user profile information to be pushed, the information of the 0th to j-1th products to be pushed, and the quantity of the 0th to j-1th products to be pushed, the third input information of the trained gift pack generation model is determined; The step of generating a push notification package for the user based on the first to T pieces of information about the products to be pushed includes: Based on the first to T items of information to be pushed and the quantity of the first to T items to be pushed, a push gift pack is generated for the user to be pushed.

16. The push method according to claim 15, characterized in that, The step of determining the fourth input information of the trained gift pack generation model based on the user profile information to be pushed, the information of the 1st to jth items to be pushed, and the quantity of the 0th to j-1th items to be pushed includes: The user profile information to be pushed, the information of the first to j items to be pushed, and the quantity of the first to j-1 items to be pushed are numerically encoded to obtain a fourth numerical feature, and the fourth numerical feature is determined as the fourth input information of the trained gift pack generation model. The step of inputting the fourth input information into the trained gift package generation model to obtain the quantity of the j-th item to be pushed includes: The fourth input information is input into the trained gift pack generation model to obtain numerical representation information representing the quantity of the j-th product to be pushed. The quantity of the j-th product to be pushed is determined based on the numerical representation information of the quantity of the j-th product to be pushed.

17. The push method according to claim 14, characterized in that, The step of selecting the target product from the pre-stored candidate products as the j-th product to be pushed in the push notification package generated for the user to be pushed includes: Select at least one target product from the pre-stored candidate products; Based on a probability sampling algorithm, the j-th item information to be pushed is selected from the target items in the push gift package generated for the user to be pushed.

18. The push method according to claim 17, characterized in that, The step of generating a push notification package for the user based on the first to T pieces of information about the products to be pushed includes: Perform step S times in the trained gift pack generation model to input the third input information into the third input information, and obtain the S groups of the 1st to Tth items to be pushed; S candidate gift packages to be pushed to the user are generated based on the first to T output product information of group S. Determine the intention parameters of the users to be pushed to purchase each of the candidate push packages, and obtain S intention parameters; A push notification package is generated for the user to be pushed to based on the S intention parameters.

19. The push method according to claim 18, characterized in that, The push notification package generated for the user to be pushed to based on the S intention parameters includes: Select the gift package with the largest intention parameter from the S candidate gift packages to be pushed to as the push gift package generated for the user to be pushed to; Alternatively, based on the package prices corresponding to the S candidate packages to be pushed and the S intention parameters, determine the expected benefits corresponding to the S candidate packages to be pushed, and select the package with the highest expected benefit from the S candidate packages to be pushed as the push package generated for the user to be pushed.

20. The push method according to claim 14, characterized in that, The step of numerically encoding the user profile information to be pushed and the information of the 0th to j-1th products to be pushed to obtain the third numerical feature includes: The user profile information to be pushed and the information of the 0th to j-1th product to be pushed are input into the second encoding model for numerical encoding to obtain the third numerical feature. The trained gift pack generation model is trained by the gift pack generation model training method of claim 7. The second encoding model is the first encoding model after optimization training in claim 6.

21. A training device for a gift package generation model, characterized in that, The training device includes: The first acquisition unit is used to acquire a training set, which includes sample user profile information corresponding to each sample user. The first calculation unit is used to determine the first input information of the gift pack generation model to be trained based on the sample user profile information and the 0th to i-1th output product information, and input the first input information into the gift pack generation model to be trained to obtain the i-th output product information in the sample gift pack generated for the sample user, where i is an integer, i traverses from 1 to N, N is the number of preset product types contained in the sample gift pack, and the 0th output product information is empty; The first generation unit is used to generate sample gift packs for the sample users based on the first to Nth output product information. A determining unit is used to determine the intention parameters of sample users to purchase the sample gift package; the intention parameters are represented by the probability of the sample users purchasing the sample gift package obtained by the sample user profile information and the sample gift package input probability calculation model; The adjustment unit is used to adjust the model parameters of the gift pack generation model to be trained, based on the principle of improving the intention parameters, until the intention parameters meet the preset conditions, and a trained gift pack generation model is obtained. The step of determining the first input information for the gift pack generation model to be trained based on the sample user profile information and the information of the 0th to i-1th output products includes: The sample user profile information and the 0th to i-1th output product information are numerically encoded to obtain the first numerical feature, and the first numerical feature is determined as the first input information of the gift pack generation model to be trained. The step of inputting the first input information into the training model to obtain the i-th output product information in the sample gift package generated for the sample user includes: The first input information is input into the training gift pack generation model to obtain the numerical representation information of the i-th output product in the sample gift pack generated for the sample user; The information of the i-th output product is determined based on the numerical representation information of the i-th output product, which is the information of the i-th output product in the sample gift package generated by the sample user.

22. A gift package delivery device, characterized in that, The pushing device includes: The second acquisition unit is used to acquire the user profile information of the user to be pushed to; The second calculation unit is used to determine the third input information of the trained gift pack generation model based on the user profile information to be pushed and the 0th to j-1th product information to be pushed, and input the third input information into the trained gift pack generation model to obtain the jth product information to be pushed in the push gift pack generated for the user to be pushed, where j is an integer, j traverses 1 to T, T is the number of preset product types contained in the push gift pack, and the 0th product information to be pushed is empty. The trained gift pack generation model is trained by the gift pack generation model training method of any one of claims 1 to 12. The second generation unit is used to generate a push gift pack for the user to be pushed based on the first to T pieces of push product information; The push unit is used to push the push gift package to the user to be pushed; The third input information for determining the trained gift pack generation model using the user profile information to be pushed and the information of the 0th to j-1th products to be pushed includes: The user profile information to be pushed and the information of the 0th to j-1th product to be pushed are numerically encoded to obtain the third numerical feature, and the third numerical feature is determined as the third input information of the trained gift pack generation model. The step of inputting the third input information into the trained gift package generation model to obtain the j-th item information to be pushed in the push gift package generated for the user to be pushed includes: The third input information is input into the trained gift pack generation model to obtain the numerical representation information of the j-th item to be pushed in the push gift pack generated for the user to be pushed; the trained gift pack generation model is trained by the gift pack generation model training method of any one of claims 1 to 6; Based on the numerical representation information of the j-th product to be pushed, it is determined to be the j-th product information in the push gift package generated by the user to be pushed.

23. An electronic device, characterized in that, include: processor; as well as A memory for storing a data processing program, which, when the electronic device is powered on and runs through the processor, executes the method as described in any one of claims 1-20.

24. A computer-readable storage medium, characterized in that, The system contains a data processing program that is executed by a processor to perform the method as described in any one of claims 1-20.

Images