Commodity recommendation method and system in commercial operation and computer program product

Abstract

The invention provides a commodity recommendation method and system in commercial operation and a computer program product, which are used for providing recommendation processing of purchased commodities and selected commodities for each shop in the commercial operation, performing exposure conversion estimation of commodities capable of being put on a shelf through an obtained shop portrait, coarsely arranging the commodities capable of being put on the shelf to obtain candidate put-on-shelf commodities of the shop, and recommending the candidate put-on-shelf commodities to the shop through the candidate put-on-shelf commodities. And pre-estimating conversion values of candidate shelf commodities through shop portraits so as to obtain shop shelf commodities and putting values of the shelf commodities in shops through fine arrangement, and finally, generating a shelf commodity processing strategy in shop operation according to two types of shelf commodities including purchased commodities and selected commodities. According to the method and the system, dynamic commodity selection is realized for all users passing through the shops, the delivery of purchased commodities is accurately controlled, the method and the system do not depend on subjective cognition of shops any more, high-reliability and high-accuracy commodity recommendation in commercial operation is realized, and high passenger flow conversion is obtained for the shops.

Description

Technical Field

[0001] The present application relates to the field of computer application technology, and specifically to a commodity recommendation method and system in commercial operations, and a computer program product. Background Art

[0002] In business operations, the selection and shelving of products in stores and the display of purchased goods all rely on the merchants' subjective consumption data analysis and the merchants' subjective observations of their own customer groups, lacking highly reliable and accurate product selection and shelving processing.

[0003] In addition, the consumption data generated by merchants will also be affected by the merchants' product display and selection in the stores.

[0004] If a merchant has put on the shelves a product that a customer group is more likely to buy, but the product is displayed in a location in the store that is not easily found, the product's placement value may be low. The product's consumption data will inevitably tell the merchant that the product's placement value is low and the product should be returned or exchanged.

[0005] As for goods that are not on the shelves but can be sold in the store, merchants are unable to accurately select products suitable for their customer groups because they do not have the corresponding consumption data to support analysis.

[0006] In summary, the selection and listing of goods in existing subway commercial operations rely on and are limited by the subjective cognition of merchants. There is an urgent need to provide a highly reliable and accurate product recommendation implementation in subway commercial operations. Summary of the invention

[0007] One purpose of the present application is to abandon the product selection and listing process in commercial operations that relies on and is limited by the subjective cognition of merchants, and to provide a highly reliable and accurate product recommendation implementation in commercial operations.

[0008] According to one aspect of an embodiment of the present application, a method for recommending goods in commercial operations is disclosed, the method comprising:

[0009] By determining all users passing through a shop during business, a shop profile of the shop is obtained, wherein the shop profile includes a user profile of each user;

[0010] The exposure conversion of the products that can be put on the shelves in the store is estimated by using the store portrait, and the products that can be put on the shelves are roughly ranked to obtain candidate products for putting on the shelves;

[0011] The estimated conversion values ​​of the candidate listed products are precisely ranked by the store portrait to obtain the listed products of the store and the placement value of the listed products in the store;

[0012] A shelf merchandise processing strategy for the store operation is generated by adapting the two types of shelf merchandise, namely, purchased merchandise and selected merchandise.

[0013] According to one aspect of an embodiment of the present application, obtaining a store portrait of the store by determining all users passing through the store during business, includes:

[0014] For shops in the business, all corresponding users are obtained according to the exit corresponding to the shop at the station;

[0015] The user portraits of all users corresponding to the exit are assigned to the shop to form a shop portrait of the shop.

[0016] According to one aspect of an embodiment of the present application, the step of estimating the exposure conversion of the products that can be put on the shelves in the store through the store portrait, roughly sorting the products that can be put on the shelves, and obtaining candidate products for putting on the shelves includes:

[0017] For the products that can be put on the shelves, the conversion value of the products that can be put on the shelves when they are exposed in the stores is estimated through the store portrait;

[0018] According to the estimated conversion value, products with high conversion value are selected from the available products as candidate products for listing.

[0019] According to one aspect of the embodiment of the present application, estimating the conversion value of the available products when they are exposed in the store through the store portrait includes:

[0020] Estimate the click rate of each user on the available products among all users who pass by the store;

[0021] The click rates are accumulated to obtain the conversion value of the exposure of the available products in the shop.

[0022] According to one aspect of an embodiment of the present application, the estimated click rate of all users passing through the store on the available products includes:

[0023] For all users who pass by the shop, the user-product pairs of the shop are constructed through user portraits and product features corresponding to the products that can be put on the shelves;

[0024] By predicting the product click-through rate of the user-product pair, the click-through rate of each user for the available products among all users corresponding to the store is obtained.

[0025] According to one aspect of an embodiment of the present application, the step of finely ranking the estimated conversion values ​​of the candidate listed products through the store portrait to obtain the listed products of the store and the placement value of the listed products in the store includes:

[0026] Secondarily estimating the click rate of all users corresponding to the store for the candidate listed products to obtain a secondary estimated value of the click rate;

[0027] Adapting the high click-through rate secondary estimated value to finely sort the candidate listing products and screen the listing products of the store;

[0028] The placement value of the listed products in the shop is calculated according to the corresponding secondary estimated click rate, click value and the number of exposures of the listed products in the shop.

[0029] According to one aspect of the embodiment of the present application, the shelf commodities include purchased commodities, and the shelf commodity processing strategy for adapting the purchased commodities and the selected commodities to the store operation includes:

[0030] The purchased goods of the store are enumerated, and a shelf commodity processing strategy is generated for the purchased goods with high investment value, in which the purchased goods are displayed in a prominent position of the store; and a shelf commodity processing strategy is generated for the purchased goods with low investment value, in which the purchased goods are not to be purchased or returned.

[0031] According to one aspect of the embodiment of the present application, the shelf commodities include selected commodities, and the shelf commodities processing strategy for the store operation generated by adapting the two types of purchased commodities and selected commodities includes:

[0032] The product selections of the store are enumerated, and a product listing processing strategy for the store's replenishment needs is generated for the product selections with high investment value.

[0033] According to one aspect of an embodiment of the present application, a commodity recommendation system for commercial operations is disclosed, the system comprising a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to implement the steps of the method described above.

[0034] According to one aspect of an embodiment of the present application, a computer program product is disclosed, including a computer program, wherein the computer program is executed by a processor to perform the steps of the method described above.

[0035] The embodiment of the present application can provide each shop in the commercial operation with recommended processing of purchased goods and selected products, so as to realize dynamic product selection for all users passing through the shop, and accurately control the delivery of purchased goods, so as to obtain high customer flow conversion for the shop.

[0036] Therefore, in the embodiments of the present application, for a shop in a business, all users passing through the shop are first determined, so as to obtain the shop portrait of the shop through the user portraits of all users, and then the exposure conversion of the shelfable goods is estimated through the shop portrait. First, the shelfable goods are roughly sorted to obtain the candidate shelfable goods of the shop, and then the conversion value of the candidate shelfable goods is estimated through the shop portrait, and then the shelfable goods and the placement value of the shelfable goods in the shop are obtained through fine sorting. Finally, a shelfable goods processing strategy in the store operation is generated that is adapted to the two types of shelfable goods, namely, purchased goods and selected goods. It no longer relies on the subjective cognition of the shop merchants, and realizes product recommendations in commercial operations with high reliability and accuracy, realizes dynamic product selection for the shop, and accurately controls the placement of purchased goods by generating a shelfable goods processing strategy adapted to the purchased goods, thereby realizing high customer flow conversion for the sale of goods in the shop.

[0037] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present application.

[0038] It should be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and other objects, features and advantages of the present application will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0040] Figure 1 A flow chart of a method for recommending goods in commercial operations according to an embodiment of the present application is shown.

[0041] Figure 2 is based on Figure 1 The corresponding embodiment shows a method flow chart describing the steps of obtaining a store portrait of a store by determining all users passing through the store.

[0042] Figure 3 is based on Figure 1 The corresponding embodiment shows a method flow chart describing the steps of estimating the exposure conversion of shelf-ready commodities in shops through shop portraits, roughly sorting the shelf-ready commodities, and obtaining candidate shelf-ready commodities.

[0043] Figure 4 is based on Figure 3 A flowchart of a method for describing the steps of estimating the conversion value of shelf-ready commodities by estimating the exposure of shelf-ready commodities in shops through shop portraits shown in the corresponding embodiment.

[0044] Figure 5 is based on Figure 4A flowchart of a method for describing the steps of estimating the click rate of all users passing through a store on the available products is shown in the corresponding embodiment.

[0045] Figure 6 is based on Figure 1 The corresponding embodiment shows a method flow chart describing the steps of obtaining the store's listed products and the placement value of the listed products in the store by accurately sorting the estimated conversion values ​​of the candidate listed products through the store portrait. DETAILED DESCRIPTION

[0046] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present application will be more comprehensive and complete and the concepts of the example embodiments will be fully conveyed to those skilled in the art. The accompanying drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0047] In addition, the described features, structures or characteristics may be combined in one or more example embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the example embodiments of the present application. However, those skilled in the art will appreciate that the technical solution of the present application may be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. may be adopted. In other cases, known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring the present application and making the various aspects of the present application obscure.

[0048] Some of the blocks shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0049] There are many routes in a city, and there are many stations on each route. There are many shops inside and outside each station based on business operations. Each shop has its own selection and display characteristics. For example, the selection of optional goods is often carried out with the help of the merchant's own subjective cognition. For example, the display location of the purchased goods is often determined with the help of the merchant's own subjective cognition.

[0050] In commercial operations, each shop is limited by subjective experience and relies heavily on subjective experience to determine what kind of goods to put in the shop and how to display the goods in the shop. There is no objective, accurate and reliable merchant product selection and shelving process, which leads to poor product placement effects in various shops in commercial operations.

[0051] Based on this, the embodiment of the present application provides a method for recommending goods in commercial operations, thereby serving commercial operations with high reliability and accuracy, so that shops in the business can achieve good performance.

[0052] See also Figure 1 , Figure 1 A flowchart of a method for recommending goods in a business operation according to an embodiment of the present application is shown. A method for recommending goods in a business operation provided by an embodiment of the present application includes:

[0053] Step S110, obtaining a shop portrait of the shop by determining all users passing through the shop during business, the shop portrait including a user portrait of each user;

[0054] Step S120, using the store portrait to estimate the exposure conversion of the products that can be put on the shelves in the store, roughly sorting the products that can be put on the shelves, and obtaining candidate products for putting on the shelves;

[0055] Step S130, using the store portrait to accurately sort the estimated conversion values ​​of the candidate listed products to obtain the store's listed products and the placement value of the listed products in the store;

[0056] Step S140, generating a shelf commodity processing strategy for store operation based on two types of shelf commodities, namely, purchased commodities and selected commodities.

[0057] These steps are described in detail below.

[0058] Through the execution of step S110, firstly, all users corresponding to the shop are obtained, and all users corresponding to the shop are all users passing through the shop in the business. For each shop in the business, the optional goods will be recommended based on all its corresponding users, and the recommended goods that have been purchased will be put on the shelves for display, so as to build a recommendation system for merchants to select goods for display on the shelves for the customer flow in the business. On the one hand, it can realize dynamic product selection for the shop and improve the conversion of customer flow. On the other hand, it can also enable the shops in the business to obtain good business, thereby assisting or even improving the shop leasing business of the owner.

[0059] Shops can be distributed inside the station, or they can be shops located in locations radiating outward from the station’s exits. For example, shops located in the station hall, shops passing by from the gate to each exit, and shops at the station’s exits.

[0060] Therefore, for many shops inside and outside the site, product recommendations for each shop can be implemented based on the embodiments of the present application, and the implemented product recommendations include but are not limited to selection recommendations for optional products and recommendations for display and shelving of purchased products.

[0061] In the product recommendations of each store, suitable products for the store are recommended through all users corresponding to the location and these users' estimated value conversion of the products exposed in the store, or the best products are selected for the store to be put on the shelves.

[0062] Determining all users who pass through a shop during business is the execution process of obtaining all users corresponding to the location of the shop. The obtained users are traveling and their travel trajectories are related to the shop, that is, the user travel profile of the user is related to the shop.

[0063] Specifically, the user travel portrait may include outbound travel trajectory, user itinerary, etc.; correspondingly, the user travel portrait is related to shops, which means that the outbound travel trajectory and user itinerary described in the user travel portrait can correspond to the location of the shop.

[0064] See also Figure 2 , Figure 2 is based on Figure 1 The corresponding embodiment shows a method flow chart describing the steps of obtaining a store portrait of a store by determining all users passing through the store during a business trip. The step S110 of obtaining a store portrait of a store by determining all users passing through the store during a business trip provided in the embodiment of the present application includes:

[0065] Step S111, for a shop in the business, all corresponding users are obtained according to the exit corresponding to the shop at the station;

[0066] Step S112, assigning user portraits of all users corresponding to the exit to the shop to form a shop portrait of the shop.

[0067] These two steps are described in detail below.

[0068] According to the exit corresponding to the store at the station, the exit travel trajectory and / or the user itinerary corresponding to the exit are obtained to obtain all users corresponding to the store. Exemplarily, all users whose user itinerary and / or exit travel trajectory correspond to the location of the store are obtained for product recommendation of the store.

[0069] Exemplarily, the locations corresponding to the user's itinerary and / or exit trajectory are the locations that can be covered by the user's exit behavior. For example, all the locations that the user passes through when walking from the track area to the exit elevator to the exit gate, as well as the location of the user in the exit direction of the exit gate, and even the location on the exit channel of the exit to which the user belongs, and all the shops distributed on it are the shops corresponding to the user.

[0070] Obtain all users whose user itineraries and / or outbound travel trajectories correspond to the locations of shops, which are obtained by aggregating the locations corresponding to the user itineraries and / or outbound travel trajectories.

[0071] Mapping the user's itinerary and / or travel trajectory to a location can be achieved by positioning the user. In addition, it can also be achieved by determining the user's preferred exit direction based on the distribution of the number of times the user exits each gate group at the site, especially the exit gate to which each user belongs.

[0072] At this point, it should be further explained that the execution process of the above step S111 includes: first, determining the user's exit travel trajectory by the distribution of the number of times the user exits each gate of the station gate group; then aggregating users for the shops covered by the exit trajectory to obtain all users corresponding to the shops, and the aggregated users correspond to the exit travel trajectory covering the shop.

[0073] The user's entry and exit behaviors at the gate will generate corresponding user trips, which are stored in the trip data. In an exemplary embodiment, the trip data exists in the form of a user trip table. Each user trip generated when the user exits the gate is stored in the constructed user trip table.

[0074] The user's itinerary describes the user's exit gate. In terms of data, the exit gate described in the user's itinerary exists in the form of an exit gate code, and the location of the gate can be obtained from the corresponding gate information.

[0075] The distribution of the number of exits at the station gate group is obtained for each user, so as to determine the exit direction or even the exit port that the user prefers, and then obtain the exit trajectory mapped by the preferred exit direction or even the exit port.

[0076] A station has several gate groups for entering and / or exiting the station, and for the gate groups available for exiting the station, the gate group includes a left half group of gates and a right half group of gates. It should be understood that users exiting from the left half group of gates tend to exit in one direction, while users exiting from the right half group of gates tend to exit in another direction.

[0077] According to the distribution of the number of times the user exits the station at each gate group, it can be determined whether the user prefers the left half of the gate group or the right half of the gate group, and then the preferred exit direction is determined. The user's exit gate is then determined by the preferred exit direction. This exit gate is based on the distribution of the number of times the user exits the station at each gate group. Therefore, this exit gate is the user's commonly used exit gate.

[0078] At this point, the user's exit trajectory is matched to the exit gate determined by the exit direction, and the exit trajectory including the user's passage from the gate to the exit gate is obtained.

[0079] To further illustrate, the distribution of the number of times a user exits each gate in a site gate group describes the number of times a single user exits gate 1 in the gate group, the number of times a single user exits gate 2 in the gate group, and so on, at gate n_zj in the gate group (n_zj is the number of gates in the entire gate group).

[0080] The user's preferred exit direction and exit port can be determined by counting and comparing the number of exits corresponding to the left half of the gate group and the right half of the gate group based on the distribution of the user's exit times at each gate group at the site, so as to determine the user's preferred exit direction and exit port.

[0081] In an exemplary embodiment, if the number of exits and exits of a user at a station gate group is not less than a set threshold, the user's preferred exit direction can be determined by counting and comparing the number of exits corresponding to the left half group of gates and the right half group of gates as described above, and then the exit in that exit direction can be determined as the user's frequently used exit.

[0082] In a gate group, if users exit from this gate group more frequently, they will mostly exit from the same side or even the same gate in this gate group. Therefore, the gate on this side, such as the left half of the gate group, or the right half of the gate group, or the gate on the side where the frequent exits are located, the corresponding exit direction is the one preferred by the users.

[0083] Therefore, for such users, the preferred exit direction and exit port can be determined directly by counting and comparing the distribution of their exit times at each gate group at the site, thereby reducing the calculation complexity and improving execution efficiency.

[0084] The number of times a user exits a station gate group is used to determine whether the user exits the station frequently through the current station gate group. The number of times a user exits a station gate group can be obtained based on the distribution statistics of the number of times the user exits each gate of the station gate group. In addition, it can also be obtained based on the user's user itinerary and gate information. Further, it can be obtained from the user gate group exit data constructed by the itinerary data constituted by the user's itinerary and the gate information, which is not limited here.

[0085] In another exemplary embodiment, if the number of exits of a user at a site gate group is less than a set threshold, it will be necessary to determine the user's preferred exit direction through the distribution of the user's exit times at each gate of the site gate group and the user's portrait through a binary classification model, and then determine the exit gate to which the user belongs in this exit direction.

[0086] Exemplarily, the set threshold may be 5, that is, if the number of exits and exits of a gate group at a station is not less than 5 times within the set time range, it means that an exit direction corresponding to the gate group and the exit gate in the exit direction are both consistent with the user's exit passage, thereby determining that the exit gate is the user's frequently used exit gate.

[0087] If the number of times a user exits a site gate group is less than the set threshold, it means that neither the site nor the gate group is frequently visited by users. Therefore, it is necessary to construct features based on richer data, that is, the distribution of the number of times a user exits each gate in the site gate group and the user portrait as mentioned above, and then implement predictions through a pre-built model.

[0088] To further illustrate, the step of determining the user's exit travel trajectory by the distribution of the number of exits of the user at each gate of the station gate group is specifically performed as follows, namely:

[0089] (1) Determine the exit gate to which each user belongs based on the distribution of the number of times the user exits the station at the station gate group and the user profile;

[0090] (2) The user's travel trajectory is mapped to the exit, and the exit trajectory of the user passing through the exit is obtained.

[0091] Regarding this execution process, it should be explained first that a binary classification model is pre-built to predict the user's preferred exit direction. Then, based on the distribution of the number of times the user exits the station gate group and the user portrait, the binary classification model can predict whether the user will go left or right after exiting the gate group, thereby determining the user's preferred exit based on the exit in the direction of travel.

[0092] For each user, the exit gate to which the user belongs will be determined based on the distribution of the number of exits at the station gate group and the user portrait.

[0093] The exit gate to which the user belongs indicates the travel trajectory of the user after exiting the station from the gate. In other words, by mapping the existing travel trajectory to the determined exit gate, the exit trajectory of the user passing through the exit gate can be obtained.

[0094] Based on this, each user will determine the exit they frequently use during their travels. Therefore, for product recommendations in shops, shops whose locations correspond to the users' exit trajectories will expose their products to these users. Therefore, the placement of products in shops will be targeted at these users. Product recommendations based on exit trajectories will help improve the accuracy of product placement in shops.

[0095] After the user's outbound travel trajectory is determined, the users can be aggregated for the shops covered by the outbound travel trajectory, thereby obtaining all users corresponding to the shops.

[0096] In other words, as the product recommendations in the shops are made, the users at the location of each shop will be aggregated to determine the target group of the shop.

[0097] The outbound travel track covers at least one store. In other words, several stores are located on the outbound travel track of the user, so the user's user trip can be mapped to the store on the outbound travel track.

[0098] For each shop, the shop's users are aggregated according to their outbound travel trajectories, that is, the users whose outbound travel trajectories cover this shop are aggregated together, that is, the aggregated users correspond to one shop.

[0099] It should be understood that for users, along with their inbound and outbound behaviors, corresponding user trips will be generated, and the user trips are descriptive data of the user trips. Therefore, according to the outbound travel trajectory, the users whose user trips correspond to the shop are aggregated for each shop, and the users whose user trips correspond to the shop are all the users corresponding to the shop.

[0100] Through the execution of step S112, after all users corresponding to the shops in the business are obtained, since each user has his or her own user portrait, the user portraits of all users are assigned to the shops, and the user portraits of all users can be used as the shop portraits of the shop.

[0101] For a shop, after obtaining all the corresponding users through the execution of step S110 and obtaining the corresponding shop portrait through the aggregation of all users, it can determine the candidate products to be put on the shelf from the products that can be put on the shelf through the estimated exposure conversion value of all users to the products put on the shelf in the execution of steps S120 and S130, and then determine the products to be put on the shelf from the candidate products to be put on the shelf, and quantify the value of each product put on the shelf in the shop.

[0102] That is to say, under the control of step S120 and step S130, secondary screening of the products will be implemented for the product exposure of the store, that is, product screening will be implemented under the control of the two execution processes of rough sorting and fine sorting respectively, so as to select products corresponding to the highest conversion value and delivery value to recommend them to the store for exposure.

[0103] In the execution of step S120, firstly, the user attention of the available products in the store is estimated for all users based on the store portrait. It can be understood that the higher the user attention, the higher the possibility of exposure conversion of the available products in the store.

[0104] Exemplarily, user attention can be measured and characterized by estimating the conversion value of all users for the available products. The conversion value of available products in a store refers to the degree of attention received by users passing by after a product is put on the store. The conversion value of the available products in the store is used to measure the degree to which users passing by the store would be attracted by the product if the product was put on the store. It should be understood that the higher the conversion value of the available products in the store, the more suitable it is for the product to be put on the store, and the higher the possibility that users passing by the store will be attracted by the product.

[0105] See also Figure 3 , Figure 3 is based on Figure 1 The corresponding embodiment shows a method flow chart describing the steps of estimating the exposure conversion of shelf-ready commodities in shops through shop portraits, roughly sorting the shelf-ready commodities, and obtaining candidate shelf-ready commodities.

[0106] The step S120 of estimating the exposure conversion of products that can be put on the shelves in the store through the store portrait, roughly sorting the products that can be put on the shelves, and obtaining candidate products for putting on the shelves provided in the embodiment of the present application includes:

[0107] Step S121, for the products that can be put on the shelves, the conversion value of the products that can be put on the shelves in the store is estimated through the store portrait;

[0108] Step S122 , selecting products with high conversion values ​​from the available products as candidate products for listing according to the estimated conversion values.

[0109] These two steps are described in detail below.

[0110] For a store, the conversion value of all users corresponding to the store for the available products can be represented by the expected click value of all users for this available product. The expected click value is used to describe the expected click value of all users corresponding to the store for the exposed products. By estimating the expected click value of all users for the exposed products on the store, it is determined which available product has a higher expected click value, and then the product with the highest expected click value can be placed in the store.

[0111] Furthermore, the click expectation value estimation for the available products for all users corresponding to the store is realized by a pre-trained click rate estimation model. For example, the click expectation of users on the available products is estimated by the click rate estimation model, and the click rate of each user on the available products is obtained.

[0112] The products that can be put on the shelves are products that are intended to be put on the shelves for exposure in the store. Exemplarily, the products that can be put on the shelves include selected products and / or purchased products.

[0113] It should be noted that in an exemplary embodiment, the conversion value estimation is performed for a store and a product that can be put on the shelf for all users corresponding to the store. The conversion value estimation, click rate estimation between each user and the product that can be put on the shelf, and the click rate accumulation for all users are performed to finally obtain the expected value of clicks of the product that can be put on the shelf in the store, that is, the conversion value.

[0114] In an exemplary embodiment, a click rate prediction model for estimating the click rate between users and available products is trained through user click records on exposed products on a travel application.

[0115] Specifically, a training data set is first constructed for the click-through rate prediction model. Each record of the training data set includes a user portrait of the user and a product feature constructed for a product. The product is exposed on the travel application and clicked by the user, that is, it corresponds to the user's click record on the exposed product on the travel application.

[0116] Thus, the training data set is constructed

[0117] Among them, the constructed training data set D train A single record in Right now The user portrait is composed of multiple parts of data, and the data sources include but are not limited to travel data, application tracking data, etc.

[0118] Specifically,

[0119] corresponds to the user's own travel data, and This corresponds to third-party supplementary data.

[0120]

[0121] The product features include descriptive features, statistical features, and generalized statistical features of the product. For example,

[0122] y ij ∈{0,1}, when the value is 0, it means that no click event occurs after product j is exposed to user i; when the value is 1, it means that a click event occurs after product j is exposed to user i.

[0123] As a result, travel applications have been d For each user i in the day, all n i Product generation records to form the training data set for the click rate prediction model.

[0124] Correspondingly, the click-through rate between users and available products is estimated to obtain the click-through rate of the users corresponding to the store on the available products through the trained click-through rate estimation model. The click-through rate estimation performed by the click-through rate estimation model requires the construction of a corresponding prediction data set, that is, for each user corresponding to the store, the prediction data set is constructed through the user portrait and the product features of the available products. Each record of the prediction data set is the constructed user-product pair.

[0125] The constructed prediction data set is

[0126] Each record, that is, the user product pair The feature list is the same as the training dataset. To further illustrate, the records in the prediction dataset represent products that can be put on the shelves. Recommend to each user who passes by the store portrait at time T, that is, user The possibility of high conversion value can be obtained by using the expected click value.

[0127] Each record in the prediction data set represents the recommendation of product j to all users corresponding to a store, i.e., users 1, 2, ..., n. userEach record will be used as the input of the click-through rate prediction model to obtain the click-through rate of each user corresponding to the shop for the available products. Similarly, the click-through rate of all users corresponding to the shop for the available products is obtained. Similarly, the click-through rate of each user corresponding to the shop for the available products is obtained, and finally the expected value of clicks on the available products in the shop is obtained.

[0128] In an exemplary embodiment, for the click rate prediction model, the constructed loss function is as follows:

[0129]

[0130] in, for The kth element of .

[0131] Iterate along the negative gradient direction of l(w) with respect to w Until the convergence condition is met: Wherein, γ is the learning rate, for example, it can be set to 0.001, ∈ is the preset convergence threshold, for example, it can be set to ∈=10 -5 .

[0132] As the training progresses, we converge to w * , the optimal model is

[0133] Therefore, through the execution of step S121, after the corresponding conversion value of each shelf-able product of the store is estimated through the shop portrait, several products with high conversion value can be screened out from the shelf-able products of the store as candidate shelf-able products under the action of step S122, thereby realizing the initial screening of the shelf-able products, excluding most of the products and obtaining a product set of hundreds of levels. The products in this product set are the candidate shelf-able products.

[0134] It should be understood that the constructed click rate prediction model is a rough ranking algorithm model suitable for the initial screening of products that can be put on the shelves, that is, for roughly ranking the products that can be put on the shelves.

[0135] See also Figure 4 , Figure 4 is based on Figure 3 A flowchart of a method for describing the steps of estimating the conversion value of shelf-ready commodities by estimating the exposure of shelf-ready commodities in shops through shop portraits shown in the corresponding embodiment.

[0136] The step S121 of estimating the conversion value of the available products in the store through the store portrait provided in the embodiment of the present application includes:

[0137] Step S301, estimating the click rate of each user on the available products among all users passing through the store;

[0138] Step S302, the cumulative click rate is used to obtain the conversion value of the shelf-available products being exposed in the store.

[0139] These two steps are described in detail below.

[0140] Among all users corresponding to the store, each user forms a user-product pair with an available product to estimate the click rate of the user on the available product. The click rate is used to represent the attractiveness of the available product to the user. The higher the click rate, the more attractive the available product is to the user. Conversely, the lower the click rate, the less attractive the available product is to the user.

[0141] For the user-product pair consisting of the user and the available products, the click rate of the user on the available products is estimated by the click rate estimation model as described above, and so on, the click rate of each user corresponding to the store on the available products is obtained.

[0142] In other words, all users corresponding to the store have corresponding click-through rates for the products that can be put on the shelves, which can numerically describe the degree of attractiveness of the products that can be put on the shelves to all users passing by the store, and then use this to evaluate the most suitable candidate products for the store.

[0143] After obtaining the click rate of each user on the available products, the accumulated click rate is used to obtain the expected value of clicks on the available products in the store, that is, the conversion value as mentioned above.

[0144] The conversion value of the products that can be put on the shelves in the store represents the degree of attraction of the products that can be put on the shelves to the user group passing by the store, and the user group passing by the store is composed of all users corresponding to the store.

[0145] For further information, see Figure 5 , Figure 5 is based on Figure 4 A flowchart of a method for describing the steps of estimating the click rate of all users passing through a store on the available products is shown in the corresponding embodiment.

[0146] The step S301 of estimating the click rate of all users passing by the store on the available products provided by the present application includes:

[0147] Step S3011, for all users passing through the store, construct user-product pairs of the store through user portraits and product features corresponding to the products that can be put on the shelves;

[0148] Step S3012, by predicting the product click-through rate of the user-product pair, the click-through rate of each user on the available products among all users corresponding to the store is obtained.

[0149] First of all, it should be explained that as users initiate entry and exit behaviors, corresponding user itineraries are generated. Many users and the user itineraries of each user will constitute itinerary data. In other words, the user will exist in the form of user itineraries in the data. Therefore, it is possible to construct a user profile describing the user's travel based on this, and then characterize the user with the user profile, which is used to construct user-product pairs with the products that can be put on the shelves.

[0150] The shops that evaluate the click-through rate of products that can be put on the shelves belong to a specific site. Therefore, the user journey used to build the user portrait is related to this specific site, thereby ensuring the accuracy of the features of the constructed user-product pairs and eliminating useless or even interfering data.

[0151] The system built for sites provides each user with travel application deployment based on user terminals. Therefore, each user only needs to interact with the travel application in the user terminal and the gate to initiate his / her own entry or exit behavior, such as entering and exiting the site by scanning the QR code provided by the travel application on the gate. The travel application running on the user terminal and the system connected to the travel application will obtain the user's own travel data such as the user's itinerary.

[0152] That is to say, the user's own travel data will be used to characterize the user in data to build the user's user profile. The user's own travel data is multiple parts of data owned by the deployed travel application, such as the user's itinerary and other travel data mentioned above.

[0153] In addition, the user's own travel data also includes user information on the travel application, tracking data on the travel application, statistical data based on travel data, user's commercial operation data on the travel application, user-related product data, etc., which are not listed here one by one.

[0154] Exemplarily, the user's tracking data on the travel application is used to describe the user's clicks on the exposed products; user information on the travel application includes gender, age, region and occupation; statistical data based on travel data may specifically include but are not limited to the total number of rides in the past 77 / 14 / 30 / 90 days, the total ride duration in the past 7 / 14 / 30 / 90 days, the average entry time of the first trip on weekdays in the past 7 / 14 / 30 / 90 days, the average exit time of the last trip on weekdays in the past 7 / 14 / 30 / 90 days, the average commuting distance per time in the past 7 / 14 / 30 / 90 days, the total number of rides, and the TOP10 high-frequency stations.

[0155] The business operation data of users on travel applications, including but not limited to: the total number of times commercial benefits are received, the total number of times commercial benefits are written off, the total amount of commercial benefits written off, the average price per customer of commercial benefits written off, the total number of times travel benefits are received, the number of times each card voucher is used, and the frequency of card use;

[0156] User-related product data may include but is not limited to: top 10 product categories viewed in the past 1 / 3 / 7 / 14 / 30 / 90 days, corresponding number of times the top 10 product categories were viewed in the past 1 / 3 / 7 / 14 / 30 / 90 days, average price of the top 10 product categories viewed in the past 1 / 3 / 7 / 14 / 30 / 90 days).

[0157] It should be understood that the user's own travel data will be adapted to the input of the travel application for product recommendations for shops to obtain a corresponding user portrait.

[0158] The source of user portrait is not limited to the user's own travel data, but can also come from third-party supplementary data. In short, the user portrait corresponds to the user's own travel data, as well as the combination of the user's own travel data and third-party supplementary data.

[0159] The third-party supplementary data includes, but is not limited to, whether the person is a pregnant woman, the degree of his or her interest in games, whether the person is a budget-conscious person, whether the person is a high net worth person, whether the person often flies, whether the person is interested in financial management, whether the person is interested in the two-dimensional world, whether the person is a budget-conscious person, whether the person is a phone addict, whether the person likes to go shopping in supermarkets, whether the person likes to take the kids out for a walk, whether the person is willing to buy a car, whether the person wants to buy a house or a hospital, whether the person likes to go to museums, whether the person likes to go to historical and cultural sites, whether the person likes to go to natural scenic areas, etc.

[0160] During the execution of step S3011, for each user corresponding to the store, a user portrait of the user is obtained through the user's own travel data, or a combination of the user's own travel data and third-party supplementary data, and the user portrait is then used as the user feature of the user in the constructed user-product pair.

[0161] In other words, the user portrait digitized by the user's own travel data, or the combination of the user's own travel data and third-party supplementary data, will serve as the user feature and constitute a user-product pair together with the product features of the product to be launched.

[0162] The commodity features of the commodities that can be put on the shelves are used to describe the features of the commodities that can be put on the shelves. Exemplarily, the commodity features of the commodities that can be put on the shelves include commodity descriptive features, statistical features, and statistical generalized features of the commodities that can be put on the shelves.

[0163] Specifically, the product characteristics of the products that can be put on the shelves include the first / second / third-level category to which the products can be put on the shelves, the price of the products that can be put on the shelves, the number of exposures of the products that can be put on the shelves in the past 1 / 3 / 7 / 14 / 30, the number of clicks on the products that can be put on the shelves in the past 1 / 3 / 7 / 14 / 30, the number of purchase conversions of the products that can be put on the shelves in the past 1 / 3 / 7 / 14 / 30, the number of days that the products that can be put on the shelves have been on the shelves, the number of exposures of the third-level category to which the products can be put on the shelves in the past 1 / 3 / 7 / 14 / 30, the number of clicks on the third-level category to which the products can be put on the shelves in the past 1 / 3 / 7 / 14 / 30, and the number of purchase conversions of the third-level category to which the products can be put on the shelves in the past 1 / 3 / 7 / 14 / 30.

[0164] Through step S3011, a user-product pair is constructed for each user in the store through the user portrait and the product features of the products that can be put on the shelves, so that the click rate of the user on the products that can be put on the shelves can be predicted based on the user-product pair.

[0165] It should be understood that shops are physical resources that actually exist offline. Users are attracted by the exposure of the available goods on the shops, which is analogous to clicks on goods exposed on the Internet. Therefore, the degree of attraction of the available goods to users through offline exposure of the shops can be accurately measured by the click rate of users on the available goods.

[0166] In step S3012, the pre-built click rate prediction model is used to predict the click rate of user-product pairs to obtain the click rate of users on the available products. It should be understood that the click rate of users on the available products is the probability that the user is attracted by the available products.

[0167] For a shop, each of its corresponding users will construct a user-product pair, and predict the click-through rate of the corresponding products through the constructed user-product pairs. By analogy, the click-through rate of the products that can be put on the shelves for each user pair among all the users corresponding to the shop can be obtained.

[0168] After the click rate of each user on the available products is estimated through the execution of step S302 for all users corresponding to the available products in the shop, step S302 can be executed to accumulate the click rates corresponding to all users and obtain the expected click value of the available products in the shop.

[0169] The click-through rate refers to a single user, while the expected click value of a product that can be listed on a store corresponds to a user group, which includes all users corresponding to the store.

[0170] The expected click value of a product available for sale on a store corresponding to a user group is used to characterize the attractiveness of the product available for sale on a store to the user group. In other words, the expected click value of a product available for sale on a store corresponding to a user group is also the interest level of the user group passing by the store in the product available for sale.

[0171] The user group formed by all users corresponding to the shop obtained through the aforementioned step S110 can be regarded as the user group owned by the shop. The product recommendation for the shop based on the user group should consider the degree of fit between the available products and the user group. The click-through rate estimation between users and available products and the cumulative calculation based on this will accurately quantify the degree of fit between the available products and the user group, and then determine what kind of products each shop should candidate for listing, which greatly ensures the accuracy of product recommendations in the shop and the effectiveness of product listing.

[0172] For product recommendations of a store, the execution of step S120 realizes a rough sorting of the products that can be put on the shelves, so as to obtain candidate products to be put on the shelves, that is, products that may be attracted by passing users, thereby further narrowing the selection range of products to be put on the shelves for product recommendations of the store.

[0173] In step S130, the candidate listing products are sorted based on the ranking, thereby obtaining the listing products of the store and accurately quantifying the placement value of the listing products in the store. It should be understood that the quantified placement value is relative to all users corresponding to the store.

[0174] Exemplarily, a refined binary classification model is pre-built and trained to perform click rate estimation for each candidate listing product by all users corresponding to the store, and obtain a secondary estimated value of the click rate of each user for the candidate listing product.

[0175] It should be understood that for a candidate product listing, each user corresponding to the store will obtain the secondary estimated value of the user's click rate for this candidate product listing through the precise ranking binary classification model. Similarly, the secondary estimated values ​​of the click rate of all users for this candidate product listing can be obtained and accumulated to obtain the secondary estimated value of the click rate of all users corresponding to the store for this candidate product listing.

[0176] Several candidate listing products corresponding to the secondary estimated values ​​of high click rates are used as the listing products of the store, so as to achieve precise ranking of the candidate listing products.

[0177] After the products on the shelves of the store are circled, the placement value of each product on the shelves in the store can be calculated based on its corresponding secondary estimated click rate, click value and the number of exposures of the product on the shelves in the store.

[0178] Specifically, see Figure 6 , Figure 6 is based on Figure 1 The corresponding embodiment shows a method flow chart describing the steps of obtaining the store's listed products and the placement value of the listed products in the store by accurately sorting the estimated conversion values ​​of the candidate listed products through the store portrait.

[0179] The step S130 of finely sorting the estimated conversion values ​​of candidate listed products by using the store portrait to obtain the listed products of the store and the placement value of the listed products in the store provided in the embodiment of the present application includes:

[0180] Step S131, secondarily estimating the click rate of all users corresponding to the store for the candidate listed products, and obtaining a second estimated value of the click rate;

[0181] Step S132, adapting the high click-through rate secondary estimated value to finely sort the candidate listing products and screen the listing products of the store;

[0182] Step S133, calculating the placement value of the listed products in the store according to the corresponding secondary estimated click rate, click value and the number of exposures of the listed products in the store.

[0183] These steps are described in detail below.

[0184] First of all, it should be explained that under the effect of the fine ranking binary classification model for implementing a secondary estimation of the click rate of the candidate listing products, the fine ranking candidate product set, i.e., the candidate listing products referred to above, Invest in the fine sorting two-classification model model select Predictions are made to obtain a secondary estimate of the click rate of the candidate listing product by all users of the specific store, thereby filtering out the listing products.

[0185] For the products selected through fine sorting, the secondary estimated click rate, click value and exposure times of the products in the store are calculated. Perform weighted summation to obtain the estimated value of the product placed in the store, that is, Among them, n user is the number of users of the shop, ctv j is the click value of the listed product. is the number of exposures of user i in time T.

[0186] To further illustrate, the problem loss function for the t-th iteration of the refined binary classification model is:

[0187]

[0188] in, is the predicted value of the ijth record in round t, f t is the tth tree, f t (x) = w q (x), q represents the representation tree f t The structure of w is a tree f t The weight of the leaf in the middle is expressed as w i To represent the weight of the i-th leaf, Ω(f t ) is about the tree f t The complexity of the regular term, then L (t) At the point The second-order approximation is:

[0189]

[0190] in,

[0191] So far, the optimal solution to the problem is:

[0192]

[0193] Among them, q is a tree with a fixed structure, γ and λ are the first-order and second-order regularization coefficients respectively.

[0194] The optimal solution of each round based on the training data set can be iteratively converged to obtain the refined binary classification model model select .

[0195] After obtaining the store's listed goods and their placement values ​​in the store through step S130, in step S140, a store's listed goods processing strategy is generated for the selected goods and / or purchased goods in the store's operation.

[0196] During the execution of step S140, all purchased commodities and unselected but available commodities of the store are enumerated, ie, selected commodities, and corresponding shelf commodity processing strategies are generated respectively.

[0197] Exemplarily, step S140 includes: enumerating the purchased goods of the store, generating a shelf commodity processing strategy for displaying the purchased goods in a prominent position in the store for the purchased goods with high investment value, and generating a shelf commodity processing strategy for no longer purchasing or returning or exchanging the purchased goods with low investment value.

[0198] Exemplarily, step S140 includes: enumerating the product selections of the store, and generating a product listing processing strategy for the store replenishment demand for the product selections with high investment value.

[0199] Based on this, the display position of the purchased goods can be adjusted. For example, goods with high estimated investment value can be displayed in a more prominent position in the store. For goods that have been purchased but with low estimated investment value, the customer can choose not to purchase them or even return or exchange them.

[0200] You can also use this as a basis to generate replenishment needs and choose to replenish products with high investment value.

[0201] At this point, we will build an intelligent product recommendation system for merchants based on the merchant traffic generated by users, allowing merchants to dynamically select products and achieve high customer traffic conversion.

[0202] And for the owners, the implementation of this application can also help merchants achieve good performance, which in turn helps to stabilize or even improve the owners' shop leasing business.

[0203] The following is a specific example to illustrate the implementation of product recommendations in commercial operations provided by this application.

[0204] First, for a shop, a shop portrait is obtained based on all its corresponding users, which is used to form user-product pairs on the shop with the constructed product features, and then the estimated click-through rate is obtained.

[0205] Specifically, the distribution of the number of times a user exits a station at a gate group is obtained through the user itinerary table and the gate information table. When a user exits a station by scanning a QR code of a travel application, the distribution of the number of times a user exits a station at a gate group describes the number of times the user scans the code at each gate, that is, the user gate group scan code table is obtained.

[0206] That is to say, the user travel table is associated with the gate information table through the gate code to obtain the user gate group code swiping table.

[0207] Based on the user gate group code scanning table, the distribution of user code scanning in this group of gates can be known, and the distribution of user code scanning in this group of gates will indicate the user's exit direction.

[0208] Combined with the gate group entrance and exit distance relationship described in the gate group entrance and exit distance relationship table, the exits pointing to different directions of the gate group are known, and the user's frequently used exits can be determined, and the user's recently frequently used exit table can be obtained.

[0209] For each exit, the shop mapped to the exit is recorded to build and maintain a shop exit relationship table.

[0210] Through the mapping of the exit to the shop, it is possible to know the shop that the user will pass by when leaving the station from an exit. In other words, all users passing by the shop are all users corresponding to the shop.

[0211] Assign user portraits of all users corresponding to the shop to the shop, further enrich its labels, and obtain a shop portrait table.

[0212] At this point, the most suitable products for the store can be found, thereby realizing dynamic product selection and shelf recommendations for the store.

[0213] It should be further explained that for all users corresponding to a shop, the commonly used exit gates can be determined by counting the users' preferred exit directions in the code scanning table of the user gate group, and then all users corresponding to the shop can be determined based on this.

[0214] In addition, if the user has fewer trips in the user itinerary, for example, the number of exits of the user at the site gate group is less than the set threshold, it will be necessary to determine the user's preferred exit direction through the distribution of the user's exit times at each gate of the site gate group and the user portrait through a binary classification model, and then determine the exit gate to which the user belongs in this exit direction.

[0215] The probability that the user belongs to each exit is obtained through a pre-trained binary classification model, and then the exit to which the user belongs is finally determined by the probability. Exemplarily, the pre-trained model can be a binary classification model, which is characterized by the distribution of the number of times the user exits each gate group at the station and the user portrait. Under this feature, the direction value corresponding to the feature is marked according to the direction of the user after exiting the station, and each direction value uniquely corresponds to an exit that can be passed from the gate group.

[0216] For example, if the user in the feature mapping passes to the left after exiting the gate group, the marked direction value is 0; correspondingly, if the user in the feature mapping passes to the right after exiting the gate group, the marked direction value is 1.

[0217] Based on this, construct a training data set

[0218] Among them, each training data set D train In a single record in It is the feature of this record, which is used to describe the user status of the corresponding user.

[0219] Specific It includes the distribution of the number of times users exit each gate in a site gate group at a certain time and the user portrait, and the user portrait can be represented by multiple user tags. is the characteristic of user i in gate group j, R m is the m-dimensional feature of user i.

[0220] For example, x ij ∈R m= [date (YYYYMM), user tag 1 of user i, user tag 2 of user i, ..., user tag n of user i label , the number of exits of user i at gate 1 of gate group j, the number of exits of user i at gate 2 of gate group j, ..., the number of exits of user i at gate n of gate group j zj The date (YYYYMM) represents the month, such as 202312, which means December 2023. The date feature indicates the time range corresponding to other features. label Refers to the nth label User tags, that is, n r Represents the number of users, n i represents the number of gate groups that user i faces when exiting the station during travel, n label Represents the number of user tags, y ij ∈{0,1} is the direction value of the feature annotation in a single record of the corresponding training data set. ij ∈{0,1} represents the final exit direction of user i under the corresponding characteristics (especially the actual number of exits).

[0221] That is, y in a single record of the training data set ij is the marked direction value. For example, if user i turns left after exiting gate group j, then y ij =0, if going right, then y ij =1.

[0222] By training the model with the constructed training data set, a model is obtained that can accurately predict the exit to which the user belongs, providing the user with accurate judgment of the exit affiliation. There is no need to obtain the location, that is, it is no longer restricted by GPS positioning and the acquisition of location permissions.

[0223] Furthermore, the training for exit prediction will be conducted for gate groups corresponding to only one exit. A training data set is constructed for the gate group corresponding to the only exit, and features are constructed based on all records related to the gate group, and direction values ​​are annotated for the constructed features based on the unique direction of the gate group.

[0224] Specifically, the number of times the user exits each gate of the gate group in the site is obtained through the travel data and the gate information, that is, the user gate group exit data of user-site-gate group-gate-exit number is obtained, which can be in a tabular form, that is, a table maintained according to fields such as user, site, gate group, gate, and exit number, and the user gate group exit data corresponding only to the only exit is obtained therefrom, and the number of times the user exits each gate of the gate group at the site is obtained from the user gate group exit data corresponding only to the only exit, and the feature is constructed based on this.

[0225] In addition, the corresponding user portraits will be extracted for feature construction, and the constructed features will be labeled with directional values ​​to finally form a training data set.

[0226] The constructed training data set shields the influence of other exits and only considers the prediction of the user's exit in the feature dimension, so that the subsequent predictions will not be affected by other factors, which will greatly improve the accuracy and reliability of the prediction.

[0227] Exemplarily, a gate group entrance and exit distance relationship will also be constructed. The gate group entrance and exit distance relationship is used to indicate the distance of each gate of the gate group in a site relative to the exits in each direction. Therefore, after predicting the most likely direction of the user after exiting the gate, it can be directly mapped to the closest exit based on the gate group entrance and exit distance relationship. This exit is the exit to which the predicted user belongs.

[0228] As mentioned above, the model trained by constructing the obtained training data set can be a binary classification model, and thus, the obtained binary classification model will be used to predict the exit to which the user belongs. Specifically, the binary classification model is used to predict the most likely direction of the user after exiting from the gate group according to the number of exits of the user at each gate on the station gate group and the user portrait, and the exit mapped by the direction is used as the exit to which the user belongs.

[0229] To further illustrate, a prediction data set is constructed based on the number of times a user exits each gate at the site gate group and the user profile. Among them, unlike the training data set, the prediction data set is not limited to the gate group corresponding to the only exit; in other words, it will not distinguish whether the gate group corresponds to the only exit, and all the user's data will be used to construct the prediction data set, and then the exit to which the user belongs will be predicted based on the characteristics of the constructed prediction data set.

[0230] Prediction dataset D predict In each record Represents the characteristics of user i in gate group j whose exit gate has not yet been determined. As mentioned above, the characteristics describe the number of exits of user i at each gate in gate group j and the user tag. The characteristics list included is similar.

[0231] Based on the obtained binary classification model, the prediction data set D predict Each record of Use the binary classification model to implement predictions one by one, and get the probability that the characteristics of user i in gate group j are biased in one direction, that is, belonging to an exit corresponding to gate group j.

[0232] For example, if the gate group j corresponds to two exits. In other words, after exiting the gate group j, user i has two directions, each direction corresponds to an exit, then the probability of leaning towards the other direction, that is, belonging to the other exit, is

[0233] In addition, if there are more than two exits in one direction of gate group j, that is, the exit in this direction is not unique, then when the binary classification model predicts that user i tends to go in a direction with more than two exits after exiting gate group j, the exit that user i is most likely to pass through will be determined based on the traffic flow of each exit, and this exit is the exit to which user i belongs.

[0234] Specifically, each exit in one direction has its flow, which can be sensed by the passenger flow counter set at the exit. The flow rate corresponding to each exit is used to determine the flow ratio of each exit, and then the exit to which user i belongs is determined based on the flow ratio.

[0235] Therefore, after the user's belonging exit is determined, the user's travel trajectory is mapped to the belonging exit according to the user's belonging exit, and the user's itinerary during travel will also be mapped to this exit.

[0236] In other words, for the exit corresponding to the travel trajectory, on the one hand, we can know which users have passed through this exit, and then we can match these users to the positions at this exit, that is, determine which users correspond to the positions at this exit, and match the user portraits of these users to the positions at this exit, such as media locations, shop locations, etc.

[0237] On the other hand, the exit corresponding to the travel trajectory provides the tracked travel trajectory within the station, breaking out of the restricted GPS positioning and the required location permissions, and can also obtain the user's in-station trajectory, enhancing the in-station tracking performance.

[0238] As the user's travel trajectory out of the station and the exit in the user's itinerary can be matched, the user is aggregated at the location at the exit, and the aggregated user is assigned a location at the exit.

[0239] The user aggregation referred to is to determine the corresponding user group for the location at the exit, and to describe and characterize the aggregated user group through the user profiles of the aggregated users.

[0240] The travel trajectory is mapped to the corresponding exit so that the corresponding user trip can be mapped to the exit, and then the user who generated the user trip is also mapped to the exit, so that the users corresponding to the same exit are aggregated.

[0241] The location mapped by the exit refers to the location on the exit. For example, in terms of location type, the location includes but is not limited to the media location, the store location, etc.; in terms of geographical location, the location mapped by the exit is any location in the direction of the exit, such as the location inside the station in the direction of the exit, the exit passage in the direction of the exit, and the location outside the station exiting from the exit.

[0242] In terms of data, users exist in the form of user portraits and their travel trajectories. Therefore, user aggregation is achieved in the form of travel trajectories and user portraits, and the user portraits corresponding to the aggregated users are assigned to the location mapped by the exit, so as to accurately locate and describe the user characteristics of the location through the assigned user portraits. For example, the user portraits of the aggregated users are assigned to shops to construct corresponding shop portraits.

[0243] For the obtained shop portrait, a click-through rate prediction recommendation algorithm will be used for each user corresponding to the shop. That is, the constructed click-through rate prediction model will estimate the click-through rate of the products to be placed in the shop, and then the click-through rate on the shop will be accumulated to obtain the expected value of clicks, that is, the shop product click-through rate estimation table will be obtained.

[0244] The expected click value of the products to be placed in the store is provided through the store product click rate estimation table, and the conversion value of the products in the store is obtained by combining the click value of the products, and the store product conversion value table is obtained, which is finally used to execute product recommendations.

[0245] It should be added that the click-through rate prediction recommendation algorithm, that is, the click-through rate prediction model, is trained based on online product exposure click data from platforms such as travel applications. The training data set and the product features of the products to be launched are obtained from the constructed product label table, which will not be elaborated here.

[0246] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation methods of the present application.

[0247] In an exemplary embodiment of the present application, a computer program medium is further provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a computer, the computer executes the method described in the above method embodiment.

[0248] According to one embodiment of the present application, a program product for implementing the method in the above method embodiment is also provided, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited to this. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0249] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0250] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0251] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0252] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0253] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0254] In addition, although the steps of the method in the present application are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0255] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation methods of the present application.

[0256] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the appended claims.

Claims

1. A method for recommending goods in commercial operations, characterized in that: The method comprises: By determining all users passing through a shop during business, a shop profile of the shop is obtained, wherein the shop profile includes a user profile of each user; The exposure conversion of the products that can be put on the shelves in the store is estimated by using the store portrait, and the products that can be put on the shelves are roughly ranked to obtain candidate products for putting on the shelves; The estimated conversion values ​​of the candidate listed products are precisely ranked by the store portrait to obtain the listed products of the store and the placement value of the listed products in the store; A shelf merchandise processing strategy for the store operation is generated by adapting the two types of shelf merchandise, namely, purchased merchandise and selected merchandise.

2. The method according to claim 1, characterized in that The step of obtaining a store portrait of the store by determining all users passing through the store during business, includes: For shops in the business, all corresponding users are obtained according to the exit corresponding to the shop at the station; The user portraits of all users corresponding to the exit are assigned to the shop to form a shop portrait of the shop.

3. The method according to claim 1, characterized in that The step of estimating the exposure conversion of the products that can be put on the shelves in the store through the store portrait, roughly sorting the products that can be put on the shelves, and obtaining candidate products for putting on the shelves includes: For the products that can be put on the shelves, the conversion value of the products that can be put on the shelves when they are exposed on the shelves in the stores is estimated through the store portrait; According to the estimated conversion value, products with high conversion value are selected from the available products as candidate products for listing.

4. The method according to claim 3, characterized in that: The estimating of the conversion value of the available products when they are exposed in the store through the store portrait includes: Estimate the click rate of each user on the available products among all users who pass by the store; The click rates are accumulated to obtain the conversion value of the exposure of the available products in the shop.

5. The method according to claim 4, characterized in that The estimated click rate of all users passing through the store for the available products includes: For all users who pass by the shop, construct user-product pairs of the shop through user portraits and product features corresponding to the products that can be put on the shelves; By predicting the product click-through rate of the user-product pair, the click-through rate of each user for the available products among all users corresponding to the store is obtained.

6. The method according to claim 1, characterized in that The step of finely sorting the estimated conversion values ​​of the candidate listed products by the store portrait to obtain the listed products of the store and the placement value of the listed products in the store includes: Secondarily estimating the click rate of all users corresponding to the store for the candidate listed products to obtain a secondary estimated value of the click rate; Adapting the high click-through rate secondary estimated value to finely sort the candidate listing products and screen the listing products of the store; The placement value of the listed products in the shop is calculated according to the corresponding secondary estimated click rate, click value and the number of exposures of the listed products in the shop.

7. The method according to claim 1, characterized in that The listed products include purchased products, and the two types of purchased products and selected products are adapted to generate the listed products processing strategy in the store operation, including: The purchased goods of the store are enumerated, and a shelf commodity processing strategy is generated for the purchased goods with high investment value, in which the purchased goods are displayed in a prominent position of the store; and a shelf commodity processing strategy is generated for the purchased goods with low investment value, in which the purchased goods are not to be purchased or returned.

8. The method according to claim 1, characterized in that The listed products include selected products, and the two types of listed products adapted to the purchased products and the selected products generate the listed products processing strategy in the store operation, including: The product selections of the store are enumerated, and a product listing processing strategy for the store's replenishment needs is generated for the product selections with high investment value.

9. A commodity recommendation system in commercial operation, the system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that The computer program is used by a processor to execute the steps of the method according to any one of claims 1 to 8.

Images