Product push method, device, medium, and equipment based on user population modeling

Abstract

The present application discloses a product push method, apparatus, medium, and equipment based on user population modeling, which relates to the field of data processing. The main purpose is to improve the existing problem of increased computing costs and time costs due to inaccurate estimation affecting the effectiveness of product recommendations and product sales, as well as estimating the conversion rate of each product in the user population. The method includes: obtaining information about the target user population, information about multiple target products, and historical push quantity data and historical conversion rate data of each target product in the target user population within a first preset time period; performing conversion rate prediction processing on the target products based on a product conversion rate prediction model to obtain predicted conversion rate data of multiple target products in the target user population within a second preset time period; and sorting the predicted conversion rate data of multiple target products in the target user population within the second preset time period according to preset rules to serve as the basis for product push.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, apparatus, medium, and device for pushing products based on user population modeling. Background Art

[0002] To boost sales, merchants often employ personalized product recommendations to stimulate user purchasing desire, recommending different products to different users. However, due to the high labor and information transmission costs associated with personalized product recommendations, a significant number of merchants opt for the method of identifying popular products to recommend products to users. This involves estimating the conversion rate of users' purchase propensity for each item in a product list over a period of time, and then promoting products with high conversion rates to users to encourage purchases.

[0003] At present, the mining of hot-selling products can be roughly divided into two categories. First, for each product, its conversion rate in the past period is counted as an estimate of the conversion rate of the product in the future period. However, for periodic consumable products with a high conversion rate in the past period does not mean that the product will have a high conversion rate in the future. Since the periodicity of the product is not taken into account, it is very easy to cause the estimated conversion rate of the product to be seriously inconsistent with the actual value, thereby affecting the effect of product recommendation and product sales; second, the purchase probability of each user for each product is estimated, and then for each product, the purchase probability of all users is averaged as the conversion rate estimate of each product. However, due to the complexity of the intermediate process and the huge amount of calculation, it requires a lot of computing power and time, resulting in increased computing and time costs. Summary of the Invention

[0004] In view of this, the present application provides a product push method, device, medium, and equipment based on user population modeling, the main purpose of which is to improve the existing method of using the past conversion rate of a product as an estimate of the future conversion rate, which affects the effect of product recommendation and product sales due to inaccurate estimation, and uses the average purchase probability of all users as the conversion rate estimate of a single product, resulting in increased computing cost and time cost.

[0005] According to one aspect of the present application, a product push method based on user population modeling is provided, comprising:

[0006] Obtain information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each of the target products within the target user group within a first preset time period;

[0007] Performing conversion rate prediction processing on each of the target products based on the trained product conversion rate prediction model to obtain predicted conversion rate data of the target products within the target user population within a second preset time period, wherein the product conversion rate prediction model is obtained based on overall modeling of the user population;

[0008] The predicted conversion rate data of the plurality of target commodities in the target user population within the second preset time period are sorted according to preset rules to serve as a basis for commodity push.

[0009] Preferably, before performing conversion rate prediction processing on the target products based on the product conversion rate prediction model that has completed model training, the method further includes:

[0010] Build an initial product conversion rate prediction model based on attention and recurrent neural networks;

[0011] Obtaining historical data within a third preset time period to generate a training set;

[0012] Training the initial product conversion rate prediction model based on the training set to obtain a product conversion rate prediction model;

[0013] If the loss function of the commodity conversion rate prediction model is less than a preset threshold, the model training is completed, and the commodity conversion rate prediction model is determined to be a commodity conversion rate prediction model that has completed model training.

[0014] Preferably, the acquiring of historical data within a third preset time period to generate a training set specifically includes:

[0015] Selecting multiple anchor points within the third preset time period;

[0016] Obtaining information about user groups at each anchor point, information about all products, and historical push quantity data and historical conversion rate data for all products within a fourth preset time period before the anchor point to generate multiple training sets;

[0017] The actual conversion rate data of the entire quantity of commodities among the user population within the fifth preset time period after each of the anchor points is obtained to determine the loss function of the commodity conversion rate prediction model.

[0018] Preferably, before performing conversion rate prediction processing on the target products based on the product conversion rate prediction model that has completed model training, the method further includes:

[0019] Converting the information of the target user group and the information of the plurality of target products, which are discrete features, into vectors based on a fully connected neural network;

[0020] Normalization is performed on the information of the target user group, the information of the plurality of target products, and the information belonging to continuous numerical features in the historical push quantity data and the historical conversion rate data of each target product in the target user group within the first preset time period.

[0021] Preferably, the preset rule is descending order, and the method further includes:

[0022] A preset number of target commodities are selected from the sorted commodity list as commodities to be pushed, and the commodities are pushed to the target user group.

[0023] Preferably, the method further comprises:

[0024] The information of the products to be pushed is obtained, and the products to be pushed are re-determined based on the preset auxiliary strategy.

[0025] Preferably, the preset auxiliary strategies include freshness auxiliary strategies and category diversity auxiliary strategies.

[0026] According to another aspect of the present application, a product push device based on user population modeling is provided, comprising:

[0027] An acquisition module, configured to acquire information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each of the target products within a first preset time period within the target user group;

[0028] a prediction module configured to perform conversion rate prediction processing on each of the target products based on a product conversion rate prediction model for which model training has been completed, to obtain predicted conversion rate data of a plurality of target products within the target user population within a second preset time period, wherein the product conversion rate prediction model is obtained based on overall modeling of the user population;

[0029] The push module is used to sort the predicted conversion rate data of the plurality of target commodities in the target user population within the second preset time period according to preset rules to serve as a basis for commodity push.

[0030] Preferably, before the prediction module, the device further includes:

[0031] A building block for constructing an initial product conversion rate prediction model based on attention and recurrent neural networks;

[0032] A generation module, configured to obtain historical data within a third preset time period to generate a training set;

[0033] A training module, configured to train the initial commodity conversion rate prediction model based on the training set to obtain a commodity conversion rate prediction model;

[0034] The determination module completes model training if the loss function of the commodity conversion rate prediction model is less than a preset threshold, and determines that the commodity conversion rate prediction model is a commodity conversion rate prediction model that has completed model training.

[0035] Preferably, the generating module specifically includes:

[0036] a selection unit, configured to select a plurality of anchor moments from the third preset time period;

[0037] A generating unit is configured to obtain information about user groups at each anchor point, information about all products, and historical push quantity data and historical conversion rate data of all products among the user groups within a fourth preset time period before the anchor point, to generate multiple training sets;

[0038] An acquisition unit is configured to acquire actual conversion rate data of the entire quantity of commodities among the user population within a fifth preset time period after each of the anchor points, so as to determine a loss function of the commodity conversion rate prediction model.

[0039] Preferably, before the prediction module, the device further includes:

[0040] A conversion module, configured to convert information belonging to discrete features in the information of the target user group and the information of the plurality of target products into vectors based on a fully connected neural network;

[0041] The normalization module is used to normalize the information of the target user group, the information of the multiple target products, and the information belonging to continuous numerical features in the historical push quantity data and historical conversion rate data of each target product in the target user group within the first preset time period.

[0042] Preferably, the preset rule is descending order, and the device further includes:

[0043] The selection module is used to select a preset number of target commodities from the sorted commodity list as commodities to be pushed, and push them to the target user group.

[0044] Preferably, the device further comprises:

[0045] The updating module is used to obtain the information of the products to be pushed and re-determine the products to be pushed in combination with the preset auxiliary strategy.

[0046] Preferably, the preset auxiliary strategies include freshness auxiliary strategies and category diversity auxiliary strategies.

[0047] According to another aspect of the present application, a storage medium is provided, in which at least one executable instruction is stored, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned product push method based on user population modeling.

[0048] According to another aspect of the present application, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0049] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned product push method based on user population modeling.

[0050] By means of the above technical solution, the technical solution provided by the embodiment of the present application has at least the following advantages:

[0051] The present application provides a method and apparatus, medium, and equipment for pushing products based on user population modeling. First, information of the target user population, information of multiple target products, and historical push quantity data and historical conversion rate data of each of the target products in the target user population within a first preset time period are obtained; secondly, conversion rate prediction processing is performed on the target products based on the product conversion rate prediction model that has completed model training, and predicted conversion rate data of multiple target products in the target user population within a second preset time period are obtained. The product conversion rate prediction model is obtained based on the overall modeling of the user population; finally, the predicted conversion rate data of multiple target products in the target user population within the second preset time period are sorted according to preset rules to serve as the basis for product pushing. Compared with the prior art, the embodiment of the present application predicts the conversion rate of the product in the user population through the product conversion rate prediction model obtained based on the overall modeling of the user population, and pushes the product according to the conversion rate. It takes into account the periodicity of the conversion rate over time, and does not need to predict the purchase probability of each user, avoiding a large amount of calculation, reducing the demand for computing power, and thus improving the efficiency of model training and prediction.

[0052] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0054] Figure 1 A flowchart of a product push method based on user population modeling provided by an embodiment of the present application is shown;

[0055] Figure 2 A schematic diagram of the commodity conversion rate prediction model structure provided in an embodiment of the present application is shown;

[0056] Figure 3 A block diagram of a product push device based on user population modeling provided by an embodiment of the present application is shown;

[0057] Figure 4 A structural diagram of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0058] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0059] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0060] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0061] Based on this, in one embodiment, Figure 1As shown, a product push method based on user population modeling is provided, and the method is applied to a computer device such as a server as an example for explanation. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, such as smart medical systems and digital medical platforms. The above method includes the following steps:

[0062] 101. Obtain information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each target product in the target user group within a first preset time period.

[0063] In the embodiment of the present application, the information of the target user group includes but is not limited to the types of goods most purchased by the target user group within a preset time period, the average number of purchases, the consumption amount, etc.; the information of the target goods includes but is not limited to the product category, product brand, product price, brand level, etc.; the first preset time period is used to represent a historical period before the current time point, such as the previous 12 weeks or the previous 24 weeks.

[0064] For example, the most purchased product categories of the target user group in the last week / 2 weeks / 1 month / 2 months / 3 months are obtained, which are 5 discrete features in total, and the average purchase frequency and consumption amount of the target user group in the last week / 2 weeks / 1 month / 2 months / 3 months are obtained, which are 10 continuous numerical features in total, as the information of the target user group; the product category and product brand are obtained, which are 2 discrete features in total, and the product price and brand level are obtained, which are 2 continuous numerical features in total, as the information of the target product. The discrete features in the target user group information and the target product information obtained above can be recorded as a i ,i=1,2...7, continuous numerical feature x i , i=1,2...12; obtain the historical push quantity data of each target product in the target user group within the first preset time period, recorded as sequence s l and historical conversion rate data, recorded as sequence c l , where the time window is set to 1 week, and each time window is adjacent and non-overlapping. l = 1, 2...L represents the lth time window in history, and L represents the first preset time period, which is set to 12 weeks. That is, there are 24 continuous numerical features in total, which serve as the push quantity and conversion rate of the target product among the target user population within the historical 12 weeks.

[0065] 102. Perform conversion rate prediction processing on target products respectively based on the product conversion rate prediction model for which model training has been completed, and obtain predicted conversion rate data of multiple target products in the target user group within a second preset time period.

[0066] Among them, the product conversion rate prediction model is obtained based on the overall modeling of the user population; the second preset time period is used to represent a future time period after the current time point, such as the next week.

[0067] In an embodiment of the present application, the conversion rate of the target product is predicted by a product conversion rate prediction model obtained by modeling the user population as a whole, so that the conversion rate of the target product in the target user population can be obtained. Compared with the existing scheme that uses the past conversion rate of the product as the estimated value of the future conversion rate, the cyclical change factor of the product sales volume is taken into consideration, which effectively improves the accuracy of predicting the conversion rate of the target product, ensures the effect of product recommendation, and thus helps to increase the sales volume of the product; in addition, compared with the existing scheme that uses the average purchase probability of all users as the estimated value of the conversion rate of a single product, the embodiment of the present application does not need to predict the purchase probability of each user, avoids a large amount of calculation, reduces the demand for computing power, and reduces computing cost and time cost.

[0068] 103. The predicted conversion rate data of the plurality of target products in the target user group within the second preset time period are sorted according to preset rules to serve as a basis for product push.

[0069] In this embodiment of the present application, the predicted conversion rate data for each target product within the second preset time period, obtained in step 102 above, is sorted according to a preset rule, wherein the preset rule can be a descending order or an ascending order. For example, in descending order, the first 20 target products in the descending order sequence are selected and pushed to the target user group; in ascending order, the first 30 target products in the ascending order sequence are filtered out from the current push, thereby improving the efficiency of the push.

[0070] Compared with the existing technology, the embodiment of the present application predicts the conversion rate of a product in a user population through a product conversion rate prediction model obtained based on overall modeling of the user population, and pushes products based on the conversion rate. It takes into account the periodicity of the conversion rate over time, and at the same time does not need to predict the purchase probability of each user, avoiding a large amount of calculation, reducing the demand for computing power, and thereby improving the efficiency of model training and prediction.

[0071] For further explanation and limitation, in an embodiment of the present application, before performing conversion rate prediction processing on the target products based on the product conversion rate prediction model that has completed model training, the embodiment method also includes: constructing an initial product conversion rate prediction model based on attention and regression neural network; obtaining historical data within a third preset time period to generate a training set; training the initial product conversion rate prediction model based on the training set to obtain a product conversion rate prediction model; if the loss function of the product conversion rate prediction model is less than a preset threshold, the model training is completed, and the product conversion rate prediction model is determined to be the product conversion rate prediction model that has completed model training.

[0072] Specifically, we first build an initial product conversion rate prediction model based on attention and recurrent neural networks. The structure is as follows: Figure 2 As shown, it includes: 1) Embedding layer, the input of this layer is discrete features, and the output is the representation vector of the discrete features. For example, if the input is the product category, all products can be divided into K = 100 categories, that is, there are 100 cases for this discrete feature, and the number of elements in the representation vector is set to D = 8. At this time, the Embedding layer will maintain a parameter matrix with D rows and K columns. When the input product category is the kth category, the layer will extract the kth column vector of the matrix as the output vector, that is, the representation vector of the kth category product. Each discrete feature in the input of the product conversion rate prediction model corresponds to an Embedding layer, and the parameter set of all Embedding layers in the model is denoted as {W E,i}, i=1,2...7, corresponding to a in the above example i , i = 1, 2...7, a total of 7 discrete features. 2) b) Stacking layer: This layer takes as input the vectors output by the Embedding layer, and outputs the concatenation of all output vectors, i.e., they are connected end to end in a fixed order, with the number of elements being the sum of the number of elements in the input vectors. 3) c) MLP layer: This is a multilayer perceptron. Taking a single hidden layer neural network as an example, its structure can be expressed as:

[0073] Input: denoted as x∈R D×1 , that is, a column vector of length D;

[0074] Hidden layer: h = σ h (W h x+b h ), where W h ∈R D ' ×D , b h ∈R D ' ×1 is the model parameter, σ h is the relu activation function, D' is the number of hidden neurons, which can be set to 256;

[0075] Output: o=σ o (W o x+b o ), where W o ∈R L×D ', b o ∈R L×1 is the model parameter, σ o is the tanh activation function.

[0076] This layer integrates various features, including information about the target product and the target user population, and outputs an attention vector with the same number of elements as the historical conversion rate sequence / historical push quantity sequence. The weight coefficient of the historical conversion rate sequence will be corrected in the form of a residual. 4) Normalization layer: This layer is used to normalize the historical push quantity sequence so that the sum of its elements is 1. This layer outputs the original weight vector, denoted as r∈R L×1 ,L=12, and then after being corrected by the attention vector o, it is used to analyze the historical conversion rate sequence c l Perform weighted averaging. 5) Element-wise multiplication layer: Its input is the original weight vector r∈R L×1 , and the attention vector o∈R L×1 The output of this layer is the element-wise product of two vectors, denoted as r res , where each element r res,l =r l o l ,l=1,2...L,L=12。The output of this layer can be used as the residual correction vector of the original weight vector. 6) Summation layer: The input of this layer is the original weight vector r∈R L×1 and its residual correction vector r res ∈R L×1 , output the corrected weight vector r'∈R L×1 , each element is the sum of the corresponding elements of the input vector, that is, r l '=r l +r res,l 7) Inner product layer: This layer uses the modified weight vector r'∈R L×1 As the weight coefficient, the historical conversion rate sequence is weighted averaged in the form of vector inner product, and the estimated conversion rate y of the corresponding product in the future time window is output, that is, This layer uses historical conversion rate data and combines it with the weight parameters generated by the attention mechanism to achieve a regression estimate of future conversion rates.

[0077] Secondly, historical data within a third preset time period is obtained to generate a training set, wherein the third preset time period may be the past year, and the training set is generated using the historical data of the past year.

[0078] Finally, the initial product conversion rate prediction model is trained until the model converges, indicating that the training is complete. Specifically, the mean square error function is used as the optimization target. Among them, y is the model output result, y gt is the true value of the conversion rate. For example, using the stochastic gradient descent algorithm as the optimization algorithm Among them, α is the preset learning rate, w represents all model parameters, including all Embedding layer parameter sets {W E,i}, i = 1, 2...7, and the MLP layer parameter set {W h ,h h ,W o ,h o When the training continues until the loss function L remains below the preset threshold for several steps, the model training is completed, and a product conversion rate prediction model that has completed model training is obtained and can be used for product conversion rate prediction.

[0079] For further explanation and limitation, in an embodiment of the present application, historical data within a third preset time period is obtained to generate a training set, specifically including: selecting multiple anchor points from the third preset time period; obtaining information on the user population at each anchor point, information on the full amount of goods, and historical push quantity data and historical conversion rate data of the full amount of goods in the user population within the fourth preset time period before the anchor point moment to generate multiple sets of training sets; obtaining the actual conversion rate data of the full amount of goods in the user population within the fifth preset time period after each of the anchor points to determine the loss function of the product conversion rate prediction model.

[0080] Exemplarily, the historical data of the past year (i.e., the third preset time period) is used as the training set, which is about 52 weeks, and is referred to as weeks 1 to 52 in chronological order. Continuing with the above example, since the time span of the historical information input to the model is L = 12 time windows, i.e., 12 weeks (the fourth preset time period), the first day of weeks 13 to 52 is selected as the anchor point moment, and 40 sets of historical data are generated as training sets. Each set of historical data contains information on all products, information on the user population at the anchor point moment, and historical information on push and conversion rates of all products in the L = 12 weeks before the anchor point moment, which are used as model inputs together; it also contains conversion rate data for all products within 1 week after the anchor point moment (the fifth preset time period), which is used as the model output of the true conversion rate, which is used to determine the loss function of the product conversion rate prediction model. For example, for the first set of historical data, the anchor point is the first day of the 13th week. Information on all products, information on the user population at the anchor point, and information on product push and conversion rates from weeks 1 to 12 are obtained as model input, and the product conversion rate of the 13th week is obtained as the model output, which is the actual conversion rate. For the second set of historical data, the anchor point is the first day of the 14th week. Information on all products, information on the user population at the anchor point, and information on product push and conversion rates from weeks 2 to 13 are obtained as model input, and the product conversion rate of the 14th week is obtained as the model output, which is the actual conversion rate. And so on... For the 40th set of historical data, the anchor point is the first day of the 52nd week. Information on all products, information on the user population at the anchor point, and information on product push and conversion rates from weeks 40 to 51 are obtained as model input, and the product conversion rate of week 52 is obtained as the model output, which is the actual conversion rate.

[0081] Preferably, in an embodiment of the present application, before performing conversion rate prediction processing on the target products respectively based on the product conversion rate prediction model that has completed model training, the embodiment method also includes: converting the information of the target user population and the information of multiple target products that belong to discrete features into vectors based on a fully connected neural network; normalizing the information of the target user population, the information of multiple target products, and the historical push quantity data and historical conversion rate data of each target product in the target user population within a first preset time period that belong to continuous numerical features.

[0082] Specifically, for all discrete features in the model input, the possible options of the feature are numbered 1, 2, ... in a fixed order, so that they can be input into the model's Embedding layer for calculation. For example, for the product category feature, all products can be divided into M = 100 categories, that is, there are M = 100 cases for this discrete feature. After sorting in a fixed order (such as pinyin), they are recorded as category 1, category 2, category 3, ..., category M (= 100), and numbered 1, 2, 3, ..., M (= 100) respectively. Subsequently, they will be input into the model's Embedding layer for calculation. For each continuous numerical feature x in the input i , normalize it, that is, according to the distribution of training set data, calculate the mean μ of the feature i and variance σ i , and transform the features into Make its mean 0 and variance 1.

[0083] Optionally, in an embodiment of the present application, the preset rule is descending order, and the embodiment method further includes: selecting a preset number of target products from the sorted product list as products to be pushed, and pushing them to the target user group. For example, using descending order, the first 20 target products in the descending sequence are selected and pushed to the target user group.

[0084] Optionally, to improve the efficiency of product push, in an embodiment of the present application, the method further includes: obtaining information about the products to be pushed, and re-determining the products to be pushed in combination with a preset auxiliary strategy. The preset auxiliary strategy includes a freshness auxiliary strategy package and a category diversity auxiliary strategy.

[0085] For example, by obtaining information about the products to be pushed, it is known that the conversion rate of Category A products is relatively low. However, considering that the product has not been recommended for a long time, there is less historical data and the estimation of the conversion rate of the product may be inaccurate. Therefore, the ranking of the product can be appropriately improved so that it has a chance to be launched. This is a freshness auxiliary strategy; or when mining hot-selling products, it is necessary to find the five products with the highest conversion rate, but in the actual estimation results, the top five products are all in the same category. At this time, the ranking of the 3rd to 5th products can be appropriately lowered so that products ranked > 5 have a chance to be launched. The types of the five products finally selected are as rich as possible to avoid being single, which is a category diversity auxiliary strategy.

[0086] The present application provides a method for pushing products based on user population modeling. First, information about the target user population, information about multiple target products, and historical push quantity data and historical conversion rate data of each target product in the target user population within a first preset time period are obtained; secondly, conversion rate prediction processing is performed on the target products based on the product conversion rate prediction model that has completed model training, and predicted conversion rate data of multiple target products in the target user population within a second preset time period are obtained. The product conversion rate prediction model is obtained based on the overall modeling of the user population; finally, the predicted conversion rate data of multiple target products in the target user population within the second preset time period are sorted according to preset rules to serve as the basis for product pushing. Compared with the prior art, the embodiment of the present application predicts the conversion rate of products in the user population through the product conversion rate prediction model obtained based on the overall modeling of the user population, and pushes products based on the conversion rate. It takes into account the periodicity of the conversion rate over time, and does not need to predict the purchase probability of each user, avoiding a large amount of calculation, reducing the demand for computing power, and thus improving the efficiency of model training and prediction.

[0087] Furthermore, as a response to the above Figure 1 The embodiment of the present application provides a product push device based on user group modeling, such as Figure 3 As shown, the device includes:

[0088] Acquisition module 21, prediction module 22, push module 23.

[0089] An acquisition module 21 is configured to acquire information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each of the target products within a first preset time period.

[0090] Prediction module 22, configured to perform conversion rate prediction processing on each target product based on the trained product conversion rate prediction model, and obtain predicted conversion rate data of multiple target products within the target user population within a second preset time period, wherein the product conversion rate prediction model is obtained based on overall modeling of the user population;

[0091] The push module 23 is configured to sort the predicted conversion rate data of the plurality of target products in the target user group within the second preset time period according to preset rules, so as to serve as a basis for product push.

[0092] In a specific application scenario, before the prediction module, the device further includes:

[0093] A building block for constructing an initial product conversion rate prediction model based on attention and recurrent neural networks;

[0094] A generation module, configured to obtain historical data within a third preset time period to generate a training set;

[0095] A training module, configured to train the initial commodity conversion rate prediction model based on the training set to obtain a commodity conversion rate prediction model;

[0096] The determination module completes model training if the loss function of the commodity conversion rate prediction model is less than a preset threshold, and determines that the commodity conversion rate prediction model is a commodity conversion rate prediction model that has completed model training.

[0097] In a specific application scenario, the generation module specifically includes:

[0098] a selection unit, configured to select a plurality of anchor moments from the third preset time period;

[0099] A generating unit is configured to obtain information about user groups at each anchor point, information about all products, and historical push quantity data and historical conversion rate data of all products among the user groups within a fourth preset time period before the anchor point, to generate multiple training sets;

[0100] An acquisition unit is configured to acquire actual conversion rate data of the entire quantity of commodities among the user population within a fifth preset time period after each of the anchor points, so as to determine a loss function of the commodity conversion rate prediction model.

[0101] In a specific application scenario, before the prediction module, the device further includes:

[0102] A conversion module, configured to convert information belonging to discrete features in the information of the target user group and the information of the plurality of target products into vectors based on a fully connected neural network;

[0103] The normalization module is used to normalize the information of the target user group, the information of the multiple target products, and the information belonging to continuous numerical features in the historical push quantity data and historical conversion rate data of each target product in the target user group within the first preset time period.

[0104] In a specific application scenario, the preset rule is descending order, and the device further includes:

[0105] The selection module is used to select a preset number of target commodities from the sorted commodity list as commodities to be pushed, and push them to the target user group.

[0106] In a specific application scenario, the device further includes:

[0107] The updating module is used to obtain the information of the products to be pushed and re-determine the products to be pushed in combination with the preset auxiliary strategy.

[0108] In a specific application scenario, the preset auxiliary strategies include a freshness auxiliary strategy and a category diversity auxiliary strategy.

[0109] The present application provides a product push device based on user population modeling, which first obtains information about the target user population, information about multiple target products, and historical push quantity data and historical conversion rate data of each of the target products in the target user population within a first preset time period; secondly, based on the product conversion rate prediction model that has completed model training, the target products are respectively subjected to conversion rate prediction processing, and the predicted conversion rate data of the multiple target products in the target user population within a second preset time period are obtained, and the product conversion rate prediction model is obtained based on the overall modeling of the user population; finally, the predicted conversion rate data of the multiple target products in the target user population within the second preset time period are sorted and processed according to preset rules to serve as the basis for product push. Compared with the prior art, the embodiment of the present application predicts the conversion rate of the product in the user population through the product conversion rate prediction model obtained based on the overall modeling of the user population, and pushes the product according to the conversion rate, which takes into account the periodicity of the conversion rate changing over time, and does not need to predict the purchase probability of each user, thus avoiding a large amount of calculation, reducing the demand for computing power, and thus improving the efficiency of model training and prediction.

[0110] According to one embodiment of the present application, a storage medium is provided, which stores at least one executable instruction. The computer-executable instruction can execute the product push method based on user population modeling in any of the above method embodiments.

[0111] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0112] Figure 4 A schematic structural diagram of a computer device provided according to an embodiment of the present application is shown. The specific embodiment of the present application does not limit the specific implementation of the computer device.

[0113] like Figure 4As shown, the computer device may include: a processor 302 , a communications interface 304 , a memory 306 , and a communication bus 308 .

[0114] The processor 302 , the communication interface 304 , and the memory 306 communicate with each other via a communication bus 308 .

[0115] The communication interface 304 is used to communicate with other devices such as clients or other servers.

[0116] The processor 302 is configured to execute the program 310 , and specifically to execute the relevant steps in the embodiment of the above-mentioned method for pushing products based on user population modeling.

[0117] Specifically, the program 310 may include program codes, which include computer operation instructions.

[0118] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in a computer device may be processors of the same type, such as one or more CPUs, or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0119] The memory 306 is used to store the program 310. The memory 306 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0120] The program 310 may be specifically configured to cause the processor 302 to perform the following operations:

[0121] Obtain information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each of the target products within the target user group within a first preset time period;

[0122] Performing conversion rate prediction processing on each of the target products based on the trained product conversion rate prediction model to obtain predicted conversion rate data of the target products within the target user population within a second preset time period, wherein the product conversion rate prediction model is obtained based on overall modeling of the user population;

[0123] The predicted conversion rate data of the plurality of target commodities in the target user population within the second preset time period are sorted according to preset rules to serve as a basis for commodity push.

[0124] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device used to push products based on user demographic modeling, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between components within the storage medium and with other hardware and software in the physical information processing device.

[0125] Each embodiment in this specification is described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. References to the same or similar parts between the various embodiments are sufficient. For system embodiments, since they largely correspond to method embodiments, their description is relatively simple. For relevant parts, references to the description of the method embodiments are sufficient.

[0126] The methods and systems of the present application may be implemented in many ways. For example, the methods and systems of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the methods according to the present application. Therefore, the present application also covers recording media that store programs for executing the methods according to the present application.

[0127] Obviously, those skilled in the art should understand that the modules or steps of the present application described above can be implemented using a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. Alternatively, they can be implemented using program code executable by the computing device, so that they can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than herein, or they can be made into separate integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0128] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Persons skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A product push method based on user population modeling, characterized in that: include: Obtain information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each of the target products within the target user group within a first preset time period; Performing conversion rate prediction processing on each of the target products based on the trained product conversion rate prediction model to obtain predicted conversion rate data of the target products within the target user population within a second preset time period, wherein the product conversion rate prediction model is obtained based on overall modeling of the user population; Sorting the predicted conversion rate data of the plurality of target products among the target user population within the second preset time period according to preset rules, so as to serve as a basis for product push; Before performing conversion rate prediction processing on the target products based on the product conversion rate prediction model that has completed model training, the method further includes: Selecting multiple anchor points within a third preset time period; Obtaining information about user groups at each anchor point, information about all products, and historical push quantity data and historical conversion rate data for all products within a fourth preset time period before the anchor point to generate multiple training sets; Obtaining actual conversion rate data of the entire quantity of products among the user population within a fifth preset time period after each of the anchor points to determine a loss function of the product conversion rate prediction model; Build an initial product conversion rate prediction model based on attention and recurrent neural networks; Training the initial product conversion rate prediction model based on the training set to obtain a product conversion rate prediction model; If the loss function of the commodity conversion rate prediction model is less than a preset threshold, the model training is completed, and the commodity conversion rate prediction model is determined to be a commodity conversion rate prediction model that has completed model training; Performing conversion rate prediction processing on the target products respectively based on the product conversion rate prediction model for which model training has been completed, and obtaining predicted conversion rate data of a plurality of target products in the target user population within a second preset time period, including: Inputting information about the target user group, information about multiple target products, and historical push quantity data and historical conversion rate data of each target product in the target user group within a first preset time period into a product conversion rate prediction model; In the Embedding layer, each discrete feature is converted into a corresponding representation vector; In the Stacking layer, the representation vectors output by the Embedding layer are carefully spliced ​​together, and the output is the splicing result of all representation vectors; In the MLP layer, various types of features are integrated and the attention vector is output; In the normalization layer, the historical push quantity sequence is normalized so that the sum of its elements is 1, and the output is the original weight vector; In the element-wise product layer, the input is the original weight vector and the attention vector, and the output is the element-wise product of the two vectors. The element-wise product serves as the residual correction vector of the original weight vector. In the summation layer, the input is the original weight vector and its residual correction vector, and the output is the corrected weight vector; In the inner product layer, the corrected weight vector is used as the weight coefficient, and the historical conversion rate sequence is weighted averaged in the form of vector inner product to output the predicted conversion rate data of the multiple target products in the target user population within the second preset time period.

2. The method according to claim 1, characterized in that Before performing conversion rate prediction processing on the target products based on the product conversion rate prediction model that has completed model training, the method further includes: Converting the information of the target user group and the information of the plurality of target products, which are discrete features, into vectors based on a fully connected neural network; Normalization is performed on the information of the target user group, the information of the plurality of target products, and the information belonging to continuous numerical features in the historical push quantity data and the historical conversion rate data of each target product in the target user group within the first preset time period.

3. The method according to claim 1, characterized in that The preset rule is descending order, and the method further includes: A preset number of target commodities are selected from the sorted commodity list as commodities to be pushed, and the commodities are pushed to the target user group.

4. The method according to claim 3, characterized in that The method further comprises: The information of the products to be pushed is obtained, and the products to be pushed are re-determined based on the preset auxiliary strategy.

5. The method according to claim 4, characterized in that The preset auxiliary strategies include freshness auxiliary strategy and category diversity auxiliary strategy.

6. A product push device based on user population modeling, characterized in that: include: An acquisition module, configured to acquire information about a target user group, information about a plurality of target products, and historical push quantity data and historical conversion rate data of each of the target products within a first preset time period within the target user group; a prediction module configured to perform conversion rate prediction processing on each of the target products based on a product conversion rate prediction model for which model training has been completed, to obtain predicted conversion rate data of a plurality of target products within the target user population within a second preset time period, wherein the product conversion rate prediction model is obtained based on overall modeling of the user population; a push module, configured to sort the predicted conversion rate data of the plurality of target products in the target user group within a second preset time period according to preset rules, so as to serve as a basis for product push; Before the prediction module, the device further includes a generation module, which specifically includes: a selection unit, configured to select a plurality of anchor moments from a third preset time period; A generating unit is configured to obtain information about user groups at each anchor point, information about all products, and historical push quantity data and historical conversion rate data of all products among the user groups within a fourth preset time period before the anchor point, to generate multiple training sets; an acquisition unit, configured to acquire actual conversion rate data of the entire quantity of products among the user population within a fifth preset time period after each of the anchor points, so as to determine a loss function of the product conversion rate prediction model; A building block for constructing an initial product conversion rate prediction model based on attention and recurrent neural networks; A training module, configured to train the initial commodity conversion rate prediction model based on the training set to obtain a commodity conversion rate prediction model; a determination module configured to complete model training if the loss function of the commodity conversion rate prediction model is less than a preset threshold, and determine that the commodity conversion rate prediction model is a commodity conversion rate prediction model that has completed model training; Prediction module, used to: Inputting information about the target user group, information about multiple target products, and historical push quantity data and historical conversion rate data of each target product in the target user group within a first preset time period into a product conversion rate prediction model; In the Embedding layer, each discrete feature is converted into a corresponding representation vector; In the Stacking layer, the representation vectors output by the Embedding layer are carefully spliced ​​together, and the output is the splicing result of all representation vectors; In the MLP layer, various types of features are integrated and the attention vector is output; In the normalization layer, the historical push quantity sequence is normalized so that the sum of its elements is 1, and the output is the original weight vector; In the element-wise product layer, the input is the original weight vector and the attention vector, and the output is the element-wise product of the two vectors. The element-wise product serves as the residual correction vector of the original weight vector. In the summation layer, the input is the original weight vector and its residual correction vector, and the output is the corrected weight vector; In the inner product layer, the corrected weight vector is used as the weight coefficient, and the historical conversion rate sequence is weighted averaged in the form of vector inner product to output the predicted conversion rate data of the multiple target products in the target user population within the second preset time period.

7. A storage medium storing at least one executable instruction, characterized in that: The executable instructions enable the processor to perform operations corresponding to the product push method based on user population modeling as described in any one of claims 1 to 5.

8. A computer device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, characterized in that the executable instruction enables the processor to perform operations corresponding to the product push method based on user population modeling according to any one of claims 1 to 5.

Images