Training sample generation method and device, model training method and device

Abstract

The embodiments of this specification provide a training sample generation method and device, a model training method and device, and a product price prediction method and device. A specific implementation method of the training sample generation method includes: obtaining the product marketing characteristics corresponding to the product in the historical marketing period, the product marketing characteristics include the product attribute characteristics of the product, and the marketing activity characteristics corresponding to the marketing activities participated in by the product in the historical marketing period; determining the product price tag corresponding to the product in the historical marketing period based on the sales data of the product in the historical marketing period; generating training samples marked with product marketing characteristics and price tags. This implementation achieves targeted information generation. The generated training samples can be used to train a prediction model for predicting the product price, and the prediction results of the prediction model can be used in risk control scenarios, which helps to achieve accurate risk warning.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and specifically, to a method and device for generating training samples, a method and device for training models, a method and device for predicting the price of goods, a method and device for processing goods, a method and device for processing marketing activities, and a method and device for recommending goods. Background Art

[0002] In the front-end marketing scenarios of e-commerce, there are often multiple marketing activities taking effect at the same time for a product, so the final transaction price of the product purchased by the user is affected by multiple marketing activities.

[0003] In practice, marketing activities are usually divided into multiple categories, such as single product direct discount, full discount, store coupons, brand coupons, category coupons, etc. Among them, the discount amounts of different types of marketing activities can be superimposed on each other, but generally only one activity of each type of marketing activity can take effect.

[0004] Therefore, it is necessary to make reasonable estimates and predictions for the transaction prices under the superposition of multiple marketing activities. This will enable the risk control system to accurately reach risk warnings based on the predicted prices with high accuracy, so as to prevent the transaction prices from being too low due to incorrect or multiple marketing activities, which may cause financial losses and personnel risks. Summary of the invention

[0005] The embodiments of this specification provide a training sample generation method and device, a model training method and device, a product price prediction method and device, a product processing method and device, a marketing activity processing method and device, and a product recommendation method and device.

[0006] In a first aspect, an embodiment of the present specification provides a method for generating training samples, the method comprising: obtaining product marketing features corresponding to a product in a historical marketing period, the product marketing features comprising product attribute features of the product, and marketing activity features corresponding to the marketing activities in which the product participated in the historical marketing period; determining a price tag corresponding to the product in the historical marketing period based on sales data of the product in the historical marketing period; generating training samples marked with the product marketing features and the price tag, the training samples being used to train a prediction model for predicting the price of the product.

[0007] In some embodiments, the marketing activities that the product participated in during the historical marketing period include marketing activities set by the merchant to which the product belongs.

[0008] In some embodiments, obtaining product marketing characteristics corresponding to a product in a historical marketing period includes: obtaining marketing activity information corresponding to each marketing activity in which the product participated in the historical marketing period; performing feature extraction on the marketing activity information to obtain marketing activity characteristics; obtaining product attribute characteristics of the product; and combining the product attribute characteristics and the marketing activity characteristics into product marketing characteristics.

[0009] In some embodiments, marketing activities have corresponding activity categories; and feature extraction is performed on marketing activity information to obtain marketing activity features, including: selecting target marketing activities that participate in generating marketing activity features from various marketing activities according to the activity categories to which each marketing activity belongs; and feature extraction is performed on the marketing activity information corresponding to each selected target marketing activity to obtain marketing activity features.

[0010] In some embodiments, based on the activity categories to which each marketing activity belongs, target marketing activities that participate in generating marketing activity features are selected from among the various marketing activities, including: for each activity category in the activity category to which each marketing activity belongs, performing the following marketing activity selection operation: if the marketing activities under the activity category are set with corresponding priorities, then based on the priorities, from among the various marketing activities that belong to the activity category, the marketing activities are selected as the target marketing activities; if the marketing activities under the activity category are not set with corresponding priorities, then based on the discount amounts, from among the various marketing activities that belong to the activity category, the marketing activities are selected as the target marketing activities.

[0011] In some embodiments, feature extraction is performed on the marketing campaign information corresponding to each selected target marketing campaign to obtain marketing campaign features, including: determining a base price corresponding to the product; based on the base price, converting the marketing campaign information corresponding to each target marketing campaign into corresponding value pairs, the value pairs including a threshold amount and a discount amount; and feature processing is performed on each value pair obtained from the conversion to obtain marketing campaign features.

[0012] In some embodiments, performing feature processing on each value pair obtained by the conversion includes: performing a division operation, a subtraction operation, a binary feature addition, and a normalization on each value pair obtained by the conversion.

[0013] In some embodiments, the activity category includes at least one of the following: direct price reduction for single products, discounts for purchases above a certain amount, store coupons, brand coupons, and category coupons.

[0014] In some embodiments, determining a base price corresponding to a commodity includes: if there is a single product direct reduction category in the activity category to which each marketing activity belongs, determining the base price based on the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category.

[0015] In some embodiments, determining the base price corresponding to the product further includes: if there is no single product direct reduction category in the activity category to which each marketing activity belongs, then determining the original price of the product as the base price.

[0016] In some embodiments, the sales data includes multiple prices received from historical orders of the product during a historical marketing period, and the number of sales corresponding to the multiple prices received; and determining the corresponding price tag of the product during the historical marketing period based on the sales data of the product during the historical marketing period, including: taking a weighted average of the multiple prices received and the corresponding number of sales to obtain an average price received; normalizing the average price received, and using the normalized value as the price tag.

[0017] In a second aspect, an embodiment of the present specification provides a model training method, which includes: obtaining a training sample set, wherein the training samples are marked with product marketing features and a price tag, the product marketing features include product attribute features of the corresponding product, and marketing activity features corresponding to the marketing activities in which the product participates in the same historical marketing period, and the price tag is determined based on the sales data of the product in the historical marketing period; using the product marketing features marked by the training samples in the training sample set as input, and using the price tag corresponding to the input product marketing features as output, training the machine learning model to be trained, and obtaining a prediction model for predicting the price of the product.

[0018] In some embodiments, the machine learning model is a tree model or a neural network.

[0019] In some embodiments, the tree model includes an extreme gradient boosting model; the neural network includes a model that combines a linear model and a deep model.

[0020] In a third aspect, an embodiment of the present specification provides a method for predicting the final price of a commodity, the method comprising: obtaining marketing activity information corresponding to the marketing activities in which the target commodity participates during a target marketing period; performing feature extraction on the marketing activity information to obtain marketing activity features; obtaining commodity attribute features of the target commodity; combining the commodity attribute features and the marketing activity features into commodity marketing features, and inputting the commodity marketing features into a prediction model trained using the method described in any implementation method of the second aspect to obtain the final price of the target commodity during the target marketing period.

[0021] In some embodiments, the marketing activities that the target product participates in during the target marketing period include marketing activities set by the merchant to which the target product belongs.

[0022] In some embodiments, marketing activities have corresponding activity categories; and feature extraction is performed on marketing activity information to obtain marketing activity features, including: selecting target marketing activities that participate in generating marketing activity features from the above-mentioned participating marketing activities according to the activity categories to which the above-mentioned participating marketing activities respectively belong; and feature extraction is performed on the marketing activity information corresponding to each selected target marketing activity to obtain marketing activity features.

[0023] In some embodiments, based on the activity categories to which the above-mentioned participating marketing activities respectively belong, target marketing activities that participate in generating marketing activity features are selected from the above-mentioned participating marketing activities, including: for each activity category in the activity categories to which the above-mentioned participating marketing activities respectively belong, the following marketing activity selection operations are performed: if the marketing activities under the activity category are set with corresponding priorities, then based on the priorities, a marketing activity is selected as the target marketing activity from the marketing activities belonging to the activity category among the above-mentioned participating marketing activities; if the marketing activities under the activity category are not set with corresponding priorities, then based on the discount amount, a marketing activity is selected as the target marketing activity from the marketing activities belonging to the activity category among the above-mentioned participating marketing activities.

[0024] In some embodiments, feature extraction is performed on the marketing campaign information corresponding to each selected target marketing campaign to obtain marketing campaign features, including: determining a base price corresponding to the target product; based on the base price, converting the marketing campaign information corresponding to each target marketing campaign into corresponding value pairs, the value pairs including a threshold amount and a discount amount; and feature processing is performed on each value pair obtained from the conversion to obtain marketing campaign features.

[0025] In some embodiments, performing feature processing on each value pair obtained by the conversion includes: performing a division operation, a subtraction operation, a binary feature addition, and a normalization on each value pair obtained by the conversion.

[0026] In some embodiments, the activity category includes at least one of the following: direct price reduction for single products, discounts for purchases above a certain amount, store coupons, brand coupons, and category coupons.

[0027] In some embodiments, determining a base price corresponding to a target product includes: if there is a single product direct reduction category in the activity categories to which the above-mentioned participating marketing activities respectively belong, then determining the base price based on the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category.

[0028] In some embodiments, determining the base price corresponding to the target product further includes: if there is no single product direct reduction category in the activity categories to which the above-mentioned participating marketing activities respectively belong, then determining the original price of the target product as the base price.

[0029] In a fourth aspect, an embodiment of the present specification provides a commodity processing method, the method comprising: obtaining a predicted price corresponding to a target commodity in a target marketing period, the predicted price being a price predicted by a method as described in any implementation method in the third aspect; if the predicted price exceeds a price threshold corresponding to the target commodity, executing a warning operation for the merchant to which the target commodity belongs.

[0030] In a fifth aspect, an embodiment of the present specification provides a method for processing a marketing activity, the method comprising: obtaining a predicted price corresponding to a target product in a target marketing period, the predicted price being a price predicted by a method as described in any implementation method in the third aspect; if the predicted price exceeds a price threshold corresponding to the target product, determining prompt information for indicating the need to modify the marketing activity combination plan; and displaying the prompt information to the merchant to which the target product belongs.

[0031] In a sixth aspect, an embodiment of the present specification provides a method for processing a marketing activity, the method comprising: obtaining a predicted price corresponding to at least one target commodity in a target marketing period, the predicted price being a price predicted by a method as described in any implementation method in the third aspect; based on the predicted price, selecting a target commodity that meets a recommendation condition from the at least one target commodity; and pushing the selected target commodity to a consumer user.

[0032] In a seventh aspect, an embodiment of the present specification provides a training sample generating device, which includes: an acquisition unit, configured to acquire commodity marketing features corresponding to commodities in historical marketing periods, the commodity marketing features including commodity attribute features of commodities and marketing activity features corresponding to marketing activities participated in by commodities in historical marketing periods; a determination unit, configured to determine the corresponding price tag of commodities in historical marketing periods based on sales data of commodities in historical marketing periods; a generation unit, configured to generate training samples marked with commodity marketing features and price tags, the training samples being used to train a prediction model for predicting the price of commodities.

[0033] In an eighth aspect, an embodiment of the present specification provides a model training device, which includes: an acquisition unit, configured to acquire a training sample set, wherein the training samples are marked with product marketing features and a price tag, the product marketing features include product attribute features of the corresponding product, and marketing activity features corresponding to the marketing activities in which the product participates in the same historical marketing period, and the price tag is determined based on the sales data of the product in the historical marketing period; a training unit, configured to use the product marketing features marked by the training samples in the training sample set as input, and use the price tag corresponding to the input product marketing features as output, to train the machine learning model to be trained, and obtain a prediction model for predicting the price of the product.

[0034] In a ninth aspect, an embodiment of the present specification provides a device for predicting the final price of a commodity, the device comprising: a first acquisition unit, configured to acquire marketing activity information corresponding to the marketing activities in which the target commodity participates during a target marketing period; a feature extraction unit, configured to perform feature extraction on the marketing activity information to obtain marketing activity features; a second acquisition unit, configured to acquire commodity attribute features of the target commodity; a prediction unit, configured to combine commodity attribute features and marketing activity features into commodity marketing features, and input the commodity marketing features into a prediction model trained by the method described in any implementation method in the second aspect, to obtain the final price of the target commodity during the target marketing period.

[0035] In a tenth aspect, an embodiment of the present specification provides a commodity processing device, which includes: an acquisition unit, configured to obtain a predicted price corresponding to a target commodity in a target marketing period, the predicted price being a price predicted by a method as described in any implementation method in the third aspect; and an early warning unit, configured to execute an early warning operation for a merchant to which the target commodity belongs if the predicted price exceeds a price threshold corresponding to the target commodity.

[0036] In the eleventh aspect, an embodiment of the present specification provides a marketing activity processing device, which includes: an acquisition unit, configured to obtain a predicted price corresponding to a target product in a target marketing period, the predicted price being a price predicted by a method as described in any implementation method in the third aspect; a determination unit, configured to determine prompt information for indicating the need to modify the marketing activity combination plan if the predicted price exceeds a price threshold corresponding to the target product; and an output unit, configured to display the prompt information to the merchant to which the target product belongs.

[0037] In a twelfth aspect, an embodiment of the present specification provides a product recommendation device, which includes: an acquisition unit, configured to obtain a predicted price corresponding to at least one target product in a target marketing period, the predicted price being a price predicted by a method as described in any implementation method in the third aspect; a selection unit, configured to select a target product that meets a recommendation condition from the at least one target product based on the predicted price; and a push unit, configured to push the selected target product to a consumer user.

[0038] In the thirteenth aspect, an embodiment of the present specification provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed in a computer, the computer is caused to execute a method described in any one of the implementations in the first, second, third, fourth, fifth and sixth aspects.

[0039] In a fourteenth aspect, an embodiment of the present specification provides a computing device, comprising a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method described in any one of the implementation methods of the first aspect, the second aspect, the third aspect, the fourth aspect, the fifth aspect and the sixth aspect is implemented.

[0040] The solution provided by the above embodiments of this specification generates training samples based on the commodity marketing characteristics and the price tags corresponding to the commodities in the historical marketing period, so as to realize targeted information generation. Among them, the price tag can represent the high-frequency price of the commodity in the historical marketing period. The high-frequency price can be understood as the high-frequency transaction price. The generated training samples can be used to characterize the relationship between the superposition of marketing activities at the commodity granularity and the high-frequency price.

[0041] By obtaining a training sample set formed by training samples generated by the training sample generation method as described above, and training a machine learning model to be trained based on the training sample set, the machine learning model can learn the degree of influence of different marketing activities on the product, and then the trained prediction model for predicting the final price of the product can predict the final price with higher accuracy.

[0042] By deploying the above-mentioned prediction model online to predict the high-frequency prices of goods under different marketing activity combinations in the future, risk warnings can be accurately delivered based on the predicted prices. This can prevent asset losses and personnel risks caused by the wrong or multiple marketing activities leading to too low prices. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings required for use in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without creative work, and this specification can also be applied to other similar scenarios based on the provided drawings. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0044] Figure 1 is an exemplary system architecture diagram to which some embodiments of the present specification may be applied;

[0045] Figure 2 is a flowchart of an embodiment of a method for generating training samples according to the present specification;

[0046] Figure 3It is a schematic diagram of the composition structure of the training samples;

[0047] Figure 4 is a schematic diagram of an application scenario of the training sample generation method according to this specification;

[0048] Figure 5 is a flowchart of another embodiment of the training sample generation method according to the present specification;

[0049] Figure 6 is a flowchart of an embodiment of a model training method according to the present specification;

[0050] Figure 7 is a schematic diagram of an application scenario of the model training method according to this specification;

[0051] Figure 8 is a flowchart of an embodiment of a method for predicting the commodity price according to the present specification;

[0052] Fig. 9 It is a schematic diagram of an application scenario of the commodity price prediction method according to this specification;

[0053] Fig.10 is a flow chart of an embodiment of a commodity processing method according to the present specification;

[0054] Fig.11 is a flow chart of an embodiment of a marketing activity processing method according to the present specification;

[0055] Fig.12 is a flow chart of an embodiment of a product recommendation method according to the present specification;

[0056] Fig.13 is a structural schematic diagram of a training sample generating device according to this specification;

[0057] Fig.14 is a structural schematic diagram of a model training device according to this specification;

[0058] Fig.15 It is a structural schematic diagram of a commodity price prediction device according to this specification;

[0059] Fig.16 is a structural schematic diagram of a commodity processing device according to the present specification;

[0060] Fig.17 is a structural schematic diagram of a marketing activity processing device according to this specification;

[0061] Fig.18 It is a structural schematic diagram of a product recommendation device according to this specification. DETAILED DESCRIPTION

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. The embodiments described are only a part of the embodiments, rather than all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0063] It should be noted that, in the absence of conflict, the embodiments in this specification and the features in the embodiments may be combined with each other.

[0064] As mentioned above, reasonable estimates and forecasts are needed for the transaction prices under the superposition of multiple marketing activities. This will enable the risk control system to accurately reach risk warnings based on the predicted prices with high accuracy, so as to prevent financial losses and personnel risks caused by incorrect or multiple marketing activities that result in too low transaction prices.

[0065] Based on this, some embodiments of this specification disclose a training sample generation method, a model training method, and a commodity price prediction method. Specifically, Figure 1 An exemplary system architecture diagram suitable for use with these embodiments is shown.

[0066] like Figure 1 As shown, an exemplary system architecture 100 is shown. System architecture 100 may include devices 101, 102, and 103. Devices 101, 102, and 103 may be hardware devices or software that support network connection to provide various network services. When the device is hardware, it may be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and servers, etc. At this time, as a hardware device, it may be implemented as a distributed device group consisting of multiple devices, or it may be implemented as a single device. When the device is software, it may be installed in the electronic devices listed above. At this time, as software, it may be implemented as, for example, multiple software or software modules for providing distributed services, or it may be implemented as a single software or software module. No specific limitation is made here.

[0067] In practice, a device can provide corresponding network services by installing corresponding client applications or server applications. After the client application is installed, the device can be embodied as a client in network communication. Correspondingly, after the server application is installed, it can be embodied as a server in network communication.

[0068] As an example, in Figure 1In the embodiment, devices 101, 102, and 103 are all embodied as servers. Specifically, device 101 may be a server installed with a server application for generating training samples. Device 102 may be a server installed with a server application for training models. Device 103 may be a server installed with a server application for predicting the price of goods. It should be noted that device 101 and device 102 may be the same device or different devices, which is not specifically limited here.

[0069] When the personnel of devices 101, 102, and 103 have a need to obtain a prediction model for predicting the price of goods, they can first prepare training samples. For example, by executing a specific training sample generation trigger operation, a training sample generation request is sent to device 101. The request may include, for example, a set of commodity identifiers, and several historical marketing time periods corresponding to each commodity identifier in the commodity identifier set. According to the request, device 101 may generate a training sample corresponding to each of the several historical marketing time periods for the commodity indicated by the commodity identifier. The training sample may be marked with commodity marketing features and a price tag. The commodity marketing features may include commodity attribute features of the commodity, and marketing activity features corresponding to the marketing activities in which the commodity participated in the historical marketing period. In addition, device 101 may store the generated training samples in a designated storage location.

[0070] Then, the above personnel can perform a specific model training trigger operation to send a model training request to the device 102. The request may include the address of the above storage location. The device 102 can obtain the training sample set from the above storage location according to the address, and train the prediction model for predicting the product price according to the training sample set.

[0071] Then, the above personnel can deploy the prediction model to the device 103. After that, the personnel who need to predict the price of the goods can send a prediction request for the price of the goods to the device 103. The device 103 can obtain the features used for prediction (such as product marketing features) according to the prediction request, and input the features into the prediction model to obtain the corresponding prediction results.

[0072] Optionally, since generating training samples and predicting the price of goods take up fewer resources than training models, the training sample generation operation and the product price prediction operation can be implemented on the front end, so devices 101 and 103 can be embodied as clients. Specifically, a client application for generating training samples can be installed on device 101. A client application for predicting the price of goods can be installed on device 103, and the client application can be integrated with the prediction model described above. In addition, device 102 can still be embodied as a server. Devices 101 and 102 are different devices.

[0073] It should be noted that the training sample generation method provided in some embodiments of this specification can be executed by device 101, and accordingly, the training sample generation device can be set in device 101. The model training method provided in some embodiments of this specification can be executed by device 102, and accordingly, the model training device can be set in device 102. In addition, the commodity arrival price prediction method provided in some embodiments of this specification can be executed by device 103, and accordingly, the commodity arrival price prediction device can be set in device 103.

[0074] It should be emphasized that Figure 1 The number of devices in the embodiment is only illustrative. Any number of devices may be provided according to implementation requirements.

[0075] Continue to refer Figure 2 , which shows a process 200 of an embodiment of a training sample generation method according to this specification, comprising the following steps:

[0076] Step 201, obtaining commodity marketing characteristics corresponding to the commodity in the historical marketing period, the commodity marketing characteristics including commodity attribute characteristics of the commodity and marketing activity characteristics corresponding to the marketing activities participated in by the commodity in the historical marketing period;

[0077] Step 202, determining the actual price tag corresponding to the product in the historical marketing period according to the sales data of the product in the historical marketing period;

[0078] Step 203: Generate training samples marked with product marketing features and price tags.

[0079] In this embodiment, the execution subject of the training sample generation method (for example Figure 1 The device 101 shown in the figure may execute step 201 in response to receiving a training sample generation request. The request may be manually triggered or automatically triggered, which is not specifically limited here.

[0080] It should be noted that the training sample generation request may include, for example, a set of commodity identifiers. Optionally, the request may include, while including the set of commodity identifiers, a number of historical marketing time periods corresponding to each commodity identifier in the set of commodity identifiers. For the commodity indicated by the commodity identifier and each of the several historical marketing time periods corresponding to the commodity identifier, the execution subject may obtain the training sample corresponding to the commodity in the historical marketing time period by executing steps 201-203.

[0081] In practice, the commodity attribute characteristics may show the commodity identification of the corresponding commodity and at least one identification of the following to which the commodity belongs: category, industry, brand, etc.

[0082] The marketing period may be a period for product promotion, such as Double 5 (May 5), 618 (June 18), Double 11 (November 11), Double 12 (December 12), etc. in real life. The historical marketing period may be a marketing period in a specific time period in the past. The specific time period in the past may be, for example, the past 2 weeks, the past 1 month, the past 2 months, the past 3 months, the past half year, or the past year, etc., which is not specifically limited here.

[0083] Marketing activities can be understood as means of promoting product sales through price cuts and discounts. The marketing activities that products participated in during the historical marketing period include the marketing activities set up by the merchant to which the products belong. Optionally, the marketing activities that products participated in during the historical marketing period may include not only the marketing activities set up by the merchant to which the products belong, but also the marketing activities set up by the managers of the e-commerce platform where the products are located. The categories of marketing activities may include, for example, but are not limited to, direct price reduction for single products, full discounts, store coupons, brand coupons and / or category coupons, etc., which are not specifically limited here.

[0084] Among them, single product direct reduction can be understood as directly reducing the original price of the product by a specific amount. Marketing activities under the single product direct reduction category can include the price of the product after the price reduction.

[0085] Full discount can be divided into two subcategories, such as full discount and full discount. Among them, full discount can be understood as when the total amount of the goods to be purchased reaches the threshold amount, a certain amount can be reduced or exempted. For example, the marketing campaign "20 yuan off for purchases over 100 yuan" can be classified into the full discount subcategory. Among them, 100 yuan is the threshold amount, and 20 yuan is the amount of reduction. Full discount can be understood as when the total number of goods to be purchased reaches the threshold number of pieces, or the total amount of goods to be purchased reaches the threshold amount, a certain discount is given. For example, the marketing campaigns "10% off for purchases over 200 yuan" and "10% off for purchases over 3 pieces" can be classified into the full discount subcategory. Among them, 3 pieces are the threshold number of pieces.

[0086] A store coupon can be understood as a coupon that can be used when purchasing any product from the store to which the store coupon belongs. Marketing activities under the store coupon category can be reflected as store coupons. Store coupons include face value, required consumption amount, validity period, etc. For example, a store coupon from store S may include a face value of 50, a required consumption amount of 300, and a validity period from January 1 to January 3, 2020. The meaning of this store coupon is that if you spend more than 300 yuan at store S from January 1 to January 3, 2020, you can use this store coupon to reduce the consumption amount by 50 yuan.

[0087] Brand coupons can be understood as coupons that can be used when purchasing any product under the brand corresponding to the brand coupon. Marketing activities under the brand coupon category can be reflected as brand coupons. Category coupons can be understood as coupons that can be used when purchasing any product under the category corresponding to the category coupon. Marketing activities under the category coupon category can be reflected as category coupons. Among them, brand coupons and category coupons can respectively include face value, required consumption amount, validity period, etc.

[0088] Based on the above description, it can be seen that store coupons, brand coupons and category coupons have different scopes of use.

[0089] In practice, the discount amounts of different marketing activities can be superimposed on each other, but usually only one of each type of marketing activity can be effective. For example, for any activity category, there are activities A and B under the activity category. Users can use the discounts of either activity A or activity B, but cannot use the discounts of both activities A and B at the same time.

[0090] In step 201, the historical marketing period can be the period in the training sample generation request as described above, or can be a preset period in the past specific time period as described above during which the corresponding product is promoted. In this step, the execution subject can use various methods to obtain the product marketing features corresponding to the historical marketing period.

[0091] As an example, the commodity marketing features corresponding to the commodity in the historical marketing period may be pre-generated and stored in a specific storage location. The execution subject may obtain the commodity marketing features from the storage location.

[0092] As another example, the commodity attribute characteristics of the commodity and the marketing activity characteristics corresponding to the historical marketing period have been pre-generated. The above-mentioned execution subject can obtain the commodity attribute characteristics and the marketing activity characteristics, and combine the two to form the commodity marketing characteristics. Here, the commodity attribute characteristics and the marketing activity characteristics can be combined into the commodity marketing characteristics according to the specified combination method.

[0093] It should be noted that the marketing activity feature can be obtained by converting the numerical pairs corresponding to the marketing activity information of each target marketing activity related thereto, and performing feature processing. The target marketing activities can be all the marketing activities in which the product corresponding to the marketing activity feature participates in the corresponding historical marketing period, or part of the marketing activities among all the marketing activities. The marketing activity information includes the activity content of the marketing activity to which it corresponds. The numerical pair may include the threshold amount and the discount amount determined according to the marketing activity information to which it corresponds. The feature processing includes the following operation items: division operation, subtraction operation, binary feature addition, and normalization. Therefore, the marketing activity feature can show the division operation result, the binary feature, the normalized subtraction operation result, and the normalized numerical pair.

[0094] In step 202, the sales data of the product in the historical marketing period may include multiple prices generated by historical orders of the product in the historical marketing period, and the number of units sold corresponding to the multiple prices. The execution entity may calculate the corresponding price tag of the product in the historical marketing period based on the multiple prices and the number of units sold.

[0095] As an example, a weighted average of the multiple prices and the corresponding sales quantities may be taken to obtain an average price, and the average price may be used as a price label.

[0096] As another example, after obtaining the average price received as described above, the average price received may be normalized, and the normalized value obtained may be used as the price received label.

[0097] In step 203, after obtaining the commodity marketing features and the price tag corresponding to the commodity in the historical marketing period, the execution subject may generate training samples marked with the commodity marketing features and the price tag. Figure 3 As shown in FIG, it shows a schematic diagram of the composition structure of the training samples. Figure 3 In the above, product marketing features consist of product attribute features and marketing activity features, and training samples consist of product marketing features and price tags.

[0098] It should be noted that the above-mentioned execution subject can store the generated training samples to a specified storage location. Alternatively, after completing the corresponding training sample generation task according to the training sample generation request as described above, the above-mentioned execution subject can automatically send each generated training sample to the model training end (for example, Figure 1 The device 102 shown in the figure) is used by the model training end to perform corresponding model training tasks according to each training sample.

[0099] Continue to view Figure 4 , Figure 4 This is a schematic diagram of an application scenario of the training sample generation method according to this embodiment. In order to prevent merchants on the e-commerce platform from suffering financial losses due to incorrect or multiple marketing activities, which lead to too low a price for the goods, and at least one of the following risks for the staff of the merchant who set up the marketing activities: increased workload, dismissal, compensation for losses, etc., it is necessary to train a prediction model for predicting the price of goods on the e-commerce platform, and deploy the prediction model online to provide merchants with a price prediction function with high prediction accuracy.

[0100] Before obtaining the prediction model, it is necessary to enter the training sample preparation stage. This application scenario shows the content related to the training sample preparation stage.

[0101] In this application scenario, the sample manager who is related to the above-mentioned e-commerce platform and is responsible for generating training samples can have a client application related to the generation of training samples installed on the terminal device used. The sample manager can perform a specific training sample generation trigger operation in the client application, so that the terminal device sends a training sample generation request to the background server of the client application according to the trigger operation. The request may include a set of product identifications and several historical marketing time periods corresponding to each product identification in the product identification set. Among them, the product identification in the product identification set is the identification of the product sold by the merchant through the above-mentioned e-commerce platform. Here, it is assumed that the product identification set includes the identification of products G1, ..., Gn. The identifications of products G1, ..., Gn all correspond to historical marketing time periods T1 and T2.

[0102] After receiving the request, the backend server can obtain the product marketing features corresponding to the product indicated by the product identifier in the historical marketing period for each product identifier in the product identifier set and each of the several historical marketing periods corresponding to the product identifier. The product marketing features include the product attribute features of the product and the marketing activity features corresponding to the marketing activities in which the product participated in the historical marketing period. In addition, the backend server can also determine the corresponding price tag of the product in the historical marketing period based on the sales data of the product in the historical marketing period. Afterwards, the backend server can generate training samples marked with the product marketing features and the price tag.

[0103] It should be pointed out that for the convenience of description, Figure 4 In the example, product G1 and historical marketing period T1 are represented as<G1,T1> , product G1 and historical marketing period T2 are represented as<G1,T2> , the product Gn and the historical marketing period T1 are expressed as<Gn,T1> , and the product Gn and the historical marketing period T2 are expressed as<Gn,T2> .

[0104] The training sample generation method provided in this embodiment can realize targeted information generation by obtaining the commodity marketing features corresponding to the commodity in the historical marketing period, and then determining the price tag corresponding to the commodity in the historical marketing period based on the sales data of the commodity in the historical marketing period, and then generating training samples marked with the marketing features of the commodity and the price tag. The generated training samples can be used to characterize the relationship between the superposition of marketing activities at the commodity granularity and the high-frequency price. In addition, the generated training samples can be used to train a prediction model for predicting the price of the commodity. The prediction accuracy can be improved by predicting the price of the commodity using the prediction model. In addition, after the prediction model is deployed online, the prediction results of the prediction model can be used in risk control scenarios, which helps the risk control system to achieve accurate contact of risk warnings, thereby improving the adoption rate of the risk control system, and preventing the transaction price from being too low due to missetting or multiple marketing activities, which may cause financial losses and personnel risks.

[0105] Further references Figure 5 , which shows a process 500 of another embodiment of a training sample generation method, comprising the following steps:

[0106] Step 501, obtaining marketing activity information corresponding to each marketing activity that the product participated in during the historical marketing period;

[0107] Step 502, selecting target marketing activities for generating marketing activity features from among the marketing activities according to the activity categories to which the marketing activities belong respectively;

[0108] Step 503, determining the base price corresponding to the product;

[0109] Step 504, according to the base price, convert the marketing activity information corresponding to each target marketing activity into a corresponding value pair, where the value pair includes a threshold amount and a discount amount;

[0110] Step 505, performing feature processing on each value pair obtained by the conversion to obtain marketing activity features;

[0111] Step 506, obtaining commodity attribute characteristics of the commodity;

[0112] Step 507, combining the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics;

[0113] Step 508, determining the actual price tag corresponding to the product in the historical marketing period according to the sales data of the product in the historical marketing period;

[0114] Step 509: Generate training samples marked with product marketing features and price tags.

[0115] In this embodiment, the execution subject of the training sample generation method (for example Figure 1 The device 101 shown in FIG. 1 may execute step 501 in response to receiving a training sample generation request. The request may be manually triggered or automatically triggered, which is not specifically limited herein.

[0116] It should be noted that for explanations of training sample generation requests, historical marketing periods, and marketing activities, please refer to Figure 2 The relevant descriptions in the corresponding embodiments will not be repeated here.

[0117] In step 501, the product identification, historical marketing period, and marketing activity information corresponding to each marketing activity participated by the product in the historical marketing period can be stored in advance. The execution subject can obtain the corresponding marketing activity information according to the product identification and the historical marketing period.

[0118] In step 502, the activity categories may include, but are not limited to, single product direct discount, full discount, store coupons, brand coupons and / or category coupons, etc., which are not specifically limited here.

[0119] For each of the activity categories to which each marketing activity belongs, the execution subject may perform the following marketing activity selection operations:

[0120] If the marketing activities under the activity category are set with corresponding priorities, then according to the priorities, a marketing activity is selected as a target marketing activity from among the marketing activities belonging to the activity category in various marketing activities;

[0121] If the marketing activities under the activity category have not been set with corresponding priorities, a marketing activity is selected as a target marketing activity from among the marketing activities belonging to the activity category in accordance with the discount amount.

[0122] Specifically, if the marketing activities under the activity category are set with corresponding priorities, the marketing activities with the highest priority among the marketing activities belonging to the activity category can be selected as the target marketing activities. If the marketing activities under the activity category are not set with corresponding priorities, a preset number of marketing activities with the largest discount amount can be selected from the marketing activities belonging to the activity category among the marketing activities. The preset number can be, for example, 1, 2, or 3, etc., which is not specifically limited here.

[0123] In step 503, if there is a single product direct reduction category in the activity category to which each marketing activity belongs, the benchmark price can be determined based on the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category. For example, the marketing activities under the single product direct reduction category are usually set with corresponding priorities. The target marketing activity belonging to the single product direct reduction category can be the marketing activity with the highest priority. The reduced price in the marketing activity information corresponding to the target marketing activity can be determined as the benchmark price.

[0124] In addition, if there is no single product direct price reduction category in the activity category to which each marketing activity belongs, the original price of the product can be determined as the base price.

[0125] In step 504, the execution entity may convert the marketing activity information corresponding to each target marketing activity into a corresponding value pair according to the determined benchmark price.

[0126] As an example, assume that the base price of product A is RMB 100. The marketing activities that product A participated in during the historical marketing period include: RMB 100 off for purchases over RMB 500 and 20% off for purchases over 3 pieces. The corresponding value pair for the marketing activity "RMB 100 off for purchases over RMB 500" can be <500, 100>, where 500 is the threshold amount and 100 is the discount amount. The corresponding value pair for the marketing activity "20% off for purchases over 3 pieces" can be <300, 60>. 300 is the threshold amount and 60 is the discount amount. It should be noted that the threshold amount = number of pieces * base price, and the number of pieces is 3. The discount amount = the threshold amount * (1-0.8).

[0127] In step 505, the execution entity may perform feature processing on each value pair obtained by the conversion to obtain marketing activity features. Specifically, the execution entity may perform division operation, subtraction operation, binary feature addition, and normalization on each value pair obtained by the conversion.

[0128] In the division operation, for each value pair, the discount amount and the threshold amount in the value pair can be divided to obtain the discount ratio. In the subtraction operation, for each value pair, the discount amount and the threshold amount in the value pair can be subtracted from the base price to obtain the difference. In the binary feature addition operation, a Boolean variable can be used to indicate whether each type of discount exists. In the normalization operation, the threshold amount and the discount amount in each value pair, as well as the difference obtained by the subtraction operation, can be divided by the base price.

[0129] In step 506, the commodity attribute feature may show the commodity identifier of the corresponding commodity and the identifier of at least one of the following items to which the commodity belongs: category, industry, brand, etc.

[0130] As an example, the commodity attribute characteristics of the commodity may be generated in advance, and the execution subject may obtain the generated commodity attribute characteristics according to the commodity identifier of the commodity.

[0131] As another example, the execution entity may obtain corresponding commodity attribute information according to the commodity identifier of the commodity, and perform feature extraction on the commodity attribute information according to a predetermined feature extraction method to obtain commodity attribute features.

[0132] In step 507, the execution entity may combine the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics according to a specified combination method.

[0133] In step 508, the sales data of the product in the predetermined historical marketing period may include multiple prices obtained from historical orders of the product in the historical marketing period, and the number of sales corresponding to the multiple prices. The execution entity may, for example, perform a weighted average of the multiple prices obtained and the corresponding number of sales to obtain an average price obtained. Then, the average price obtained is normalized, and the normalized value is used as the price obtained label. When normalizing the average price obtained, the average price obtained may be divided by the base price.

[0134] In step 509, the execution entity may generate training samples marked with product marketing features and price tags.

[0135] from Figure 5 It can be seen that Figure 2 Compared with the corresponding embodiments, the training sample generation method provided in this embodiment highlights the steps of expanding the method for obtaining the marketing characteristics of the product, especially the steps of obtaining the marketing activity characteristics. The scheme provided in this embodiment can realize the feature abstraction of the marketing activity information in character form and uniformly convert it into a dual-attribute numerical form of threshold + discount amount. By adopting a numerical feature representation method for marketing activity information, an effective data foundation can be laid for subsequent machine learning modeling, thereby effectively ensuring the prediction accuracy of the product's price.

[0136] Further references Figure 6 , which shows a process 600 of an embodiment of a model training method, comprising the following steps:

[0137] Step 601, obtaining a training sample set, wherein the training sample includes product marketing features and price tags, the product marketing features include product attribute features of the corresponding product and marketing activity features corresponding to the marketing activities in the same historical marketing period in which the product participates, and the price tags are determined according to the sales data of the product in the historical marketing period;

[0138] Step 602, taking the commodity marketing features marked by the training samples in the training sample set as input, taking the price tag corresponding to the input commodity marketing features as output, training the machine learning model to be trained, and obtaining a prediction model for predicting the price of the commodity.

[0139] In this embodiment, the execution subject of the model training method (for example Figure 1 The device 102 shown in the figure may execute step 601 in response to receiving a model training request. The model training request may be manually triggered or automatically triggered, which is not specifically limited here.

[0140] As an example, if the model training request is automatically triggered, the model training request may be, for example, the generator of the training sample set (e.g. Figure 1 The device 101 shown in the figure automatically sends the training sample to the above-mentioned execution entity after completing the training sample generation task related to the training sample set.

[0141] Among them, the model training request may, for example, include the address of the storage location where the training sample set is located, and any of the following items: the machine learning model to be trained, and the address of the storage location where the model is located.

[0142] The training samples in the training sample set are Figure 2 or Figure 5 The training samples generated by the training sample generation method provided in the corresponding embodiment. For the commodity marketing feature marked by any training sample in the training sample set, the marketing activity feature included in the commodity marketing feature can be obtained by performing feature processing on the numerical value pairs obtained by converting the marketing activity information corresponding to each target marketing activity related to the marketing activity feature. The feature processing includes the following operation items: division operation, subtraction operation, binary feature addition, and normalization. Therefore, the marketing activity feature can show the division operation result, the binary feature, the normalized subtraction operation result, and the normalized numerical value pair.

[0143] The machine learning model to be trained may be an untrained machine learning model or an incompletely trained machine learning model, and no specific limitation is made here.

[0144] Specifically, the machine learning model may be a tree model or a neural network, etc. The tree model may include but is not limited to an XGBoost (eXtreme Gradient Boosting) model. The neural network may include but is not limited to a Wide-Deep model. The Wide-Deep model combines a linear model and a deep model.

[0145] XGBoost has the advantages of high efficiency, accuracy, and model interactivity. For example, with the help of OpenMP, XGBoost can automatically use the multi-cores of a single CPU (Central Processing Unit) for parallel computing, so it has the advantage of high efficiency. Among them, OpenMP is a multi-threaded programming scheme for shared memory parallel systems. In addition, the main reason for the improvement in accuracy is that the XGBoost model, compared with the traditional GBDT (Gradient Boosting Decision Tree), adds control over model complexity and later pruning processing, making the learned model less prone to overfitting. In addition, in terms of the interactivity of the model, the gradient and Hessian matrix of the objective function can be calculated, so users can customize the objective function when training the model. XGBoost allows users to customize the error measurement method during cross-validation.

[0146] Regarding the Wide-Deep model, it should be pointed out that Wide-Deep can also be expressed as Wide&Deep, and there is no limitation on the writing method. Among them, the Wide-Deep model aims to enable the trained model to simultaneously acquire memorization and generalization capabilities. Memory ability can be interpreted as learning those features that often appear at the same time and exploring the co-occurrence in historical data. Generalization ability is based on migration correlation and exploring new feature combinations that have hardly appeared before. Specifically from the perspective of model definition, Wide refers to the generalized linear model (Wide Linear Model), and Deep refers to the deep neural network (Deep Netural Network, DNN). In practice, the objective function used by the Wide-Deep model may include but is not limited to MSE (Mean Square Error).

[0147] In step 601, the execution entity may, for example, obtain a training sample set according to the model training request as described above.

[0148] In step 602, the execution entity may take the commodity marketing features marked by the training samples in the training sample set as input, and take the final price tag corresponding to the input commodity marketing features as output, train the machine learning model to be trained, and obtain a prediction model for predicting the final price of the commodity.

[0149] As an example, the above-mentioned execution entity can perform the following model training operations: input the product marketing features marked by each training sample in the training sample set into the machine learning model to be trained in turn, and obtain the predicted price received corresponding to the product marketing features marked by each training sample in the training sample set; compare the price received label corresponding to the product marketing features marked by each training sample in the training sample set with the predicted price received, and obtain the prediction accuracy of the machine learning model after this round of training; determine whether the prediction accuracy is greater than a preset accuracy threshold; if it is greater than the preset accuracy threshold, use the machine learning model after this round of training as a prediction model for predicting the product price received.

[0150] In addition, if the prediction accuracy is not greater than the preset accuracy threshold, the machine learning model after this round of training can be used as the machine learning model to be trained, and the above model training operation can be continued.

[0151] It should be noted that when the machine learning model to be trained is a Wide-Deep model, the machine learning model includes a DNN and a linear model, wherein the inputs of the DNN and the linear model are different.

[0152] For example, for a product marketing feature marked by any training sample in the training sample set, the content input to the DNN may include the product attribute feature included in the product marketing feature, and the normalized value pair in the marketing activity feature included in the product marketing feature. The content input to the linear model may include the division operation result, binary feature and normalized subtraction operation result in the marketing activity feature. It should be understood that the input of the DNN and the linear model can be designed according to actual needs and is not specifically limited here.

[0153] Continue to view Figure 7 , Figure 7 is a schematic diagram of an application scenario of the model training method according to this embodiment. Figure 4 Take the e-commerce platform in the corresponding application scenario as an example. This application scenario shows the content related to the model training stage. The model training stage is: Figure 4 The corresponding application scenario illustrates the next stage of the training sample preparation phase.

[0154] In this application scenario, the model management personnel who are responsible for model training and are related to the above-mentioned e-commerce platform may have client applications related to model training installed on their terminal devices. Figure 4 The sample managers in the corresponding application scenarios can be the same person or different people, and there is no specific limitation here.

[0155] The model manager can perform a specific model training trigger operation in the client application related to model training, so that the terminal device sends a model training request to the background server of the client application according to the trigger operation. The request may include, for example, the address of the storage location where the training sample set is located, and the machine learning model to be trained. The training sample set is a set formed by the training samples generated in the above-mentioned training sample preparation stage.

[0156] After receiving the request, the backend server can obtain the training sample set and the machine learning model to be trained according to the request. Then, the backend server can use the product marketing features marked by the training samples in the training sample set as input, and the price tag corresponding to the input product marketing features as output to train the machine learning model to be trained, and obtain a prediction model for predicting the price of the product.

[0157] The model training method provided in this embodiment can enable the machine learning model to learn the degree of influence of different marketing activities on the product by obtaining the training sample set as described above, and training the machine learning model to be trained based on the training sample set, so that the prediction model obtained by training for predicting the product's final price can predict the final price with high accuracy. In addition, after the prediction model is deployed online, the prediction results of the prediction model can be used in risk control scenarios, which helps the risk control system to achieve accurate risk warning, thereby improving the adoption rate of the risk control system and preventing the transaction price from being too low due to incorrect or multiple marketing activities, which may cause financial losses and personnel risks.

[0158] Further references Figure 8 , which shows a process 800 of an embodiment of a method for predicting the commodity price, including the following steps:

[0159] Step 801, obtaining marketing activity information corresponding to each marketing activity that the target product participates in during the target marketing period;

[0160] Step 802, extracting features from the marketing activity information to obtain marketing activity features;

[0161] Step 803, obtaining commodity attribute characteristics of the target commodity;

[0162] Step 804, the commodity attribute characteristics and the marketing activity characteristics are combined into commodity marketing characteristics, and the commodity marketing characteristics are input into a prediction model for predicting the commodity final price, so as to obtain the final price corresponding to the target commodity in the target marketing period.

[0163] In this embodiment, the execution subject of the commodity price prediction method (for example Figure 1The device 103 shown in FIG. 1 may execute step 801 in response to receiving a price prediction request related to a target commodity and a target historical marketing period. The request may be manually triggered or automatically triggered, which is not specifically limited here.

[0164] As an example, if the request is manually triggered, the request may be triggered by a staff member of the merchant to which the target product belongs who is responsible for setting up marketing activities, by executing a specific product price prediction trigger operation. If the request is automatically triggered, the request may be automatically sent by a thread or device responsible for detecting marketing activities set by the merchant for the product in response to detecting a marketing activity set by the staff member for the target product that takes effect during the target marketing period.

[0165] The request may include a commodity identification of the target commodity and a target marketing period, wherein the target marketing period may be a period for promoting the target commodity in the future.

[0166] It should be noted that the above prediction model is based on Figure 6 The model obtained by training using the model training method provided in the corresponding embodiment.

[0167] In step 801, the execution entity may, for example, obtain marketing activity information corresponding to the marketing activities that the target product participates in during the target marketing period according to the product identifier of the target product and the target historical marketing period.

[0168] The various marketing activities that the target product participates in during the target marketing period may include marketing activities set by the merchant to which the target product belongs. Optionally, the various marketing activities that the target product participates in during the target marketing period may include marketing activities set by the merchant to which the target product belongs and marketing activities set by the administrator of the e-commerce platform where the target product is located.

[0169] In step 802, the execution entity may perform feature extraction on all the acquired marketing activity information. Alternatively, the execution entity may perform feature extraction on part of the marketing activity information in the entire marketing activity information. It should be understood that the feature extraction process in this step is similar to the feature extraction process corresponding to the marketing activity features marked by the training samples used in the model training phase of the prediction model.

[0170] The execution process of step 803 and step 804 of combining the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics is the same as that of step 804. Figure 5 Steps 506-507 in the corresponding embodiment are similar and will not be described in detail here.

[0171] In addition, in step 804, the way of expressing the price of the target product in the target marketing period is consistent with the way of expressing the price tag in the training sample used in the model training stage of the above-mentioned prediction model. For example, if the price tag is an unnormalized price, the price of the target product in the target marketing period is also an unnormalized price. If the price tag is a normalized price, the price of the target product in the target marketing period is also a normalized price.

[0172] It should be understood that the composition structure of the commodity marketing features corresponding to the target commodity in the target marketing period in this embodiment is consistent with the composition structure of the commodity marketing features in the training samples used in the model training stage of the above-mentioned prediction model.

[0173] In some optional implementations of this embodiment, the marketing activity may have a corresponding activity category. The above-mentioned execution subject may extract features from the marketing activity information to obtain marketing activity features in the following manner: according to the activity categories to which each marketing activity belongs, select target marketing activities that participate in generating marketing activity features from each marketing activity; extract features from the marketing activity information corresponding to each selected target marketing activity to obtain marketing activity features.

[0174] In some optional implementations of this embodiment, for each activity category in the activity category to which each marketing activity belongs, the above-mentioned execution subject may perform the following marketing activity selection operation: if the marketing activities under the activity category are set with corresponding priorities, then according to the priorities, a marketing activity is selected as the target marketing activity from the marketing activities belonging to the activity category in the various marketing activities; if the marketing activities under the activity category are not set with corresponding priorities, then according to the discount amount, a marketing activity is selected as the target marketing activity from the marketing activities belonging to the activity category in the various marketing activities. It should be noted that the specific details of the marketing activity selection operation are different from Figure 5 The marketing activity selection operation described in the corresponding embodiment is similar and will not be repeated here.

[0175] In some optional implementations of the present embodiment, the above-mentioned execution entity may adopt the following method to extract features from the marketing campaign information corresponding to each selected target marketing campaign to obtain marketing campaign features: determine the base price corresponding to the target product; based on the base price, convert the marketing campaign information corresponding to each target marketing campaign into a corresponding numerical pair, the numerical pair including a threshold amount and a discount amount; perform feature processing on each numerical pair obtained by the conversion to obtain marketing campaign features.

[0176] In some optional implementations of this embodiment, feature processing is performed on each value pair obtained by conversion, specifically, division operation, subtraction operation, binary feature addition, and normalization are performed on each value pair obtained by conversion. Here, the feature processing process is similar to Figure 5 The feature processing process performed in the corresponding embodiments is similar and will not be repeated here.

[0177] In some optional implementations of this embodiment, the categories of marketing activities may include, for example, but are not limited to, direct discounts on single products, full discounts, store coupons, brand coupons and / or category coupons, etc., which are not specifically limited here.

[0178] In some optional implementations of the present embodiment, the above-mentioned execution entity may determine the base price corresponding to the target commodity in the following manner: if there is a single product direct reduction category in the activity category to which each marketing activity belongs, then the base price is determined according to the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category; if there is no single product direct reduction category in the activity category to which each marketing activity belongs, then the original price of the target commodity is determined as the base price.

[0179] In some optional implementations of this embodiment, the execution entity may feed back the predicted price of the target product in the target historical marketing period to the merchant to which the target product belongs. Specifically, when the price is a normalized price, the execution entity may convert the price into a non-normalized price and feed back the price to the merchant.

[0180] In some optional implementations of this embodiment, the execution subject may also send the corresponding price of the target product in the target historical marketing period to the risk control system, and the risk control system may perform corresponding risk control operations based on the price. For example, the risk control system may determine whether the price is lower than the corresponding price threshold of the target product. If it is lower than the threshold, it will enter the corresponding approval flow, or perform an early warning operation.

[0181] For the approval process, for example, the merchant to which the target product belongs can review whether the purchase price meets expectations. For the warning operation, for example, warning information can be generated and fed back to the merchant through a specific transmission channel. The transmission channel can be, for example, SMS, email, phone, or a specific client application (for example, a client application used when setting up a marketing campaign for the target product).

[0182] It should be noted that the risk control system can be deployed locally on the above-mentioned execution entity or on other devices, and is not specifically limited here. When the risk control system is deployed locally on the above-mentioned execution entity, the above-mentioned execution entity has the function of the risk control system, and the risk control operations performed by the risk control system can also be understood as the risk control operations performed by the above-mentioned execution entity.

[0183] Since this embodiment can accurately predict the impact of a newly emerging marketing activity on a high-frequency purchase price by extracting its corresponding marketing activity features and using the above-mentioned prediction model, the risk control system can accurately reach risk warnings based on the purchase price predicted by the above-mentioned prediction model, thereby avoiding risk omissions and false alarms as much as possible.

[0184] Continue to view Fig. 9 , Fig. 9 is a schematic diagram of an application scenario of the commodity price prediction method according to this embodiment. Figure 7 Take the e-commerce platform in the corresponding application scenario as an example. This application scenario shows the content related to the product price prediction stage. The product price prediction stage is: Figure 7 The corresponding application scenario shows the next stage of the model training phase.

[0185] In this application scenario, the staff responsible for setting up marketing activities of the merchants stationed on the above-mentioned e-commerce platform can install a prediction tool related to the prediction of the product price on the terminal device used. In addition, the terminal device can also be installed with a commodity management client application corresponding to the above-mentioned e-commerce platform for merchants to use. The prediction tool can be, for example, a plug-in in the commodity management client application, etc., which is not specifically limited here.

[0186] After the above staff uses the commodity management client application to set a marketing campaign for commodity G that will take effect in a future marketing period T, they can use the forecasting tool to perform a specific commodity arrival price forecast trigger operation, so that the above terminal device sends a corresponding commodity arrival price forecast request to the background server of the forecasting tool according to the trigger operation. Here, the request may include the commodity identification of commodity G and the marketing period T. It should be noted that the forecasting model for predicting the arrival price of commodities as described above is deployed on the background server.

[0187] After receiving the request for product price prediction, the backend server can obtain the marketing activity information corresponding to each marketing activity in which product G participates in the marketing period T according to the request. The various marketing activities include the marketing activities set by the above-mentioned staff for product G, as well as other marketing activities other than the marketing activities. Then, the backend server can use a specific feature extraction method to extract features from the acquired marketing activity information to obtain marketing activity features. After that, the backend server can obtain the product attribute features of product G. Then, the backend server can combine the product attribute features and the marketing activity features into product marketing features. Then, the backend server can input the product marketing features into the above-mentioned prediction model to obtain the corresponding price of product G in the marketing period T.

[0188] The commodity price prediction method provided in this embodiment can predict the price with high accuracy by using the above prediction model. The price is then transmitted to the risk control system, so that the risk control system can accurately reach the risk warning according to the price, thereby improving the adoption rate of the risk control system and preventing the transaction price from being too low due to wrong or multiple marketing activities, which may cause financial losses and personnel risks.

[0189] As mentioned above, the prediction results of the above prediction model can be used in risk control scenarios. In addition, the prediction results of the above prediction model can also be used by merchants to modify marketing activity combination plans, and can also be used to recommend products to consumer users.

[0190] Based on this, the embodiments of this specification also provide a commodity processing method, a marketing activity processing method, and a commodity recommendation method. Below, the commodity processing method, the marketing activity processing method, and the commodity recommendation method are described one by one.

[0191] See also Fig.10 , which shows a process 1000 of an embodiment of a commodity processing method. The execution subject of the method may be the risk control system as described above. The method comprises the following steps:

[0192] Step 1001, obtaining the predicted price of the target product in the target marketing period;

[0193] Step 1002: If the predicted final price exceeds the final price threshold corresponding to the target product, a warning operation is performed for the merchant to which the target product belongs.

[0194] In this embodiment, the predicted hand price can be adopted Figure 8 The corresponding embodiment provides a method for predicting the commodity price at the time of purchase that predicts the price at the time of purchase.

[0195] For the explanation of the target products and target marketing periods in this embodiment, please refer to Figure 8 The relevant descriptions in the corresponding embodiments are not repeated here. It should be noted that the target product may have a corresponding price threshold. The price threshold may be set by the merchant to which the target product belongs.

[0196] Specifically, it can be obtained from Figure 8 The execution subject of the solution provided by the corresponding embodiment obtains the predicted price of the target product in the target marketing period. The predicted price can be actively sent by the execution subject. Then, the obtained predicted price can be compared with the corresponding price threshold. If the predicted price does not exceed the price threshold, no warning operation is required. If the predicted price exceeds the price threshold, it is necessary to perform a warning operation on the merchant to which the target product belongs.

[0197] In practice, early warning information can be generated and fed back to the merchant to which the target product belongs through a specific transmission channel. The early warning information can be information in the form of text, voice and / or image. The transmission channel can be, for example, text messages, emails, phone calls, or specific client applications (for example, client applications used when setting up marketing activities for the target product).

[0198] The commodity processing method provided in this embodiment obtains the predicted price corresponding to the target commodity in the target marketing period, and then, in response to the predicted price exceeding the price threshold corresponding to the target commodity, performs an early warning operation for the merchant to which the target commodity belongs, thereby achieving accurate contact of risk early warnings, thereby improving the adoption rate of the risk control system and preventing capital losses and personnel risks caused by too low transaction prices due to incorrect or multiple marketing activities.

[0199] Continue to view Fig.11 As shown, it shows a process 1100 of an embodiment of a marketing activity processing method. The execution subject of the method can be the risk control system as described above, or Figure 8 The execution subject of the commodity price prediction method provided in the corresponding embodiment is not specifically limited here. The method includes the following steps:

[0200] Step 1101, obtaining the predicted price of the target product in the target marketing period;

[0201] Step 1102: if the predicted final price exceeds the final price threshold corresponding to the target product, then a prompt message for prompting that the marketing activity combination plan needs to be modified is determined;

[0202] Step 1103: Display prompt information to the merchant to which the target product belongs.

[0203] In this embodiment, the predicted hand price can be adopted Figure 8The corresponding embodiment provides a method for predicting the commodity price at the time of purchase that predicts the price at the time of purchase.

[0204] For the explanation of the target products and target marketing periods in this embodiment, please refer to Figure 8 The relevant descriptions in the corresponding embodiments are not repeated here. It should be noted that the target product may have a corresponding price threshold. The price threshold may be set by the merchant to which the target product belongs.

[0205] Specifically, in step 1101, if the execution subject of the solution provided in this embodiment is the above-mentioned risk control system, then Figure 8 The execution subject of the solution provided in the corresponding embodiment obtains the predicted price of the target product in the target marketing period, wherein the predicted price may be actively sent by the execution subject.

[0206] If this embodiment and Figure 8 The execution entities of the solutions provided by the corresponding embodiments are the same, so the execution entity of the solution provided by this embodiment can locally obtain the predicted price corresponding to the target product in the target marketing period.

[0207] In step 1102, the prompt information may be a pre-set general prompt information. As an example, the content of the prompt information may include, for example, "The predicted price of the ** product is too low, and it is recommended to modify the marketing activity combination plan." Therefore, the pre-set prompt information may be obtained. Optionally, a preset prompt information generation algorithm may be used to generate prompt information based on the obtained predicted price.

[0208] In step 1103, if the execution subject of the solution provided in this embodiment belongs to the client, the client may be the client used by the merchant to which the target product belongs, and thus the execution subject may directly display the prompt information to the merchant. If the execution subject of the solution provided in this embodiment belongs to the server, the execution subject may send the prompt information to the client used by the merchant, and the client may display the prompt information to the merchant.

[0209] The marketing activity processing method provided in this embodiment obtains the predicted price corresponding to the target product in the target marketing period, and then determines prompt information for prompting the need to modify the marketing activity combination plan in response to the predicted price exceeding the price threshold corresponding to the target product, and then displays the prompt information to the merchant to which the target product belongs, so as to prompt the merchant to modify the marketing activity combination plan in time to prevent the transaction price from being too low due to incorrect or multiple marketing activities, thereby causing financial losses and personnel risks.

[0210] Continue to view Fig.12, which shows a process 1200 of an embodiment of a product recommendation method. The execution subject of the method may be a server having a product recommendation function. Specifically, the execution subject of the method may be a server of the e-commerce platform as described above. The method includes the following steps:

[0211] Step 1201, obtaining the predicted price of at least one target commodity in a target marketing period;

[0212] Step 1202, based on the predicted price, select a target product that meets the recommendation condition from at least one target product;

[0213] Step 1203: Push the selected target product to the consumer user.

[0214] In this embodiment, the predicted hand price can be adopted Figure 8 The corresponding embodiment provides a method for predicting the commodity price at the time of purchase that predicts the price at the time of purchase.

[0215] For the explanation of the target products and target marketing periods in this embodiment, please refer to Figure 8 The relevant descriptions in the corresponding embodiments will not be repeated here.

[0216] Specifically, in step 1201, at least one target product may come from the same merchant or from different merchants, which is not specifically limited here. When the predicted prices corresponding to the at least one target product in the target marketing period are pre-stored locally in the execution subject, the execution subject can obtain the predicted prices locally. When the execution subject does not store the predicted prices, for example, it can obtain Figure 8 The executor of the solution provided by the corresponding embodiment obtains the predicted price.

[0217] In step 1202, the recommendation condition is pre-set. The target product may have a corresponding price threshold. The price threshold may be set by the merchant to which the target product belongs. The recommendation condition may include, for example: the predicted price does not exceed the corresponding price threshold.

[0218] Optionally, the recommendation condition may include, for example: the predicted price does not exceed the price threshold, and the difference between the predicted price and the price threshold is less than a preset value. When any target product meets the recommendation condition, it can be represented that the predicted price of the target product is close to the corresponding price threshold, and the target product has a relatively large discount.

[0219] It should be understood that the recommendation conditions can be designed according to actual needs, and the content of the recommendation conditions is not specifically limited here.

[0220] In step 1203, the selected target product may be pushed to the consumer user. It should be understood that the consumer user here may be all the consumer users of the e-commerce platform, or some of the consumer users. As an implementation method, when the consumer users here are some of the consumer users of the e-commerce platform, some of the consumer users may be randomly selected from the consumer user set of the e-commerce platform.

[0221] Optionally, based on the historical behavior data of the consumer users in the consumer user set and the product information of the selected target product, a consumer user who prefers the selected target product can be selected from the consumer user set. Then, the target product preferred by the selected consumer user is pushed to the selected consumer user. Specifically, the relevant information of the target product can be sent to the client used by the consumer user so that the client displays the relevant information. The relevant information may include product attribute information of the target product. Furthermore, the relevant information may include information such as the predicted price of the target product corresponding to the target historical marketing period while including the product attribute information.

[0222] The commodity recommendation method provided in this embodiment can achieve targeted information push by obtaining the predicted price of at least one target commodity corresponding to the target marketing period, and then selecting the target commodity that meets the recommendation condition from the at least one target commodity based on the predicted price, and then pushing the selected target commodity to the consumer user. Through this solution, it is possible to push commodities with relatively large discounts to consumer users, which can save users' search costs for commodities with relatively large discounts, improve user experience, and improve user conversion rate.

[0223] Further references Fig.13 As an implementation of the methods shown in some of the above figures, this specification provides an embodiment of a training sample generation device, which is similar to Figure 2 Corresponding to the method embodiment shown, the device can be applied to Figure 1 Device 101 is shown.

[0224] like Fig.13As shown, the training sample generating device 1300 of this embodiment may include: an acquisition unit 1301, a determination unit 1302 and a generation unit 1303. The acquisition unit 1301 is configured to acquire the commodity marketing features corresponding to the commodity in the historical marketing period, and the commodity marketing features include the commodity attribute features of the commodity and the marketing activity features corresponding to the marketing activities participated in by the commodity in the historical marketing period; the determination unit 1302 is configured to determine the corresponding price tag of the commodity in the historical marketing period according to the sales data of the commodity in the historical marketing period; the generation unit 1303 is configured to generate training samples marked with the commodity marketing features and the price tag, and the training samples are used to train the prediction model for predicting the price of the commodity.

[0225] In this embodiment, the specific processing of the acquisition unit 1301, the determination unit 1302 and the generation unit 1303 and the technical effects thereof can be referred to in Figure 2 The relevant descriptions of step 201, step 202 and step 203 in the corresponding embodiment are not repeated here.

[0226] In some optional implementations of this embodiment, the marketing activities that the product participated in during the historical marketing period include the marketing activities set by the merchant to which the product belongs.

[0227] In some optional implementations of the present embodiment, the acquisition unit 1301 may include: a first acquisition sub-unit (not shown in the figure), configured to acquire marketing activity information corresponding to each marketing activity in which the product participated in a historical marketing period; a feature extraction sub-unit (not shown in the figure), configured to perform feature extraction on the marketing activity information to obtain marketing activity features; a second acquisition sub-unit (not shown in the figure), configured to acquire product attribute features of the product; and a generation sub-unit (not shown in the figure), configured to combine product attribute features and marketing activity features into product marketing features.

[0228] In some optional implementations of the present embodiment, the marketing activities have corresponding activity categories; and the feature extraction subunit may include: a selection module (not shown in the figure), configured to select target marketing activities that participate in generating marketing activity features from various marketing activities according to the activity categories to which the various marketing activities belong; a feature extraction module (not shown in the figure), configured to perform feature extraction on the marketing activity information corresponding to the selected target marketing activities, to obtain marketing activity features.

[0229] In some optional implementations of the present embodiment, the selection module is further configured to: for each activity category in the activity category to which each marketing activity belongs, perform the following marketing activity selection operation: if the marketing activities under the activity category are set with corresponding priorities, then based on the priorities, the marketing activities are selected as the target marketing activities from the marketing activities belonging to the activity category in various marketing activities; if the marketing activities under the activity category are not set with corresponding priorities, then based on the discount amount, the marketing activities are selected as the target marketing activities from the marketing activities belonging to the activity category in various marketing activities.

[0230] In some optional implementations of the present embodiment, the feature extraction module may include: a determination submodule (not shown in the figure), configured to determine a base price corresponding to the commodity; a conversion submodule (not shown in the figure), configured to convert the marketing activity information corresponding to each target marketing activity into corresponding numerical pairs according to the base price, the numerical pairs including a threshold amount and a discount amount; a feature processing submodule (not shown in the figure), configured to perform feature processing on each numerical pair obtained by the conversion to obtain marketing activity features.

[0231] In some optional implementations of this embodiment, the feature processing submodule may be further configured to: perform division operations, subtraction operations, binary feature addition, and normalization on each pair of numerical values ​​obtained by conversion.

[0232] In some optional implementations of this embodiment, the activity category may include at least one of the following: direct price reduction for single products, full discount, store coupons, brand coupons, and category coupons.

[0233] In some optional implementations of this embodiment, the determination submodule may be further configured to: if there is a single product direct reduction category in the activity category to which each marketing activity belongs, then the base price is determined based on the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category.

[0234] In some optional implementations of this embodiment, the determination submodule may be further configured to: if there is no single product direct reduction category in the activity category to which each marketing activity belongs, the original price of the product is determined as the base price.

[0235] In some optional implementations of this embodiment, the sales data may include multiple final prices generated by historical orders of the product within a predetermined historical marketing period, and the number of sales corresponding to the multiple final prices; and the determination unit 1302 may be further configured to: take a weighted average of the multiple final prices and the corresponding number of sales to obtain an average final price; normalize the average final price, and use the normalized value as the final price label.

[0236] The training sample generation device provided in this embodiment obtains the commodity marketing features corresponding to the commodity in the historical marketing period through the acquisition unit, and then determines the price tag corresponding to the commodity in the historical marketing period according to the sales data of the commodity in the historical marketing period through the determination unit, and then generates training samples marked with the marketing features of the commodity and the price tag through the generation unit, so as to achieve targeted information generation. The generated training samples can be used to characterize the relationship between the superposition of marketing activities at the commodity granularity and the high-frequency price. In addition, the generated training samples can be used to train a prediction model for predicting the price of the commodity. The prediction accuracy can be improved by predicting the price of the commodity using the prediction model. In addition, after the prediction model is deployed online, the prediction results of the prediction model can be used in risk control scenarios, which helps the risk control system to achieve accurate contact of risk warnings, thereby improving the adoption rate of the risk control system, and preventing the transaction price from being too low due to missetting or multiple marketing activities, which may cause asset losses and personnel risks.

[0237] Further references Fig.14 As an implementation of the methods shown in some of the above figures, this specification provides an embodiment of a model training device, which is similar to Figure 6 Corresponding to the method embodiment shown, the device can be applied to Figure 1 Device 102 is shown.

[0238] like Fig.14 As shown, the model training device 1400 of this embodiment may include: an acquisition unit 1401 and a training unit 1402. The acquisition unit 1401 is configured to acquire a training sample set, wherein the training sample is marked with a commodity marketing feature and a price tag, the commodity marketing feature includes the commodity attribute feature of the corresponding commodity, and the marketing activity feature corresponding to the marketing activity in which the commodity participates in the same historical marketing period, and the price tag is determined according to the sales data of the commodity in the historical marketing period; the training unit 1402 is configured to use the commodity marketing feature marked by the training sample in the training sample set as input, and the price tag corresponding to the input commodity marketing feature as output, to train the machine learning model to be trained, and obtain a prediction model for predicting the price of the commodity.

[0239] In this embodiment, the specific processing of the acquisition unit 1401 and the training unit 1402 and the technical effects thereof can be referred to in Figure 6 The relevant descriptions of step 601 and step 602 in the corresponding embodiment are not repeated here.

[0240] In some optional implementations of this embodiment, the machine learning model may be a tree model or a neural network.

[0241] In some optional implementations of this embodiment, the tree model may include an extreme gradient boosting XGBoost model; the neural network may include a Wide-Deep model that combines a linear model and a deep model.

[0242] The model training device provided in this embodiment obtains the training sample set as described above through the acquisition unit, and trains the machine learning model to be trained according to the training sample set through the training unit, so that the machine learning model can learn the degree of influence of different marketing activities on the goods, and then the prediction model obtained by training for predicting the price of the goods can predict the price with high accuracy. In addition, after the prediction model is deployed online, the prediction results of the prediction model can be used in risk control scenarios, which helps the risk control system to achieve accurate risk warning, thereby improving the adoption rate of the risk control system and preventing the transaction price from being too low due to wrong or multiple marketing activities, which may cause financial losses and personnel risks.

[0243] Further references Fig.15 As an implementation of the methods shown in some of the above figures, this specification provides an embodiment of a commodity price prediction device, which is similar to Figure 8 Corresponding to the method embodiment shown, the device can be applied to Figure 1 Device 103 is shown.

[0244] like Fig.15 As shown, the commodity price prediction device 1500 of this embodiment may include: a first acquisition unit 1501, a feature extraction unit 1502, a second acquisition unit 1503 and a prediction unit 1504. The first acquisition unit 1501 is configured to acquire marketing activity information corresponding to the marketing activities that the target commodity participates in during the target marketing period; the feature extraction unit 1502 is configured to extract features from the marketing activity information to obtain marketing activity features; the second acquisition unit 1503 is configured to acquire commodity attribute features of the target commodity; the prediction unit 1504 is configured to combine the commodity attribute features and the marketing activity features into commodity marketing features, and input the commodity marketing features into a method such as Figure 6 The prediction model trained by the method described in the corresponding embodiment obtains the corresponding price of the target product in the target marketing period.

[0245] In this embodiment, the specific processing of the first acquisition unit 1501, the feature extraction unit 1502, the second acquisition unit 1503 and the prediction unit 1504 and the technical effects thereof can be referred to in Figure 8 The relevant descriptions of step 801, step 802, step 803 and step 804 in the corresponding embodiment are not repeated here.

[0246] In some optional implementations of this embodiment, the marketing activities that the target commodity participates in during the target marketing period may include marketing activities set by the merchant to which the target commodity belongs.

[0247] In some optional implementations of the present embodiment, the marketing activities have corresponding activity categories; and the feature extraction unit 1502 may include: a selection subunit (not shown in the figure), configured to select target marketing activities that participate in generating marketing activity features from the above-mentioned participating marketing activities according to the activity categories to which the above-mentioned participating marketing activities respectively belong; a feature extraction subunit (not shown in the figure), configured to perform feature extraction on the marketing activity information corresponding to each selected target marketing activity, to obtain marketing activity features.

[0248] In some optional implementations of the present embodiment, the selection sub-unit may be further configured to: for each of the activity categories to which the above-mentioned participating marketing activities respectively belong, perform the following marketing activity selection operation: if the marketing activities under the activity category are set with corresponding priorities, then based on the priorities, the marketing activities are selected as the target marketing activities from the marketing activities belonging to the activity category among the above-mentioned participating marketing activities; if the marketing activities under the activity category are not set with corresponding priorities, then based on the discount amount, the marketing activities are selected as the target marketing activities from the marketing activities belonging to the activity category among the above-mentioned participating marketing activities.

[0249] In some optional implementations of the present embodiment, the feature extraction subunit may include: a determination module (not shown in the figure), configured to determine a base price corresponding to the target product; a conversion module (not shown in the figure), configured to convert the marketing activity information corresponding to each target marketing activity into corresponding numerical pairs according to the base price, the numerical pairs including a threshold amount and a discount amount; a feature processing module (not shown in the figure), configured to perform feature processing on each numerical pair obtained by the conversion to obtain marketing activity features.

[0250] In some optional implementations of this embodiment, the feature processing module may be further configured to: perform division operations, subtraction operations, binary feature addition, and normalization on each pair of numerical values ​​obtained by conversion.

[0251] In some optional implementations of this embodiment, the activity category may include at least one of the following: direct price reduction for single products, full discount, store coupons, brand coupons, and category coupons.

[0252] In some optional implementations of this embodiment, the determination module can be further configured to: if there is a single product direct reduction category in the activity categories to which the above-mentioned participating marketing activities respectively belong, then the base price is determined according to the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category.

[0253] In some optional implementations of this embodiment, the determination module may be further configured to: if there is no single product direct reduction category in the activity categories to which the above-mentioned participating marketing activities respectively belong, the original price of the target product is determined as the base price.

[0254] The commodity price prediction device provided in this embodiment can predict the price with high accuracy by using the above prediction model. By subsequently transmitting the price to the risk control system, the risk control system can accurately reach the risk warning according to the price, thereby improving the adoption rate of the risk control system and preventing the transaction price from being too low due to wrong or multiple marketing activities, which may cause financial losses and personnel risks.

[0255] Further references Fig.16 As an implementation of the methods shown in some of the above figures, this specification provides an embodiment of a commodity processing device, which is similar to Fig.10 The device can be applied to the wind control system as described above.

[0256] like Fig.16 As shown, the commodity processing device 1600 of this embodiment may include an acquisition unit 1601 and an early warning unit 1602. The acquisition unit 1601 is configured to acquire the predicted price of the target commodity in the target marketing period. The predicted price is based on Figure 8 The method described in the corresponding embodiment predicts the final price; the warning unit 1602 is configured to execute a warning operation for the merchant to which the target product belongs if the predicted final price exceeds the final price threshold corresponding to the target product.

[0257] In this embodiment, the specific processing of the acquisition unit 1601 and the early warning unit 1602 and the technical effects thereof can be referred to in Fig.10 The relevant descriptions of step 1001 and step 1002 in the corresponding embodiment are not repeated here.

[0258] The commodity processing device provided in this embodiment obtains the predicted price corresponding to the target commodity in the target marketing period through the acquisition unit, and responds to the predicted price exceeding the price threshold corresponding to the target commodity through the early warning unit. For the merchant to which the target commodity belongs, an early warning operation is executed, thereby achieving accurate contact of risk early warning, thereby improving the adoption rate of the risk control system and preventing capital losses and personnel risks caused by too low transaction prices due to incorrect or multiple marketing activities.

[0259] Further references Fig.17 As an implementation of the methods shown in some of the above figures, this specification provides an embodiment of a marketing activity processing device, and the device embodiment is Fig.11 The device can be applied to the wind control system as described above, or Figure 8 The execution entity of the commodity price prediction method provided in the corresponding embodiment is not specifically limited here.

[0260] like Fig.17 As shown, the marketing activity processing device 1700 of this embodiment may include an acquisition unit 1701, a determination unit 1702, and an output unit 1703. The acquisition unit 1701 is configured to acquire the predicted price of the target product in the target marketing period. The predicted price is calculated using the following formula: Figure 8 The method described in the corresponding embodiment predicts the final price; the determination unit 1702 is configured to determine the prompt information for prompting the need to modify the marketing activity combination plan if the predicted final price exceeds the final price threshold corresponding to the target product; the output unit 1703 is configured to display the prompt information to the merchant to which the target product belongs.

[0261] In this embodiment, the specific processing of the acquisition unit 1701, the determination unit 1702 and the output unit 1703 and the technical effects thereof can be referred to respectively. Fig.11 The relevant descriptions of step 1101, step 1102 and step 1103 in the corresponding embodiment are not repeated here.

[0262] The marketing activity processing device provided in this embodiment obtains the predicted price corresponding to the target commodity in the target marketing period through the acquisition unit, determines the prompt information for prompting the need to modify the marketing activity combination plan in response to the predicted price exceeding the price threshold corresponding to the target commodity through the determination unit, and displays the prompt information to the merchant to which the target commodity belongs through the output unit, so as to prompt the merchant to modify the marketing activity combination plan in time to prevent the asset loss and personnel risk caused by the transaction price being too low due to the wrong or multiple marketing activities.

[0263] Further references Fig.18 As an implementation of the methods shown in some of the above figures, this specification provides an embodiment of a product recommendation device. Fig.12 The method embodiment shown corresponds to the embodiment shown in the figure. The device can be applied to a server with a product recommendation function. Specifically, the device can be applied to the e-commerce platform as described above.

[0264] like Fig.18 As shown, the commodity recommendation device 1800 of this embodiment may include an acquisition unit 1801, a selection unit 1802, and a push unit 1803. The acquisition unit 1801 is configured to acquire the predicted price corresponding to at least one target commodity in the target marketing period. The predicted price is calculated using the following formula: Figure 8The method described in the corresponding embodiment predicts the final price; the selection unit 1802 is configured to select a target product that meets the recommendation condition from the at least one target product based on the predicted final price; the push unit 1803 is configured to push the selected target product to the consumer user.

[0265] In this embodiment, the specific processing of the acquisition unit 1801, the selection unit 1802 and the pushing unit 1803 and the technical effects thereof can be referred to respectively. Fig.12 The relevant descriptions of step 1201, step 1202 and step 1203 in the corresponding embodiment are not repeated here.

[0266] The commodity recommendation device provided in this embodiment can achieve targeted information push by obtaining the predicted price of at least one target commodity in the target marketing period through the acquisition unit, selecting the target commodity that meets the recommendation condition from the at least one target commodity based on the predicted price through the selection unit, and pushing the selected target commodity to the consumer user through the push unit. Through this solution, it is possible to push commodities with relatively large discounts to consumer users, which can save the user's search cost for commodities with relatively large discounts, improve user experience, and improve user conversion rate.

[0267] The embodiments of this specification also provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed in a computer, the computer is caused to execute the methods respectively shown in the above method embodiments.

[0268] The embodiments of the present specification also provide a computing device, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the methods shown in the above method embodiments are implemented.

[0269] The embodiments of this specification also provide a computer program product, which, when executed on a data processing device, enables the data processing device to implement the methods respectively shown in the above method embodiments.

[0270] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the multiple embodiments disclosed in this specification can be implemented by hardware, software, firmware, or any combination thereof. When implemented by software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.

[0271] In some cases, the actions or steps described in the claims may be performed in a different order than in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0272] The specific implementation methods described above further illustrate in detail the purposes, technical solutions and beneficial effects of the multiple embodiments disclosed in this specification. It should be understood that the above description is only the specific implementation methods of the multiple embodiments disclosed in this specification, and is not used to limit the protection scope of the multiple embodiments disclosed in this specification. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the multiple embodiments disclosed in this specification should be included in the protection scope of the multiple embodiments disclosed in this specification.

Claims

1. A training sample generation method, comprising: Obtaining commodity marketing features corresponding to a commodity in a historical marketing period, the commodity marketing features including commodity attribute features of the commodity and marketing activity features corresponding to multiple marketing activities participated in by the commodity in the historical marketing period, the marketing activity features being obtained by performing feature processing on value pairs obtained by converting marketing activity information corresponding to the multiple marketing activities, respectively, the value pairs including a threshold amount and a discount amount; Determine, based on the sales data of the commodity in the historical marketing period, the corresponding price tag of the commodity in the historical marketing period; A training sample marked with the commodity marketing feature and the final price tag is generated, and the training sample is used to train a prediction model for predicting the final price of the commodity.

2. The method according to claim 1, wherein: The marketing activities that the commodity participated in during the historical marketing period include the marketing activities set by the merchant to which the commodity belongs.

3. The method according to claim 1, wherein: The obtaining of the commodity marketing characteristics corresponding to the commodity in the historical marketing period includes: Obtaining marketing activity information corresponding to each marketing activity in which the product participated in the historical marketing period; Extracting features from the marketing activity information to obtain features of the marketing activity; Obtaining commodity attribute characteristics of the commodity; The commodity attribute characteristics and the marketing activity characteristics are combined into the commodity marketing characteristics.

4. The method according to claim 3, wherein: Marketing activities have corresponding activity categories; as well as The extracting features of the marketing activity information to obtain the marketing activity features includes: According to the activity categories to which the various marketing activities respectively belong, selecting from the various marketing activities a target marketing activity that participates in generating the marketing activity features; Feature extraction is performed on the marketing activity information corresponding to each selected target marketing activity to obtain the marketing activity feature.

5. The method according to claim 4, wherein: The step of selecting target marketing activities that participate in generating the marketing activity features from the various marketing activities according to the activity categories to which the various marketing activities respectively belong includes: For each of the activity categories to which each marketing activity belongs, the following marketing activity selection operation is performed: If the marketing activities under the activity category are set with corresponding priorities, then according to the priorities, a marketing activity is selected as a target marketing activity from the marketing activities belonging to the activity category among the various marketing activities; If the marketing activities under the activity category have not been set with corresponding priorities, then according to the discount amount, a marketing activity is selected as the target marketing activity from the marketing activities belonging to the activity category among the various marketing activities.

6. The method according to claim 4, wherein: The feature extraction of the marketing activity information corresponding to each selected target marketing activity to obtain the marketing activity features includes: Determine the base price corresponding to the commodity; According to the base price, converting the marketing activity information corresponding to each target marketing activity into a corresponding value pair, wherein the value pair includes a threshold amount and a discount amount; Perform feature processing on each value pair obtained by conversion to obtain the marketing activity feature.

7. The method according to claim 6, wherein: The feature processing of each value pair obtained by conversion includes: The converted values ​​are divided, subtracted, binary features are added, and normalized.

8. The method according to claim 6, wherein: Activity categories include at least one of the following: direct price reduction for single products, discounts on purchases over a certain amount, store coupons, brand coupons, and category coupons.

9. The method according to claim 8, wherein: The determining of the base price corresponding to the commodity includes: If there is a single product direct reduction category in the activity categories to which the various marketing activities respectively belong, the base price is determined according to the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category.

10. The method according to claim 9, wherein: The determining of the base price corresponding to the commodity further includes: If there is no single product direct price reduction category in the activity categories to which the various marketing activities belong, the original price of the product will be determined as the base price.

11. The method according to any one of claims 1 to 10, wherein: The sales data includes multiple prices of the product generated by historical orders during the historical marketing period, and the number of units sold corresponding to the multiple prices; as well as The determining, based on the sales data of the commodity in the historical marketing period, the corresponding price tag of the commodity in the historical marketing period includes: Taking a weighted average of the multiple prices and the corresponding sales quantities to obtain an average price; The average price is normalized, and the normalized value is used as the price tag.

12. A model training method, comprising: A training sample set is obtained, wherein the training samples are marked with commodity marketing features and price tags, wherein the commodity marketing features include commodity attribute features of the corresponding commodity and marketing activity features corresponding to multiple marketing activities in which the commodity participates in the same historical marketing period, wherein the marketing activity features are obtained by performing feature processing on numerical pairs obtained by converting marketing activity information corresponding to the multiple marketing activities, wherein the numerical pairs include a threshold amount and a discount amount; and the price tag is determined according to sales data of the commodity in the historical marketing period; The commodity marketing features marked by the training samples in the training sample set are taken as input, and the price tag corresponding to the input commodity marketing features is taken as output, and the machine learning model to be trained is trained to obtain a prediction model for predicting the price of the commodity.

13. The method according to claim 12, wherein: The machine learning model is a tree model or a neural network.

14. The method according to claim 13, wherein: The tree model includes an extreme gradient boosting model; The neural network includes a model that combines a linear model and a deep model.

15. A method for predicting the price of a commodity, comprising: Obtaining marketing activity information corresponding to each marketing activity that the target product participates in during the target marketing period; Extracting features from the marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; The commodity attribute characteristics and the marketing activity characteristics are combined into commodity marketing characteristics, and the commodity marketing characteristics are input into a prediction model trained by the method as claimed in claim 12 to obtain the corresponding price of the target commodity in the target marketing period.

16. The method according to claim 15, wherein: The marketing activities in which the target commodity participates during the target marketing period include marketing activities set by the merchant to which the target commodity belongs.

17. The method according to claim 15, wherein: Marketing activities have corresponding activity categories; as well as The extracting features of the marketing activity information to obtain marketing activity features includes: According to the activity categories to which the participating marketing activities respectively belong, selecting a target marketing activity that participates in generating the marketing activity features from the participating marketing activities; Feature extraction is performed on the marketing activity information corresponding to each selected target marketing activity to obtain the marketing activity feature.

18. The method according to claim 17, wherein: The step of selecting target marketing activities that participate in generating the marketing activity features from the participating marketing activities according to the activity categories to which the participating marketing activities belong respectively includes: For each of the activity categories to which the participating marketing activities belong, the following marketing activity selection operation is performed: If the marketing activities under the activity category are set with corresponding priorities, then according to the priorities, a marketing activity is selected as a target marketing activity from the marketing activities belonging to the activity category among the participating marketing activities; If the marketing activities under the activity category have not been set with corresponding priorities, then according to the discount amount, a marketing activity is selected as the target marketing activity from the marketing activities belonging to the activity category among the participating marketing activities.

19. The method according to claim 17, wherein: The feature extraction of the marketing activity information corresponding to each selected target marketing activity to obtain the marketing activity features includes: Determine a base price corresponding to the target product; According to the base price, converting the marketing activity information corresponding to each target marketing activity into a corresponding value pair, wherein the value pair includes a threshold amount and a discount amount; Perform feature processing on each value pair obtained by conversion to obtain the marketing activity feature.

20. The method according to claim 19, wherein: The feature processing of each value pair obtained by conversion includes: Perform division operations, subtraction operations, binary feature addition, and normalization on each pair of values ​​obtained from the conversion.

21. The method according to claim 19, wherein: Activity categories include at least one of the following: direct price reduction for single products, discounts on purchases over a certain amount, store coupons, brand coupons, and category coupons.

22. The method according to claim 21, wherein: The determining of the base price corresponding to the target product includes: If there is a single product direct reduction category in the activity categories to which the participating marketing activities respectively belong, the base price is determined according to the reduced price in the marketing activity information corresponding to the target marketing activity under the single product direct reduction category.

23. The method according to claim 22, wherein: The determining of the base price corresponding to the target product also includes: If there is no single product direct price reduction category in the activity categories to which the participating marketing activities respectively belong, the original price of the target product is determined as the base price.

24. A commodity processing method, comprising: Obtain the predicted price of the target product during the target marketing period. The predicted price is the price predicted by the following method: Obtaining marketing activity information corresponding to each of the marketing activities that the target product participates in during the target marketing period; Extracting features from the marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; The commodity attribute features and the marketing activity features are combined into commodity marketing features, and the commodity marketing features are input into a prediction model trained by the method according to claim 12 to obtain the corresponding price of the target commodity in the target marketing period; If the predicted final price exceeds the final price threshold corresponding to the target product, an early warning operation is performed for the merchant to which the target product belongs.

25. A marketing activity processing method, comprising: Obtain the predicted price of the target product during the target marketing period. The predicted price is the price predicted by the following method: Obtaining marketing activity information corresponding to each of the marketing activities that the target product participates in during the target marketing period; Extracting features from the marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; The commodity attribute features and the marketing activity features are combined into commodity marketing features, and the commodity marketing features are input into a prediction model trained by the method according to claim 12 to obtain the corresponding price of the target commodity in the target marketing period; If the predicted final price exceeds the final price threshold corresponding to the target product, a prompt message is determined to prompt that the marketing activity combination plan needs to be modified; The prompt information is displayed to the merchant to which the target product belongs.

26. A product recommendation method, comprising: Obtain the predicted price of at least one target product in a target marketing period. The predicted price of each target product in the target marketing period is the price predicted in the following way: Obtaining marketing activity information corresponding to each of the marketing activities that the target product participates in during the target marketing period; Extract features from marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; Combining the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics, and inputting the commodity marketing characteristics into the prediction model trained by the method according to claim 12 to obtain the corresponding price of the target commodity in the target marketing period; Based on the predicted price, selecting a target product that meets a recommendation condition from the at least one target product; Push selected target products to consumer users.

27. A training sample generating device, comprising: an acquisition unit configured to acquire commodity marketing features corresponding to a commodity in a historical marketing period, the commodity marketing features including commodity attribute features of the commodity and marketing activity features corresponding to a plurality of marketing activities in which the commodity participated in the historical marketing period, the marketing activity features being obtained by performing feature processing on a numerical value pair obtained by converting marketing activity information corresponding to the plurality of marketing activities, respectively, and the numerical value pair including a threshold amount and a discount amount; A determination unit is configured to determine a price tag corresponding to the commodity in the historical marketing period according to the sales data of the commodity in the historical marketing period; The generating unit is configured to generate training samples marked with the commodity marketing features and the final price tag, wherein the training samples are used to train a prediction model for predicting the final price of the commodity.

28. A model training device, comprising: The acquisition unit is configured to acquire a training sample set, wherein the training samples are marked with commodity marketing features and a price tag, wherein the commodity marketing features include commodity attribute features of the corresponding commodity and marketing activity features corresponding to multiple marketing activities in which the commodity participates in the same historical marketing period, wherein the marketing activity features are obtained by performing feature processing on numerical pairs obtained by converting marketing activity information corresponding to the multiple marketing activities, wherein the numerical pairs include a threshold amount and a discount amount; and the price tag is determined according to sales data of the commodity in the historical marketing period; The training unit is configured to take the commodity marketing features marked by the training samples in the training sample set as input, take the price tag corresponding to the input commodity marketing features as output, train the machine learning model to be trained, and obtain a prediction model for predicting the price of the commodity.

29. A commodity price prediction device, comprising: A first acquisition unit is configured to acquire marketing activity information corresponding to each marketing activity in which the target product participates in a target marketing period; a feature extraction unit configured to extract features from the marketing activity information to obtain marketing activity features; A second acquisition unit is configured to acquire commodity attribute characteristics of the target commodity; The prediction unit is configured to combine the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics, and input the commodity marketing characteristics into a prediction model trained using the method as claimed in claim 12 to obtain the corresponding price of the target commodity in the target marketing period.

30. A commodity processing device, comprising: The acquisition unit is configured to acquire a predicted price corresponding to the target commodity in the target marketing period, wherein the predicted price is a price predicted by: Obtaining marketing activity information corresponding to each of the marketing activities that the target product participates in during the target marketing period; Extract features from marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; Combining the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics, and inputting the commodity marketing characteristics into the prediction model trained by the method according to claim 12 to obtain the corresponding price of the target commodity in the target marketing period; The early warning unit is configured to execute an early warning operation for the merchant to which the target product belongs if the predicted price exceeds the price threshold corresponding to the target product.

31. A marketing activity processing device, comprising: The acquisition unit is configured to acquire a predicted price corresponding to the target commodity in the target marketing period, wherein the predicted price is a price predicted by: Obtaining marketing activity information corresponding to each of the marketing activities that the target product participates in during the target marketing period; Extract features from marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; Combining the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics, and inputting the commodity marketing characteristics into the prediction model trained by the method according to claim 12 to obtain the corresponding price of the target commodity in the target marketing period; a determination unit configured to determine prompt information for prompting that the marketing activity combination plan needs to be modified if the predicted price exceeds the price threshold corresponding to the target product; The output unit is configured to display the prompt information to the merchant to which the target product belongs.

32. A commodity recommendation device, comprising: The acquisition unit is configured to acquire the predicted price of at least one target commodity corresponding to the target marketing period, wherein the predicted price of each target commodity corresponding to the target marketing period is the price predicted by: Obtaining marketing activity information corresponding to each of the marketing activities that the target product participates in during the target marketing period; Extract features from marketing activity information to obtain marketing activity features; Acquire commodity attribute characteristics of the target commodity; Combining the commodity attribute characteristics and the marketing activity characteristics into commodity marketing characteristics, and inputting the commodity marketing characteristics into the prediction model trained by the method according to claim 12 to obtain the corresponding price of the target commodity in the target marketing period; a selection unit configured to select a target product that meets a recommendation condition from the at least one target product based on the predicted price; The push unit is configured to push the selected target product to the consumer user.

33. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed in a computer, the computer is caused to execute the method according to any one of claims 1 to 26.

34. A computing device comprising a memory and a processor, wherein: The memory stores executable codes, and when the processor executes the executable codes, the method according to any one of claims 1 to 26 is implemented.

Images