Scene perception-oriented retail AI intelligent recommendation system and method

Abstract

The invention provides a scene perception-oriented retail AI intelligent recommendation system and method, and the method comprises the steps: constructing retail scene information provided for a user by each retail channel according to real-time retail data uploaded by different retail channels in combination with retail commodity information corresponding to each retail channel, and analyzing the corresponding real-time retail data to generate a commodity recommendation sequence corresponding to the retail channel, determining interested commodities of the user according to browsing data of the user in the selected retail channel and corresponding retail scene information, and adjusting the commodity recommendation sequence in real time. The browsing duration of the user for each retail commodity is deduced according to the browsing data, and the corresponding commodity recommendation reason is generated and displayed in combination with the commodity advantages corresponding to each retail commodity, so that the recommendation content is enriched, the user can find more characteristic commodities meeting requirements in different channels, and the user experience is improved. The comprehensiveness and individuation degree of recommendation are improved, and the satisfaction degree of the user to the recommendation system is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent retail management, and particularly relates to a scene-aware retail AI intelligent recommendation system and method. BACKGROUND

[0002] In today's retail industry, with the rapid development of e-commerce and the increasing diversification of consumer demand, providing personalized and accurate product recommendations for customers has become a key factor in improving the competitiveness of retail enterprises. Traditional retail recommendation systems often only rely on customers' historical purchase data or simple product attributes for recommendations, but ignore the influence of the specific shopping scene on customers' purchase decisions. For example, the same customer may have different preferences for products in different locations and at different times, or the best-selling products may vary in different time periods in different scenarios. However, traditional recommendation systems lack the ability to perceive and utilize scene information, resulting in recommendations that often do not match customers' actual needs, reducing customer shopping experience and purchase conversion rates, and indirectly reducing customer spending. Over time, traditional recommendation systems have gradually exited the market. Therefore, there is a need for a scene-aware and intelligent retail recommendation technology in the market.

[0003] Therefore, the present application provides a scene-aware retail AI intelligent recommendation system and method. SUMMARY

[0004] The present application provides a scene-aware retail AI intelligent recommendation system and method, which solves the problem of existing retail recommendation systems that cannot effectively perceive customer shopping scenes and have low recommendation accuracy, and realizes accurate product recommendations based on scene awareness.

[0005] The present application provides a scene-aware retail AI intelligent recommendation system, which includes: A scene training module for constructing retail scene information provided by each retail channel for users based on real-time retail data uploaded by different retail channels and corresponding retail product information for each retail channel; A product planning module for analyzing corresponding real-time retail data using the retail scene information and generating a product recommendation order for the corresponding retail channel; A recommendation display module for obtaining browsing data of a user in a selected retail channel, determining the user's interest products in combination with the corresponding retail scene information, and adjusting the product recommendation order in real time; An intelligent optimization module for deriving the user's browsing duration for each retail product based on the browsing data, generating and displaying corresponding product recommendation reasons in combination with the corresponding product advantages of each retail product.

[0006] In an implementable manner, Also comprising: A data processing module is configured to acquire a commodity display structure corresponding to each of the retail channels respectively, and determine a channel attribute corresponding to each of the retail channels, determine a data collection method corresponding to the retail channel according to the channel attribute, and construct a unified data collection scheme of all the retail channels; Real-time scene data corresponding to each of the retail channels is collected respectively by using the unified data collection scheme, and when the real-time scene data contains a request field, the real-time scene data is mapped into a preset metadata template to generate metadata corresponding to the request field; A request position corresponding to the request field is determined according to the metadata, the request field is labeled in the corresponding commodity display structure, and real-time commodity dynamic data corresponding to the retail channel is generated; Request result data corresponding to each of the request positions is derived according to the real-time scene data; The real-time commodity dynamic data and the request result data corresponding to each of the retail channels are fused respectively to generate real-time retail data of the retail channel and upload to the scene training module.

[0007] In an implementable manner, The scene training module comprises: An information processing unit is configured to acquire commodity dynamic data corresponding to each of the retail channels respectively, and construct a sales-replenishment process corresponding to different retail commodities in each of the retail channels in combination with a commodity display structure corresponding to each of the retail channels, and generate retail commodity information of the retail channel; A sales analysis unit is configured to match the real-time retail data corresponding to each of the retail channels and the retail commodity information corresponding thereto respectively, obtain a plurality of sales events corresponding to each of the retail commodities, and acquire a sales timestamp corresponding to each of the sales events respectively; A dynamic analysis unit is configured to acquire a sales address corresponding to each of the retail channels, derive a sales cause of each of the sales events in combination with a specific scene feature corresponding to each of the sales addresses, and generate sales dynamic information of each of the retail channels in combination with the corresponding sales timestamp; A scene construction unit is configured to derive a consumption tendency of a user in each of the retail channels according to the sales dynamic information by using an AI technology, match the consumption tendency with the specific scene feature corresponding thereto, and construct retail scene information of the retail channel by using a target specific scene feature that is successfully matched.

[0008] In an implementable manner, The commodity planning module comprises: The modeling analysis unit is configured to construct a retail scene model corresponding to each retail channel according to retail scene information and real-time retail data corresponding to the retail channel, and to count a single-day sales volume and a single-day sales frequency corresponding to each retail commodity in the retail scene model; The commodity layering unit is configured to perform sales layering on the retail commodities according to the single-day sales volume, to derive a next sales time period of each sales layer in combination with the single-day sales frequency corresponding to each retail commodity, and to set a corresponding overall recommendation order for the corresponding retail commodity according to the next sales time period corresponding to each sales layer. The sampling comparison unit is configured to sample commodities for each retail channel respectively to obtain a sales volume difference of a same retail commodity corresponding to different retail channels, to obtain a current recommendation order of the same retail commodity corresponding to different retail channels, to take the first current recommendation order as a fixed recommendation order of the same retail commodity, to adjust each current recommendation order by using the sales volume difference, and to obtain a derived commodity recommendation order corresponding to each same retail commodity. The order determination unit is configured to adjust the overall recommendation order by using the derived commodity recommendation order, to screen unique retail commodities in each retail channel that do not contain the derived commodity recommendation order, to analyze a commodity similarity between each unique retail commodity and the same retail commodity corresponding thereto by using AI technology, to adjust the overall recommendation order according to the commodity similarity, and to generate a commodity recommendation order corresponding to each retail commodity.

[0009] In an implementable manner, Further comprising: Counting a first commodity quantity corresponding to a same retail commodity and a second commodity quantity corresponding to a unique retail commodity in each retail channel respectively; Screening a target retail channel in which the first commodity quantity is less than the second commodity quantity; Obtaining a current commodity recommendation order corresponding to the target retail channel, inputting the current commodity recommendation order into the retail scene model corresponding thereto for first sales simulation, and obtaining a first estimated single-day sales volume corresponding to the target retail channel; When the first estimated single-day sales volume is less than an average single-day sales volume of the target retail channel, iteratively adjusting the current commodity recommendation order according to an overall recommendation order corresponding to the target retail channel, and performing second sales simulation by using the retail scene model to obtain a second estimated single-day sales volume corresponding to each iteration result; Obtaining an optimal iteration result with the highest second estimated single-day sales volume, and generating a commodity recommendation order corresponding to each retail commodity in the target retail channel.

[0010] In an implementable mode, The recommendation display module comprises: The data acquisition unit is configured to determine a starting browsing time of the user in the retail channel, acquire the starting browsing time in real time, and obtain corresponding browsing data of the user in the retail channel. The product matching unit is configured to derive real-time display product information of the retail channel based on real-time browsing responses of the retail channel to the browsing data, and determine a browsing duration of the user for the real-time display product information based on the browsing data. The scene fusion unit is configured to identify a real-time display scene corresponding to each of the real-time display product information, analyze interest degrees of the user for each display product based on the corresponding browsing duration, and determine an interest product of the user in the retail scene. The real-time adjustment unit is configured to set a priority for the interest product for priority display, acquire a plurality of product features corresponding to remaining products in the retail scene, and set a corresponding recommendation right for each of the remaining products based on a feature similarity between each of the remaining products and the interest product, and perform recommendation display.

[0011] In an implementable mode, The intelligent optimization module comprises: The browsing analysis unit is configured to construct a browsing track of the user in the retail channel based on the browsing data, determine a plurality of browsed products of the user and a browsing duration corresponding to each of the browsed products, and generate a browsing duration feature of the user. The depth analysis unit is configured to generate a browsing preference feature of the user based on browsing operations of the user for each of the browsed products, and derive a browsing interest dimension of the user for each retail product in the retail channel based on the browsing duration feature and the browsing preference feature. The advantage matching unit is configured to match a product advantage corresponding to each of the retail products with the browsing interest dimension to obtain a recommendation advantage corresponding to each of the retail products, perform display rendering on the recommendation advantage based on the browsing preference feature, and generate and display a product recommendation reason corresponding to each of the retail products.

[0012] In an implementable mode, Further comprising: When the user purchases a retail product, a remaining quantity to be sold corresponding to the retail channel is generated. A remaining sales duration corresponding to the retail product is derived based on the remaining quantity to be sold, and a restocking guide corresponding thereto is generated and displayed.

[0013] The application provides a scene-aware retail AI intelligent recommendation method, comprising: Step 1: constructing retail scene information provided by each retail channel for users according to real-time retail data uploaded by different retail channels and retail product information corresponding to each retail channel; Step 2: analyzing the real-time retail data corresponding to the retail scene information to generate a product recommendation sequence corresponding to the retail channel; Step 3: obtaining browsing data of a user in a selected retail channel, determining the user's interest product in combination with the corresponding retail scene information, and adjusting the product recommendation sequence in real time; Step 4: deriving the user's browsing time length for each retail product according to the browsing data, generating and displaying the corresponding product recommendation reason in combination with the product advantages corresponding to each retail product.

[0014] In an implementable manner, The step 3 comprises: Step 31: constructing a retail scene model corresponding to each retail channel according to the retail scene information and real-time retail data corresponding to each retail channel, and counting the single-day sales volume and single-day sales frequency corresponding to each retail product in the retail scene model; Step 32: performing sales stratification on the retail products according to the single-day sales volume, and deriving the next sales time period of the corresponding sales stratification in combination with the single-day sales frequency corresponding to each retail product, and setting the corresponding overall recommendation sequence for the corresponding retail product according to the next sales time period corresponding to each sales layer; Step 33: sampling the products in each retail channel respectively to obtain the sales volume difference of the same retail product corresponding to different retail channels, obtaining the current recommendation sequence of the same retail product corresponding to different retail channels, taking the first current recommendation sequence as the fixed recommendation sequence of the same retail product, adjusting each current recommendation sequence by using the sales volume difference, and obtaining the derived product recommendation sequence corresponding to each same retail product; Step 34: adjusting the overall recommendation sequence by using the derived product recommendation sequence, screening unique retail products in each retail channel that do not contain the derived product recommendation sequence, analyzing the product similarity between each unique retail product and the corresponding same retail product by using AI technology, adjusting the overall recommendation sequence according to the product similarity, and generating the product recommendation sequence corresponding to each retail product.

[0015] The implementable beneficial effects of the technical scheme are as follows: in order to accurately recommend goods to users and improve the interest of users in goods, firstly, a dedicated retail scene is constructed according to real-time data and goods information of different retail channels, so that the system can accurately capture the scene characteristics of each channel, and the goods recommendation is adapted to the channel scene from the source, and the fitting degree of recommendation and scene is improved, then real-time retail data is analyzed based on the retail scene information, and a goods recommendation sequence is generated, so as to ensure that the recommendation sequence conforms to the operation logic of the channel and the potential demand of the user in the scene, avoid the low efficiency problem of the general recommendation mode in the specific channel, further combine the browsing data of the user in the selected channel and the corresponding scene information, and adjust the recommendation sequence in real time, so as to quickly capture the interest change of the user in the current scene, dynamically update the recommendation according to the user interest, improve the attention of the user to the recommended goods, and then derive the browsing time of the user to the goods, combine the advantages of the goods to generate a recommendation reason, so that the recommendation not only shows the goods, but also accurately delivers the core selling point that moves the user, enhances the persuasiveness of the recommendation, improves the purchase willingness of the user, finally, the scene is constructed for different retail channels and the recommendation is optimized, so as to avoid the fragmentation of the user due to the abruptness of the recommendation style, improve the consistency and smoothness of the cross-channel shopping experience, in this way, the suitable goods can be recommended to the user according to the hobby and actual use of the user, all retail channels can be adapted, the demand of the user can be responded in time, and the efficiency of the goods sales and the satisfaction of the user are improved.

[0016] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application can be realized and attained by means of the instrumentalities particularly pointed out in the written description and the appended drawings.

[0017] The technical scheme of the present application will be further described in detail below by means of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation on the present application. In the drawings: Figure 1 It is a composition schematic diagram of a scene-aware retail AI intelligent recommendation system in an embodiment of the present application; Figure 2 It is a working flow schematic diagram of a scene-aware retail AI intelligent recommendation method in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The preferred embodiments of the present application are described below in combination with the drawings, and it should be understood that the preferred embodiments described herein are only used as an illustration and explanation of the present application, and do not serve as a limitation on the present application.

[0020] Embodiment 1: The embodiment provides a scene-aware retail AI intelligent recommendation system, which comprises the following modules. Figure 1 As shown in the figure, the system comprises the following modules. A scene training module, configured to construct retail scene information provided by each retail channel for users according to real-time retail data uploaded by different retail channels and retail commodity information corresponding to each retail channel; A commodity planning module, configured to analyze the real-time retail data corresponding to the retail scene information to generate a commodity recommendation sequence of the retail channel; A recommendation display module, configured to acquire browsing data of a user in a selected retail channel, determine the interested commodity of the user in combination with corresponding retail scene information, and adjust the commodity recommendation sequence in real time; An intelligent optimization module, configured to derive the browsing time length of each retail commodity of the user according to the browsing data, generate and display a corresponding commodity recommendation reason in combination with the commodity advantages corresponding to each retail commodity.

[0021] In this example, the retail channels include online channels and offline channels, and each channel includes several different channels. In this example, the real-time retail data represent data generated by the retail channel during retail work. In this example, the retail scene information represents the purchase scene provided by the retail channel for the user, for example, a concise scene, a climate-related scene, a trend-related scene, etc. In this example, the browsing data represent data generated by the user during purchase in the retail channel, and the selected retail channel represents the retail channel currently used by the user. In this example, since the browsing time length of each user is different, the display mode of the commodity recommendation reason is also different, for example, the browsing time length of user A is greater than 1 minute, so all the advantages of the retail commodity are displayed, and the browsing time length of user B is less than 30 seconds, so the outstanding advantages of the retail commodity are displayed.

[0022] The working principle and beneficial effects of the technical solution are as follows: in order to accurately recommend goods to users and improve the interest of users in goods, first, an exclusive retail scene is constructed according to real-time data and goods information of different retail channels, so that the system can accurately capture the scene characteristics of each channel, and the goods recommendation is adapted to the channel scene from the source, thereby improving the fitting degree of the recommendation and the scene, then real-time retail data is analyzed based on the retail scene information to generate a goods recommendation sequence, so as to ensure that the recommendation sequence conforms to the operation logic of the channel and the potential needs of users in the scene, thereby avoiding the low efficiency of the general recommendation mode in a specific channel, further combining the browsing data of users in the selected channel and the corresponding scene information, and adjusting the recommendation sequence in real time, so as to quickly capture the interest changes of users in the current scene, dynamically update the recommendation according to the interest of users, improve the attention of users to the recommended goods, and then derive the browsing time of users to the goods, and generate a recommendation reason combining the advantages of the goods, so that the recommendation not only shows the goods, but also accurately delivers the core selling points that move users, enhances the persuasiveness of the recommendation, and improves the purchase willingness of users, finally, the scene is constructed for different retail channels and the recommendation is optimized, so as to avoid the fragmentation of users due to the abruptness of the recommendation style, improve the consistency and smoothness of the cross-channel shopping experience, in this way, the suitable goods can be recommended to users according to their hobbies and actual use, all retail channels can be adapted, the needs of users can be responded in time, and the efficiency of goods sales and the satisfaction of users can be improved.

[0023] Embodiment 2 Based on the embodiment 1, the scene-aware retail AI intelligent recommendation system further comprises: A data processing module is configured to acquire the goods display structure corresponding to each retail channel respectively, and determine the channel attribute corresponding to each retail channel, determine the data acquisition mode corresponding to the retail channel according to the channel attribute, and construct a unified data acquisition scheme for all retail channels. Real-time scene data corresponding to each retail channel is collected by using the unified data acquisition scheme, and when the real-time scene data contains a request field, the real-time scene data is mapped into a preset metadata template to generate metadata corresponding to the request field. According to the metadata, the request position corresponding to the request field is determined, and the request field is labeled in the corresponding goods display structure to generate real-time goods dynamic data corresponding to the retail channel. According to the real-time scene data, the request result data corresponding to each request position is derived. The real-time goods dynamic data and the request result data corresponding to each retail channel are fused respectively to generate real-time retail data of the retail channel and upload the real-time retail data to the scene training module.

[0024] In this example, the commodity display structure represents the structure of the displayed commodities in the retail channel; In this example, the channel attribute includes online attribute and offline attribute; In this example, the unified data collection scheme represents a scheme for collecting real-time scene data of all retail channels; In this example, the request field represents the character corresponding to the data segment when the user operates in the retail channel; In this example, the real-time scene data represents real-time data generated in the retail channel; In this example, the request position represents the destination position when the user operates in the retail channel; In this example, the metadata represents data containing information of the structure, content, source and format of the request field in the real-time scene data; In this example, the preset metadata template is composed of structure, content, source and format; In this example, the real-time commodity dynamic data represents data presented when the retail commodity in the retail channel changes, and the operation of obtaining real-time commodity dynamic data accurately locates the position of data related to user demand in the commodity display system, realizes the association and integration of request data and actual commodity display scene, and enables the system to clearly master the specific position and dynamic information of the commodity in each channel which the user pays attention to; In this example, the request result data represents data generated when the request position responds to the user's request.

[0025] The working principle and beneficial effects of the above technical solution are as follows: in order to ensure the effectiveness and accuracy of real-time retail data, eliminate data loss and data redundancy, first, the channel attribute of each retail channel is determined, and the corresponding data collection method is customized according to the channel attribute, then a unified data collection scheme is constructed, when the real-time scene data contains the request field, the module maps it to the preset metadata template to generate metadata, and then determines the request position and marks it in the commodity display structure, generates real-time commodity dynamic data, derives the request result data corresponding to the request position, and fuses the real-time commodity dynamic data and the request result data to generate real-time retail data, so that the data uploaded to the scene training module not only contains the information of the commodity itself, but also deeply fuses multi-dimensional information such as user request, commodity display position, scene dynamic, etc., which significantly improves the scene correlation of the data, and finally the construction of the unified data collection scheme considers the characteristics of different retail channels, so that the system can flexibly adapt to the newly added retail channels, greatly enhancing the expansibility of the system and the adaptability to the changes of the retail channels, providing a standardized and consistent data basis for subsequent scene training, commodity recommendation and other links.

[0026] Embodiment 3: On the basis of embodiment 1, the scene-aware retail AI intelligent recommendation system comprises a scene training module, which comprises: An information processing unit is configured to acquire dynamic data of each retail channel, and construct a sales-replenishment process of each retail channel based on the dynamic data and a corresponding commodity display structure. A sales analysis unit is configured to match the real-time retail data and the corresponding retail commodity information of each retail channel to obtain a plurality of sales events corresponding to each retail commodity, and acquire a sales timestamp corresponding to each sales event. A dynamic analysis unit is configured to acquire a sales address corresponding to each retail channel, derive a sales cause of each sales event based on a specific scene feature corresponding to the sales address, and generate sales dynamic information of each retail channel based on the corresponding sales timestamp. A scene construction unit is configured to derive a consumption tendency of a user in each retail channel based on the sales dynamic information, match the consumption tendency with the specific scene feature, and construct retail scene information of the retail channel based on a target specific scene feature that is successfully matched.

[0027] In this example, the sales-replenishment process represents a dynamic process of sales and replenishment of retail commodities in a retail channel. In this example, the retail commodity information represents dynamic information of retail commodities in a retail channel. In this example, a retail scene generates a corresponding sales event for each retail commodity sold. In this example, the sales cause represents a factor that contributes to a sales event, such as promotion, climate, user preference, and subsidies. In this example, the sales timestamp represents the completion time of a sales event. The timestamp can also reflect the time distribution characteristics of commodity sales, laying a data foundation for subsequent derivation of sales causes based on scene features, and helping to capture the association between commodity sales and time scenes. In this example, the consumption tendency represents the reason for a user to consume in a retail channel.

[0028] The working principle and beneficial effects of the above technical solution are: by analyzing the commodity dynamic data and commodity display structure, the sales-replenishment process of different retail commodities is constructed, which can clearly present the whole process dynamics of the commodities from sales to replenishment, then the real-time retail data is matched with the retail commodity information to obtain a plurality of sales events of each retail commodity and the corresponding sales time stamp, the sales cause of the sales event is deduced in combination with the specific scene characteristics of the sales address, the sales dynamic information is generated in combination with the sales time stamp, finally the user consumption tendency is deduced by using AI technology, and is matched with the specific scene characteristics, and finally the retail scene information is constructed. In this way, it is ensured that the retail scene information can truly reflect the association between the user's consumption habits and the scene in the channel, so that the subsequent commodity recommendation can better meet the real needs of the user in the scene.

[0029] Embodiment 4: On the basis of embodiment 1, the scene-aware retail AI intelligent recommendation system comprises a commodity planning module, which comprises: A modeling analysis unit is configured to construct a retail scene model corresponding to each retail channel according to the real-time retail data and the retail scene information corresponding to each retail channel, and to count the daily sales volume and the daily sales frequency corresponding to each retail commodity in the retail scene model; A commodity layering unit is configured to perform sales layering on the retail commodities according to the daily sales volume, and to deduce the next sales time period of the corresponding sales layer in combination with the daily sales frequency corresponding to each retail commodity, and to set a corresponding overall recommendation order for the corresponding retail commodity according to the next sales time period corresponding to each sales layer; A sampling comparison unit is configured to sample the commodities in each retail channel respectively to obtain the sales volume difference of the same retail commodity corresponding to different retail channels, to obtain the current recommendation order of the same retail commodity corresponding to different retail channels, to take the first current recommendation order as the fixed recommendation order of the same retail commodity, to adjust each current recommendation order by using the sales volume difference, and to obtain the deduced commodity recommendation order corresponding to each same retail commodity; A sequence determination unit is configured to adjust the overall recommendation order by using the deduced commodity recommendation order, to screen unique retail commodities in each retail channel that do not contain the deduced commodity recommendation order, to analyze the commodity similarity between each unique retail commodity and the corresponding same retail commodity by using AI technology, to adjust the overall recommendation order according to the commodity similarity, and to generate the commodity recommendation order corresponding to each retail commodity.

[0030] In this example, the retail scene model represents a model of the scene of the retail channel constructed in a virtual space, and the model can combine abstract scene information with specific sales data; In this example, the single-day sales volume represents the average daily sales volume of a retail product in the previous week, and the single-day sales frequency represents the average daily sales frequency of a retail product in the previous week. In this example, the sales stratification represents the process of clustering retail products according to sales volume or sales frequency. In this example, the next sales time period represents the time period when the user next purchases a retail product in the sales stratification. In this example, the overall recommendation order represents the result of sorting the sales stratification according to the order of the next sales time period. In this example, the product sampling represents the process of sampling retail products that are sold in two or more sales channels at the same time. In this example, the larger the recommendation weight, the higher the recommendation order, for example, hot-selling products are given priority in their high-frequency sales period. In this example, the sales volume difference represents the difference in sales volume of the same retail product in different retail channels. In this example, the current recommendation order represents the recommendation order of the same retail product in different retail channels at the current time. In this example, the derived product recommendation order represents the result of adjusting the recommendation order of the same retail product in each retail channel based on the fixed recommendation order.

[0031] The working principle and beneficial effects of the above technical solution are as follows: a retail scene model is constructed based on retail scene information and real-time retail data, and the single-day sales volume and sales frequency of the product are counted, which can accurately capture the sales situation of the product in different retail channels. Then, the product is stratified according to the single-day sales volume, and the next sales time period is derived based on the sales frequency and the overall recommendation order is set, which can give different sales hotness products the corresponding recommendation weight at the appropriate time, improving the timeliness and pertinence of the recommendation, and helping to improve the conversion rate of the product. Then, the sales volume difference of the same product in different channels is analyzed, and the current recommendation order is adjusted based on the fixed recommendation order to obtain the derived product recommendation order. Finally, the overall recommendation order is adjusted using the derived product recommendation order, and for unique products in each channel, the recommendation order is adjusted by analyzing the similarity of the unique product to similar products using AI technology, ensuring that unique products are not missed. This not only enriches the recommendation content, but also allows users to discover more characteristic products that meet their needs in different channels, improving the comprehensiveness and personalization of the recommendation, and enhancing the user's satisfaction with the recommendation system.

[0032] Embodiment 5: Based on embodiment 4, the scene-aware retail AI intelligent recommendation system comprises: counting a first quantity of the same retail goods and a second quantity of the unique retail goods in each of the retail channels respectively; screening a target retail channel with a first quantity of the same retail goods less than a second quantity of the unique retail goods; obtaining a current recommended sequence of goods corresponding to the target retail channel, inputting the current recommended sequence of goods into the retail scene model for a first time to obtain a first estimated daily sales volume of the target retail channel; when the first estimated daily sales volume is less than an average daily sales volume of the target retail channel, iteratively adjusting the current recommended sequence of goods according to an overall recommended sequence of goods corresponding to the target retail channel, and using the retail scene model for a second time to obtain a second estimated daily sales volume corresponding to each of the iteratively adjusted results; obtaining an optimal iteratively adjusted result with the highest second estimated daily sales volume, and generating a recommended sequence of goods corresponding to each of the retail goods in the target retail channel.

[0033] In this example, the unique retail goods are the goods sold by only one retail channel. In this example, the iteratively adjusting means adjusting a group of the current recommended sequence of goods each time.

[0034] The working principle and beneficial effects of the above technical solution are as follows: the target retail channel with a higher proportion of unique goods is screened by counting the first quantity of the same retail goods and the second quantity of the unique retail goods, so that the retail channel with characteristic goods as the core competitiveness can be accurately locked, and then the current recommended sequence of goods is input into the retail scene model for sales simulation. By comparing the first estimated daily sales volume with the average daily sales volume, the rationality of the current recommended strategy is determined. When the estimated value is low, the overall recommended sequence is iteratively adjusted and repeatedly simulated, and finally the optimal scheme with the highest second estimated daily sales volume is selected, so as to determine the recommended sequence of goods in the target retail channel. The generation of the recommended sequence is more scientific and objective, which can reduce the deviation of human decision-making, quickly find the recommended strategy suitable for the target channel, reduce the operation cost, and improve the overall operation efficiency.

[0035] Embodiment 6: Based on the embodiment 1, the recommendation display module of the scene-aware retail AI intelligent recommendation system comprises: a data acquisition unit configured to determine a starting browsing time of the user in the retail channel according to the real-time retail data, acquire the starting browsing time in real time to obtain the browsing data corresponding to the user in the retail channel. A commodity matching unit is configured to derive real-time display commodity information of the retail channel based on real-time browsing responses of the retail channel to the browsing data, and determine a browsing duration of the user to the real-time display commodity information based on the browsing data; A scene fusion unit is configured to identify a real-time display scene corresponding to each of the real-time display commodity information, analyze interest degrees of the user to each display commodity based on the corresponding browsing duration, and determine an interest commodity of the user in the retail scene; A real-time adjustment unit is configured to set a priority for the interest commodity for preferential display, acquire a plurality of commodity features corresponding to remaining commodities in the retail scene, and set a corresponding recommendation right for each of the remaining commodities based on a feature similarity between each of the remaining commodities and the interest commodity, and perform recommended display.

[0036] The working principle and beneficial effects of the above technical solution are as follows: In order to better recommend commodities to the user, real-time collection is performed starting from a moment when the user starts browsing commodity traffic, the browsing track of the user in the retail channel is completely recorded, the completeness and timeliness of the browsing data are ensured, real-time display commodity information is derived based on real-time responses of the retail channel to the browsing data, the attention degree of the user to the commodity is quantified based on the browsing duration, the scene corresponding to the real-time display commodity is further identified, the interest degree of the user is analyzed based on the browsing duration, the deep fusion of the commodity display scene and the user interest is achieved, the priority is set for the interest commodity and preferential display is performed, and the recommendation right is set based on the feature similarity between the remaining commodities and the interest commodity, and the dynamic optimization of the recommendation order is achieved. In this way, the recommendation range is expanded on the basis of meeting the core needs of the user, the diversity is taken into account, and the effect of the recommended display and the shopping experience of the user are significantly improved.

[0037] Embodiment 7 On the basis of the embodiment 1, the intelligent optimization module of the scene-aware retail AI intelligent recommendation system comprises: A browsing analysis unit is configured to construct a browsing track of the user in the retail channel based on the browsing data, determine a plurality of browsed commodities of the user and a browsing duration corresponding to each of the browsed commodities, and generate a browsing duration feature of the user. A deep analysis unit is configured to generate a browsing preference feature of the user based on browsing operations of the user to each of the browsed commodities, and derive a browsing interest dimension of the user to each retail commodity in the retail channel based on the browsing duration feature and the browsing preference feature. An advantage matching unit is configured to match the product advantage corresponding to each retail product with the browsing interest dimension to obtain a recommended advantage corresponding to each retail product, and render the recommended advantage according to the browsing preference feature to generate and display a product recommendation reason corresponding to each retail product.

[0038] In this example, the browsing trajectory represents the trajectory generated by the user when browsing in the retail channel; In this example, since the user is constantly browsing products, the browsing duration feature is constantly changing with the browsing progress of the user; In this example, the browsing preference feature represents the preference of the user when browsing products, for example, user C likes to browse products related to animation; In this example, the rendering of the display represents the result of beautifying the recommended advantage.

[0039] The working principle and beneficial effects of the above technical solution are as follows: by constructing the user browsing trajectory, determining the browsed products and corresponding browsing duration, and generating the browsing duration feature, the browsing behavior pattern of the user can be comprehensively and meticulously captured, and then the browsing interest dimension of the user is derived by combining the browsing preference feature generated by the browsing operation and the browsing duration feature, which realizes multi-dimensional interpretation of the user's interest, matches the product advantage with the user's browsing interest dimension, selects the recommended advantage that best fits the user's interest, ensures that the recommendation reason hits the user's focus point, and finally renders the recommended advantage according to the browsing preference feature, which can present the recommendation reason in a way that better fits the user's preference. Through such a way, the attractiveness of the recommendation reason is enhanced, which is more likely to impress the user, thereby improving the click-through rate and purchase rate of the product and optimizing the overall recommendation effect.

[0040] Embodiment 8: Based on embodiment 1, the scene-aware retail AI intelligent recommendation system further comprises: When the user purchases a retail product, a remaining quantity to be sold corresponding to the retail channel is generated; According to the remaining quantity to be sold, a remaining sales duration corresponding to the retail product is derived, and a restocking guide is generated and displayed.

[0041] The working principle and beneficial effects of the above technical solution are as follows: the restocking time period of each retail product is analyzed in advance, and relevant personnel are reminded to restock as soon as possible to avoid affecting the sales quantity due to goods.

[0042] Embodiment 9: The embodiment provides a scene-aware retail AI intelligent recommendation method, as shown in Figure 2 The embodiment provides a scene-aware retail AI intelligent recommendation method, as shown in Step 1: Constructing retail scene information provided by each retail channel for users according to real-time retail data uploaded by different retail channels and retail product information corresponding to each retail channel; Step 2: Analyzing the real-time retail data corresponding to the retail scene information to generate a product recommendation order corresponding to the retail channel; Step 3: Obtaining user browsing data in a selected retail channel, determining the user's interest products in combination with the corresponding retail scene information, and adjusting the product recommendation order in real time; Step 4: Deriving the user's browsing time for each retail product according to the browsing data, generating and displaying the corresponding product recommendation reasons in combination with the product advantages corresponding to each retail product.

[0043] In this example, the retail channels include online channels and offline channels, and each channel includes several different channels; In this example, real-time retail data refers to data generated by retail channels during retail work; In this example, retail scene information refers to the purchase scene provided by the retail channel to the user, such as a concise scene, a climate-related scene, a trend-related scene, etc. In this example, browsing data refers to data generated by the user when purchasing in the retail channel, and the selected retail channel refers to the retail channel currently used by the user; In this example, since the browsing time of each user is different, the display method of the product recommendation reason is also different, for example, the browsing time of user A is greater than 1 minute, so all the advantages of the retail product are displayed, and the browsing time of user B is less than 30 seconds, so the outstanding advantages of the retail product are displayed.

[0044] The working principle and beneficial effects of the above technical solution are as follows: in order to accurately recommend goods to users and improve the interest of users in goods, first, a dedicated retail scene is constructed according to real-time data and goods information of different retail channels, so that the system can accurately capture the scene characteristics of each channel, and the goods recommendation is adapted to the channel scene from the source, thereby improving the fitting degree of the recommendation and the scene, then real-time retail data is analyzed based on the retail scene information to generate a goods recommendation sequence, so as to ensure that the recommendation sequence conforms to the operation logic of the channel and the potential needs of users in the scene, thereby avoiding the low efficiency of the general recommendation mode in a specific channel, further combining the browsing data of users in the selected channel and the corresponding scene information, and adjusting the recommendation sequence in real time, so as to quickly capture the interest changes of users in the current scene, dynamically update the recommendation according to the interest of users, improve the attention of users to the recommended goods, and then derive the browsing time of users to the goods, and generate a recommendation reason combining the advantages of the goods, so that the recommendation not only shows the goods, but also accurately delivers the core selling points that move users, enhances the persuasiveness of the recommendation, and improves the purchase willingness of users, finally, the scene is constructed for different retail channels and the recommendation is optimized, so as to avoid the fragmentation of users due to the abruptness of the recommendation style, improve the consistency and smoothness of the cross-channel shopping experience, in this way, the suitable goods can be recommended to users according to their hobbies and actual use, all retail channels can be adapted, the needs of users can be responded in time, and the efficiency of goods sales and the satisfaction of users can be improved.

[0045] Embodiment 10: Based on the embodiment 9, the scene-aware retail AI intelligent recommendation method comprises the following steps: Step 31: constructing a retail scene model corresponding to each retail channel according to the retail scene information and real-time retail data corresponding to each retail channel, and counting the single-day sales volume and single-day sales frequency corresponding to each retail good in the retail scene model; Step 32: performing sales stratification on the retail goods according to the single-day sales volume, and deriving the next sales time period of each sales stratification in combination with the single-day sales frequency corresponding to each retail good, and setting the corresponding overall recommendation sequence for the corresponding retail goods according to the next sales time period corresponding to each sales stratification; Step 33: sampling goods for each retail channel respectively to obtain the sales volume difference of the same retail good in different retail channels, obtaining the current recommendation sequence of the same retail good in different retail channels, taking the first current recommendation sequence as the fixed recommendation sequence of the same retail good, adjusting each current recommendation sequence by using the sales volume difference, and obtaining the derived goods recommendation sequence corresponding to each same retail good; Step 34: adjusting the overall recommendation sequence by using the derived commodity recommendation sequence, and screening unique retail commodities in each retail channel that do not contain the derived commodity recommendation sequence, using AI technology to analyze the commodity similarity between each unique retail commodity and the corresponding same retail commodity, adjusting the overall recommendation sequence according to the commodity similarity, and generating a commodity recommendation sequence corresponding to each retail commodity.

[0046] In this example, the retail scene model represents a model of a scene about a retail channel constructed in a virtual space, which can combine abstract scene information with specific sales data; In this example, the single-day sales volume represents the average daily sales volume of a retail commodity in the previous week, and the single-day sales frequency represents the average daily sales frequency of a retail commodity in the previous week; In this example, the sales stratification represents a process of clustering analysis of retail commodities according to sales volume or sales frequency; In this example, the next sales time period represents the time period for the user to purchase the retail commodity in the sales stratification next time; In this example, the overall recommendation sequence represents the result of sorting the sales stratification according to the sequence of the next sales time period; In this example, commodity sampling represents a process of sampling retail commodities that are sold in two or more sales channels at the same time; In this example, the larger the recommendation weight, the earlier the recommendation sequence, for example, hot-selling commodities are given priority in their high-frequency sales period; In this example, the sales volume difference represents the quantity difference of the sales volume of the same retail commodity in different retail channels; In this example, the current recommendation sequence represents the recommendation sequence of the same retail commodity in different retail channels at the current time; In this example, the derived commodity recommendation sequence represents the result of adjusting the recommendation sequence of the corresponding same retail commodity in each retail channel with the fixed recommendation sequence as the first sequence.

[0047] The working principle and beneficial effects of the technical scheme are as follows: the retail scene model is constructed according to the retail scene information and real-time retail data, and the single-day sales volume and sales frequency of the goods are counted, so that the sales situation of the goods in different retail channels can be accurately captured; then, the goods are divided into different layers according to the single-day sales volume, the next sales time period is deduced in combination with the sales frequency, and the overall recommendation order is set, so that the goods with different sales hotness can obtain corresponding recommendation weights at appropriate times, the timeliness and pertinence of the recommendation are improved, the conversion rate of the goods is improved, the sales volume difference of the same goods in different channels is analyzed, the current recommendation order is adjusted based on the fixed recommendation order, the deduced goods recommendation order is obtained, the overall recommendation order is adjusted by using the deduced goods recommendation order, and the recommendation order of the unique goods of each channel is adjusted by analyzing the similarity of the unique goods and similar goods through AI technology, so that the unique goods are not missed, the recommendation content is enriched, the users can find more characteristic goods meeting the needs in different channels, the comprehensiveness and individuality of the recommendation are improved, and the satisfaction of the users to the recommendation system is improved.

[0048] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A scenario-aware retail AI intelligent recommendation system, characterized in that, include: The scenario training module is used to construct retail scenario information for users for each retail channel based on real-time retail data uploaded from different retail channels and the retail product information corresponding to each retail channel. The product planning module is used to analyze the corresponding real-time retail data using the retail scenario information and generate a product recommendation order corresponding to the retail channel. The recommendation display module is used to obtain the user's browsing data in the selected retail channels, combine it with the corresponding retail scenario information to determine the user's interested products, and adjust the product recommendation order in real time. The intelligent optimization module is used to deduce the user's browsing time for each retail product based on the browsing data, generate corresponding product recommendation reasons based on the advantages of each retail product, and display them.

2. The scene-aware retail AI intelligent recommendation system as described in claim 1, characterized in that, Also includes: The data processing module is used to obtain the product display structure corresponding to each of the retail channels, determine the channel attributes corresponding to each of the retail channels, determine the data collection method corresponding to the retail channel based on the channel attributes, and construct a unified data collection scheme for all the retail channels. The unified data collection scheme is used to collect real-time scene data corresponding to each of the retail channels. When the real-time scene data contains a request field, the real-time scene data is mapped to a preset metadata template to generate the metadata corresponding to the request field. The request location corresponding to the request field is determined based on the metadata, the request field is marked in the corresponding product display structure, and real-time product dynamic data corresponding to the retail channel is generated. Based on the real-time scene data, derive the request result data corresponding to each request location; The real-time product dynamic data and the request result data corresponding to each retail channel are merged to generate real-time retail data for the retail channel and uploaded to the scenario training module.

3. The scene-aware retail AI intelligent recommendation system as described in claim 1, characterized in that, The scene training module includes: The information processing unit is used to acquire the product dynamic data corresponding to each of the retail channels, construct the sales-replenishment process corresponding to different retail products in each of the retail channels in combination with the product display structure corresponding to each of the retail channels, and generate the retail product information of the retail channels. The sales analysis unit is used to match the real-time retail data and the corresponding retail product information for each of the retail channels to obtain several sales events for each retail product, and to obtain the sales timestamp for each sales event. The dynamic analysis unit is used to obtain the sales address corresponding to each of the retail channels, deduce the sales motives of each sales event by combining the specific scenario characteristics corresponding to each sales address, and generate sales dynamic information of each retail channel by combining the corresponding sales timestamp. The scenario construction unit is used to use AI technology to deduce the user's consumption tendency in each of the retail channels based on the sales dynamic information, match the consumption tendency with the corresponding specific scenario features, and construct the retail scenario information of the retail channel using the successfully matched target specific scenario features.

4. The scene-aware retail AI intelligent recommendation system as described in claim 1, characterized in that, The product planning module includes: The modeling and analysis unit is used to construct a retail scenario model corresponding to each retail channel based on the retail scenario information and real-time retail data, and to count the daily sales volume and daily sales frequency of each retail product in the retail scenario model. The product stratification unit is used to stratify the retail products according to the daily sales volume, and deduce the next sales period of the corresponding sales stratification based on the daily sales frequency of each retail product, and set the corresponding overall recommendation order for the corresponding retail products according to the next sales period of each sales stratification. The sampling comparison unit is used to sample products for each of the retail channels, obtain the sales volume difference of the same retail product in different retail channels, obtain the current recommendation order of the same retail product in different retail channels, take the earliest current recommendation order as the fixed recommendation order of the same retail product, adjust each current recommendation order using the sales volume difference, and obtain the deduced product recommendation order corresponding to each of the same retail products. The order determination unit is used to adjust the overall recommendation order using the derived product recommendation order, and to filter out unique retail products in each retail channel that do not contain the derived product recommendation order. It uses AI technology to analyze the product similarity between each unique retail product and the corresponding same retail product, adjusts the overall recommendation order according to the product similarity, and generates a product recommendation order corresponding to each retail product.

5. The scene-aware retail AI intelligent recommendation system as described in claim 4, characterized in that, Also includes: The quantity of the first product corresponding to the same retail product in each of the aforementioned retail channels and the quantity of the second product corresponding to the unique retail product are counted separately. Filter target retail channels where the quantity of the first product is less than the corresponding quantity of the second product; Obtain the current product recommendation order corresponding to the target retail channel, input the current product recommendation order into the corresponding retail scenario model to conduct the first sales simulation, and obtain the first estimated daily sales volume corresponding to the target retail channel; When the first estimated daily sales volume is less than the average daily sales volume of the target retail channel, the current product recommendation order is iteratively adjusted according to the overall recommendation order corresponding to the target retail channel, and a second sales simulation is performed using the retail scenario model to obtain the second estimated daily sales volume corresponding to each iterative adjustment result. Obtain the optimal iterative adjustment result with the highest estimated daily sales volume in the second round, and generate the product recommendation order for each retail product in the target retail channel.

6. The scene-aware retail AI intelligent recommendation system as described in claim 1, characterized in that, The recommendation display module includes: The data acquisition unit is used to determine the start time of the user's browsing of goods in the retail channel based on the real-time retail data, and to collect the start time of the browsing in real time to obtain the browsing data corresponding to the user in the retail channel. The product matching unit is used to obtain the real-time browsing response of the retail channel to the browsing data, deduce the real-time displayed product information of the retail channel, and determine the browsing duration of the user on the real-time displayed product information based on the browsing data; The scene fusion unit is used to identify the real-time display scene corresponding to each of the real-time displayed product information, analyze the user's interest in each displayed product in combination with the corresponding browsing time, and determine the user's interest products in the retail scene. The real-time adjustment unit is used to prioritize the display of the interest products, and at the same time, it acquires several product features corresponding to the remaining products in the retail scenario, sets corresponding recommendation weights for each remaining product based on the feature similarity between each remaining product and the interest products, and recommends and displays them.

7. The scene-aware retail AI intelligent recommendation system as described in claim 1, characterized in that, The intelligent optimization module includes: The browsing analysis unit is used to construct the user's browsing trajectory in the retail channel based on the browsing data, determine several items that the user has browsed and the browsing duration corresponding to each item, and generate the user's browsing duration characteristics. The deep analysis unit is used to obtain the user's browsing operations for each of the browsed products, generate the user's browsing preference features, and use the browsing duration features and the browsing preference features to deduce the user's browsing interest dimension for each retail product in the retail channel. The advantage matching unit is used to match the advantages of each retail product with the browsing interest dimension to obtain the recommended advantages of each retail product, display and render the recommended advantages according to the browsing interest characteristics, generate and display the product recommendation reasons for each retail product.

8. The scene-aware retail AI intelligent recommendation system as described in claim 1, characterized in that, Also includes: After the user purchases retail goods, a remaining unsold quantity corresponding to the retail channel is generated; Based on the remaining unsold quantity, the remaining sales duration corresponding to the retail product is deduced, and a corresponding replenishment guide is generated and displayed.

9. A scenario-aware retail AI intelligent recommendation method, characterized in that, include: Step 1: Construct retail scenario information for users for each retail channel by combining real-time retail data uploaded from different retail channels with the retail product information corresponding to each retail channel; Step 2: Analyze the corresponding real-time retail data using the retail scenario information to generate a product recommendation order for the corresponding retail channel; Step 3: Obtain the user's browsing data in the selected retail channels, combine it with the corresponding retail scenario information to determine the user's interested products, and adjust the product recommendation order in real time; Step 4: Based on the browsing data, deduce the user's browsing time for each retail product, and generate corresponding product recommendation reasons based on the advantages of each retail product, and display them.

10. The scene-aware retail AI intelligent recommendation method as described in claim 9, characterized in that, Step 3 includes: Step 31: Construct a retail scenario model corresponding to each retail channel based on the retail scenario information and real-time retail data, and calculate the daily sales volume and daily sales frequency of each retail product in the retail scenario model; Step 32: Based on the daily sales volume, the retail products are segmented for sales, and the next sales period for the corresponding sales segment is derived by combining the daily sales frequency of each retail product. Based on the next sales period for each sales segment, the corresponding overall recommendation order is set for the retail products. Step 33: Sample products for each of the retail channels to obtain the sales volume difference of the same retail product in different retail channels, obtain the current recommendation order of the same retail product in different retail channels, take the earliest current recommendation order as the fixed recommendation order of the same retail product, adjust each current recommendation order using the sales volume difference, and obtain the deduced product recommendation order for each of the same retail products. Step 34: Adjust the overall recommendation order using the derived product recommendation order, and filter out unique retail products in each retail channel that do not contain the derived product recommendation order. Use AI technology to analyze the product similarity between each unique retail product and the corresponding same retail product, adjust the overall recommendation order based on the product similarity, and generate the product recommendation order corresponding to each retail product.

Images