Commodity recommendation method and system for offline retail scene

Abstract

The invention discloses a commodity recommendation method and system for an offline retail scene, belongs to the technical field of intelligent recommendation, and is used for solving the technical problem that the matching degree between commodities recommended by an existing commodity recommendation scheme and actual demands of consumers is not high. The method comprises the steps of obtaining personal basic data of a consumer; acquiring consumption behavior data of consumers by using an image recognition technology and a target tracking technology, and associating the consumption behavior data with the personal basic data; integrating and generating behavior feature vectors of the consumers according to the associated data; determining a neighbor user corresponding to the consumer based on the behavior feature vector, generating a first recommendation list through consumption data of the neighbor user in a preset time period, extracting interested commodities of the consumer based on the behavior feature vector, and generating a second recommendation list based on commodity attributes of the interested commodities; and according to the first recommendation list and the second recommendation list, generating a recommended commodity list through a preset weight.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation technology, and in particular to a product recommendation method and system for offline retail scenarios. Background Technology

[0002] In the retail industry, with the rapid development of e-commerce platforms and offline digital retail, consumers face a massive number of product choices every day. Traditional product recommendation systems often make recommendations based on general product attributes and sales volume, failing to fully consider the personalized behavioral characteristics of consumers.

[0003] In offline retail scenarios, existing technologies struggle to accurately capture and analyze consumer product selection behavior on shelves, and cannot fully leverage the advantages of dual-screen environments for efficient recommendations. Consumer behavior data, such as picking up, viewing, and comparing products on shelves, contains rich information about their needs; however, the lack of effective image recognition and target tracking technologies to translate this behavioral data into recommendations results in a low match between recommended products and consumers' actual needs, reducing the consumer shopping experience and the merchant's sales efficiency. Summary of the Invention

[0004] This application provides a product recommendation method and system for offline retail scenarios to solve the following technical problem: current product recommendation systems do not recommend products that match consumers' actual needs very well.

[0005] The technical solution adopted in this application is as follows:

[0006] On the one hand, this application provides a product recommendation method for offline retail scenarios. The method includes: acquiring basic personal data of consumers, which includes at least identity attribute data and historical consumption data; collecting consumer consumption behavior data using image recognition technology and target tracking technology, and associating the consumption behavior data with the basic personal data; generating a consumer behavior feature vector based on the associating data; determining the consumer's neighboring users based on the behavior feature vector, and generating a first recommendation list using the neighboring users' consumption data within a preset time period; extracting the consumer's interested products based on the behavior feature vector, and generating a second recommendation list based on the product attributes of the interested products; and generating a recommended product list based on the first recommendation list and the second recommendation list using preset weights.

[0007] In one possible implementation of this application, obtaining a consumer's basic personal data includes: obtaining or generating an identity tag corresponding to the consumer based on the consumer's facial recognition results; obtaining the consumer's identity attribute data and historical consumption data through the identity tag, wherein the identity attribute data includes at least one of gender, age, occupation, and smoking history, and the historical consumption data includes at least one of the following: product name, consumption time, consumption location, consumption quantity, and consumption price.

[0008] In one possible implementation of this application, consumer behavior data is collected using image recognition technology and target tracking technology, including: generating a facial recognition result of the consumer using image recognition technology, and tracking the consumer's facial target using target tracking technology; when the consumer's consumption action is captured, identifying the consumption target corresponding to the consumption action using image recognition technology, and identifying the number of times and the time of occurrence of the consumption action using target tracking technology; and generating consumer behavior data based on the consumption target, the number of times, and the time of occurrence of the consumption action.

[0009] In one possible implementation of this application, associating the consumer behavior data with the personal basic data includes: adding the consumer's identity tag to the consumer behavior data; after adding the identity tag and associating, the method further includes: performing at least one of the following on all data under the identity tag: cleaning, noise reduction, missing value processing, and standardization processing.

[0010] In one possible implementation of this application, generating a consumer behavior feature vector based on the associated data integration includes: extracting a portion of the consumption data corresponding to the consumption target from the historical consumption data, and generating consumption preference data corresponding to the consumption target based on the portion of the consumption data; supplementing the consumption preference data based on the frequency and timing of the consumption actions in the consumption behavior data; and generating the behavior feature vector based on the consumption preference data.

[0011] In one possible implementation of this application, before generating the first recommendation list, the method further includes: using a collaborative filtering algorithm to calculate the similarity between the consumer and other consumers through the consumer's behavioral feature vector, so as to identify neighboring users among the other consumers; and generating the first recommendation list based on the historical consumption data and consumption preference data of the neighboring users.

[0012] In one possible implementation of this application, before generating the second recommendation list, the method further includes: identifying the consumption target in the consumption behavior data as products of interest; extracting the product attributes of the products of interest, and obtaining other products with similar attributes to the product attributes; and generating the second recommendation list through the other products.

[0013] In one possible implementation of this application, after generating the recommended product list, the method further includes: displaying the recommended product list on a checkout display screen and / or a shopping guide display screen, wherein the shopping guide display screen is a shopping guide display screen along the consumer's passageway.

[0014] In one possible implementation of this application, after generating the recommended product list, the method further includes: adjusting the behavioral feature vector, the first recommended list, and the second recommended list based on the consumer's consumption feedback information regarding the recommended product list, wherein the consumption feedback information includes the interaction information between the consumer and the checkout display screen and / or the shopping guide display screen, as well as the consumption information generated by the consumer regarding the recommended product list.

[0015] On the other hand, this application also provides a product recommendation system for offline retail scenarios. The system includes: a data acquisition module for acquiring basic personal data of consumers, the basic personal data including at least identity attribute data and historical consumption data; the data acquisition module for collecting consumer consumption behavior data using image recognition technology and target tracking technology, and associating the consumption behavior data with the basic personal data; a behavior analysis module for generating a consumer behavior feature vector based on the associating data; a product recommendation module for determining the consumer's neighboring users based on the behavior feature vector, generating a first recommendation list based on the neighboring users' consumption data within a preset time period, and extracting the consumer's interested products based on the behavior feature vector, and generating a second recommendation list based on the product attributes of the interested products; and the product recommendation module for generating a recommended product list based on the first recommendation list and the second recommendation list using preset weights.

[0016] This application provides a product recommendation method and system for offline retail scenarios, which has the following beneficial effects:

[0017] By introducing image recognition and target tracking technologies, it is possible to accurately capture consumers' product selection behavior in offline retail scenarios. Compared with traditional product systems, this greatly improves the ability to collect and analyze consumer offline behavior data, more comprehensively grasps consumers' personalized needs and preferences, significantly improves the accuracy and relevance of product recommendations, enhances the matching degree between recommended products and consumers' actual needs, and optimizes consumers' shopping experience.

[0018] By fully leveraging the characteristics of dual-screen environments, differentiated and interconnected recommendation displays can be achieved. Based on consumers' behavioral habits and focus on different screens, more relevant recommendations or products can be provided, enhancing the attractiveness and effectiveness of product recommendations. This enables personalized product recommendations, saves labor costs, increases consumers' purchase opportunities, and ultimately improves merchants' sales conversion rates and consumer loyalty.

[0019] Employing an improved hybrid recommendation algorithm that combines the timeliness and seasonality of products with online and offline consumer behavior data and dual-screen interaction data, it can better adapt to the dynamic changes in the product market and diverse consumption scenarios. Through a feedback optimization module, it continuously collects consumer feedback, constantly optimizing the recommendation algorithm and consumer behavior feature model, enabling the product recommendation solution / system to be adaptive and learning-capable, constantly evolving with consumer needs and market changes, and maintaining the efficiency and accuracy of product recommendations. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0021] Figure 1 A flowchart illustrating a product recommendation method for an offline retail scenario provided in this application;

[0022] Figure 2 This application provides an architecture diagram for a product recommendation system in an offline retail scenario. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0024] This application provides a product recommendation method and system for offline retail scenarios. By introducing image recognition and target tracking technologies, it accurately captures consumers' product selection behavior in offline retail scenarios. Combined with the characteristics of dual-screen environments, it achieves more accurate and personalized product recommendations, effectively improving the matching degree between recommended products and consumers' actual needs, and helping to improve consumer shopping satisfaction and retailer sales performance.

[0025] The method in this application will be described in detail below with reference to the accompanying drawings.

[0026] Figure 1 This application provides a flowchart of a product recommendation method for offline retail scenarios, such as... Figure 1 As shown, the product recommendation method in this application includes at least the following steps:

[0027] Step 101: Obtain the consumer's basic personal data.

[0028] Taking a consumer entering a retail store as an example, after the consumer enters the store, facial data is collected via camera, and facial recognition is performed. Upon successful recognition, the consumer's identity tag is obtained. Then, based on the obtained identity tag, the retail store system extracts the consumer's corresponding identity attribute data and historical consumption data. The identity attribute data includes at least one of gender, age, occupation, and smoking history, while the historical consumption data includes at least one of the following: product name, purchase time, purchase location, purchase quantity, and purchase price.

[0029] Of course, if the consumer's face is identified but no corresponding identity tag is found in the retail store system during the above process, a new identity tag will be generated for the consumer.

[0030] Step 102: Collect consumer behavior data using image recognition and target tracking technologies, and link the consumer behavior data with personal basic data.

[0031] After recognizing a consumer's face using image recognition technology, target tracking technology is then used to track that face, essentially tracking the consumer's movement within the retail store. When a consumer makes a purchase action—meaning they pick up an item from the shelf—detected by cameras or other devices, image recognition technology identifies the item corresponding to that action, determining the purchase target—the item the consumer wants to buy or is interested in. Then, using that item as the target, target tracking technology is used to identify the frequency and timing of each purchase action.

[0032] Furthermore, based on the recognition and tracking results of image recognition technology and target tracking technology, the consumption target, frequency, and time of occurrence corresponding to the consumption action are combined to generate consumption behavior data.

[0033] Add the consumer's corresponding identity tag to the generated consumer behavior data, that is, complete the association between the consumer's basic personal data and consumer behavior data, and use the data under the identity tag as the initial data to be collected.

[0034] In one possible implementation of this application, to facilitate subsequent product recommendation analysis, the initially collected data needs to be preprocessed. This preprocessing includes at least one of the following: cleaning, denoising, missing value handling, and standardization. Specifically, cleaning removes duplicate and erroneous data records; denoising eliminates abnormal fluctuations in data; for missing values, methods such as mean imputation and regression prediction are used to complete them; and standardization converts data of different types and ranges into a unified format and scale for subsequent analysis.

[0035] Those skilled in the art will recognize that the image recognition and target tracking technologies involved in the aforementioned schemes can be implemented using existing algorithms, and will not be elaborated upon here.

[0036] Step 103: Generate consumer behavior feature vectors based on the correlated data integration.

[0037] The preprocessed data is then subjected to in-depth analysis to construct a consumer behavior characteristic model, resulting in a behavior feature vector. This vector serves as the basis for generating the subsequent product recommendation list. Specifically, based on purchase history data, the model analyzes consumers' purchase frequency, purchase cycle, and purchase preferences (such as preferred brands, categories, and price ranges). Simultaneously, offline shopping path data and product picking data are utilized, combined with image recognition and target tracking technology analysis results, to further dissect consumers' shopping habits and potential interests. For example, by analyzing the time consumers spend in front of a certain type of product shelf, the order and frequency of product picking, the model determines their preference for different brands or specifications of that product category. Finally, the above analysis results are integrated into a consumer behavior feature vector, used to represent consumers' personalized behavioral characteristics, i.e., their consumption preferences for products.

[0038] Step 104: Determine the neighboring users corresponding to the consumer based on the behavioral feature vector, and generate the first recommendation list based on the consumption data of the neighboring users within a preset time period.

[0039] In this application, a collaborative filtering algorithm is used to calculate the similarity between the current consumer and other consumers based on the aforementioned determined consumer behavior feature vector. Based on the similarity calculation results, neighboring users are identified among other consumers. Then, the historical consumption data and consumption preference data of the neighboring users are obtained, and a first recommendation list is generated based on this.

[0040] It should be clarified that the first recommendation list generated here can be a recommendation list containing multiple categories of products generated based on the data of neighboring users, or it can be a recommendation list of a single category of products corresponding to the current consumer's consumption goals generated based on the data of neighboring users. When generating the list, it is possible to take into account the different types of products consumed by the users presenting the argument.

[0041] It should also be clarified that the implementation process of the aforementioned collaborative filtering algorithm is a known process, and will not be described in detail in this embodiment.

[0042] Step 105: Extract the products of interest to consumers based on behavioral feature vectors, and generate a second recommendation list based on the product attributes of the products of interest.

[0043] The "product of interest" here refers to the consumer's purchase target corresponding to the aforementioned steps, i.e., the product the consumer picks up from the shelf, indicating that the consumer intends to buy that product. At this point, the product attributes of the product of interest are obtained, such as product name, ingredients, efficacy, target audience, brand, etc. Then, other products with similar attributes are retrieved from the retail store system, and a second recommendation list is generated based on these other products.

[0044] Step 106: Based on the first recommendation list and the second recommendation list, generate a list of recommended products using preset weights.

[0045] After obtaining the first and second recommendation lists, the two recommendation lists are integrated through a weighted fusion method, and the weights are dynamically adjusted according to different scenarios and consumer needs to obtain the final list of recommended products.

[0046] In one possible implementation of this application, after obtaining the final list of recommended products, effective display is needed to increase consumer purchases. Therefore, this can be achieved on a smart checkout display screen. When consumers check out at a dual-screen machine or self-checkout machine, interactive recommendations are made based on recommendation records and already purchased items. This is especially important for easily accessible items placed near the checkout area. Simultaneously, different display methods are used based on consumers' screen interaction habits. For example, for consumers who frequently use touchscreens, more clickable entry points to detailed product information are provided. Alternatively, fixed shopping guide displays within the supermarket can be used. These displays, located in frequently used aisles or rest areas, showcase comprehensive recommended products based on overall consumer behavior analysis, as well as information on popular promotions. The shopping guide displays use large images, videos, and other visually impactful formats to attract consumer attention. Furthermore, by linking with the smart shopping cart display screen, when consumers approach the screen, a brief overview of the highlighted recommended products is simultaneously displayed on the shopping cart screen, facilitating comparison and viewing.

[0047] Furthermore, after consumers generate consumption data based on the displayed product recommendation list, they can also provide positive feedback based on this data. For example, the system can collect consumer feedback on the recommendation results, including but not limited to whether they clicked on recommended products, whether they purchased them, and their reviews. Based on this feedback, the weighting of the first and second recommendation lists when generating the final product recommendation list can be adjusted, and / or the process of generating consumer behavior feature vectors can be optimized to continuously improve the accuracy and effectiveness of product recommendations, thereby enhancing the match between recommended products and consumers' actual needs.

[0048] Example 1

[0049] Based on the above description, the product recommendation method proposed in this embodiment includes:

[0050] Data Acquisition: In a partnered offline supermarket, high-definition cameras are installed every 3 meters above the shelves, and cameras are also deployed in the main aisles and corners of the supermarket, creating a comprehensive image acquisition network. When a consumer enters the supermarket, the cameras use facial recognition technology to identify the consumer and establish a unique identifier. When a consumer selects goods, such as picking up a brand of milk from the milk shelf, the camera captures this action, image recognition technology identifies the product name, and target tracking technology records the time and number of times the milk is picked up. This behavioral data is then associated with the consumer's identity and transmitted to the data acquisition module. Simultaneously, the smart shopping cart displays and fixed guide displays within the supermarket collect real-time data on consumer touch operations, gaze placement, and other interactive data. At the retailer's POS platform, the data acquisition module connects to the platform's database via an API interface to obtain real-time data such as the consumer's purchase history, browsing history, and search keywords.

[0051] Data preprocessing: The collected raw data is transmitted to the data preprocessing module to clean the data, remove duplicate behavior records and erroneous timestamps; the mean imputation method is used to complete the missing product evaluation data; and different types of data are standardized to conform to a unified format.

[0052] Behavioral Analysis: The behavioral analysis module analyzes the preprocessed data. For example, if Consumer B repeatedly picks up and examines a certain brand of yogurt in the store but doesn't purchase it, combined with their high purchase frequency of yogurt in their online shopping history, the analysis suggests they may be looking for a yogurt product that better suits their needs. Simultaneously, the analysis shows that Consumer B frequently clicks to view yogurt product details on the smart shopping cart display and pays more attention to promotional information when lingering in front of the sales guide display, further refining their behavioral characteristic model.

[0053] Recommendation Algorithm: The recommendation algorithm module employs a hybrid recommendation algorithm. In the collaborative filtering algorithm, based on consumer B's behavioral feature vector, 10 neighboring users with high similarity to them are found. Considering the current summer season, purchase behavior related to cold drinks is given higher weight. Simultaneously, due to consumer B's specific offline behavior regarding yogurt, the weight of yogurt-related products is increased. Recent purchases of yogurt and related drinks by neighboring users are used as initial recommendations. In the content-based recommendation algorithm, the yogurt category attributes that consumer B is interested in are analyzed, recommending other brands of yogurt with similar ingredients and effects, prioritizing products with good display effects and interactive feedback on the dual-screen device. The recommendation results from the two algorithms are then fused with weights of 0.6 (collaborative filtering algorithm) and 0.4 (content-based recommendation algorithm) to obtain the final recommended product list.

[0054] Recommended Display: Recommended products are displayed in a dual-screen environment. When consumer B proceeds to checkout, the shopping cart screen displays recommended yogurt products in real time, including images, names, flavor descriptions, and current promotional offers. Simultaneously, a nearby shopping guide screen displays popular recommended yogurt products and promotional videos in large image format. Consumer B can click to view detailed information on the shopping cart screen and obtain a more intuitive visual experience on the shopping guide screen, achieving a dual-screen interactive recommendation system.

[0055] Feedback optimization: Collect feedback from consumer B on recommended products. If consumer B clicks on and purchases a recommended brand of yogurt, record this information in the feedback optimization module and appropriately increase the weight of the content-based recommendation algorithm in the hybrid recommendation algorithm. At the same time, further optimize the behavioral feature model of consumer B.

[0056] Based on the same inventive concept, this application also provides a product recommendation system for offline retail scenarios, the architecture of which is as follows: Figure 2 As shown.

[0057] Figure 2 This application provides an architecture diagram for a product recommendation system in an offline retail scenario. (For example...) Figure 2 As shown, the product recommendation system 20 for offline retail scenarios in this application specifically includes:

[0058] Data acquisition module 201 acquires basic personal data of consumers, including at least identity attribute data and historical consumption data;

[0059] The data acquisition module 201 uses image recognition technology and target tracking technology to collect consumer behavior data and associates the consumer behavior data with the personal basic data;

[0060] Behavior analysis module 202 generates consumer behavior feature vectors based on the correlated data.

[0061] The product recommendation module 203 determines the neighboring users corresponding to the consumer based on the behavioral feature vector, generates a first recommendation list based on the consumption data of the neighboring users within a preset time period, and extracts the products of interest of the consumer based on the behavioral feature vector, and generates a second recommendation list based on the product attributes of the products of interest.

[0062] The product recommendation module generates a list of recommended products based on the first recommendation list and the second recommendation list, using preset weights.

[0063] In one possible implementation of this application, the aforementioned product recommendation system may further include:

[0064] The recommendation display module 204 displays the list of recommended products on the checkout display screen and / or the shopping guide display screen, wherein the shopping guide display screen is the shopping guide display screen along the consumer's passageway;

[0065] The feedback optimization module 205 adjusts the behavioral feature vector, the first recommendation list, and the second recommendation list based on the consumer's consumption feedback information regarding the recommended product list. The consumption feedback information includes the interaction information between the consumer and the checkout display screen and / or the shopping guide display screen, as well as the consumption information generated by the consumer regarding the recommended product list.

[0066] Of course, the product recommendation system in this application can also add or remove modules according to actual needs, and can also be configured to include other modules, or different forms of other modules. For example, the product recommendation system can also be configured to include:

[0067] The data acquisition module consists of two parts: First, based on the retail POS platform, it continuously collects consumers' purchase history data (purchased product name, purchase time, purchase quantity, purchase price, etc.), consumer characteristic attribute data (gender, age, occupation, smoking history, etc.), and geographical location LBS data (store, time, etc.). Second, in offline retail scenarios, it utilizes high-definition cameras installed above shelves, in aisle corners, etc., combined with image recognition and target tracking technologies to collect consumers' behavioral data in real time. Specifically, image recognition technology identifies consumers' physical features and establishes consumer identification; target tracking technology continuously tracks consumers' movement trajectory between shelves, their actions of picking up products, and their target products. When a consumer picks up a fast-moving consumer good, the camera captures the action, and the system records data such as the product name, the time of picking up, and the number of times the product is picked up; simultaneously, the consumer's identity is confirmed through the consumer membership system, and this behavioral data is integrated with the consumer's online behavioral data and other offline behavioral data.

[0068] Data preprocessing module: Receives raw behavioral data from the data acquisition module and performs data cleaning, denoising, missing value handling, and standardization. Cleaning removes duplicate and erroneous data records; denoising eliminates abnormal fluctuations; for missing values, methods such as mean imputation and regression prediction are used to complete them; standardization converts data of different types and ranges into a unified format and scale for subsequent analysis.

[0069] The behavior analysis module performs in-depth analysis of preprocessed data to build a consumer behavior characteristic model. First, based on purchase history data, it analyzes consumers' purchase frequency, purchase cycle, and purchase preferences (such as preferred brands, categories, and price ranges). Second, utilizing offline shopping path data and product picking data, combined with image recognition and target tracking technology analysis results, it further dissects consumers' shopping habits and potential interests. For example, by analyzing the time consumers spend in front of a certain type of product shelf, the order and frequency of product picking, it determines their preference for different brands or specifications of that type of product. Finally, the above analysis results are integrated into a consumer behavior feature vector to represent consumers' personalized behavioral characteristics.

[0070] The recommendation algorithm module employs a hybrid recommendation algorithm combining a modified collaborative filtering algorithm and a content-based recommendation algorithm. The collaborative filtering algorithm calculates the similarity between consumers based on their behavioral feature vectors, identifying user groups (neighboring users) with high similarity to the target consumer. Simultaneously, it considers the timeliness and seasonality of fast-moving consumer goods (FMCG), as well as consumer behavior data in offline scenarios, assigning different weights to consumption behaviors in different time periods and behavioral types. For example, in summer, higher weight is given to the purchase of cold drinks; for FMCG in specific scenarios such as the rainy season, weight is increased in rainy weather; and for product types that consumers frequently pick up but do not purchase offline, the weight of related products is appropriately increased during recommendation. Based on the purchase history and preferences of neighboring users, a preliminary list of recommended products is generated for the target consumer, which is defined as the recall layer (coarse screening). The content-based recommendation algorithm analyzes the attribute data of FMCG (such as product name, ingredients, efficacy, target audience, brand, etc.), matching product attributes with consumer behavioral characteristics. For product categories and attributes that consumers show interest in, other FMCG with similar attributes are recommended. Simultaneously, by combining consumer interaction data in a dual-screen environment, products that display well on the corresponding screen and conform to consumer operating habits are prioritized for recommendation; this step is designated as the ranking layer (fine ranking). Finally, the recommendation results of the two algorithms are integrated through a weighted fusion method, dynamically adjusting the weights according to different scenarios and consumer needs to obtain the final list of recommended products.

[0071] Recommendation Display Module: In dual-screen kiosks and self-checkout display environments, differentiated recommendations are displayed based on consumer behavior data and recommendation algorithm results, defined as re-ranking (enhanced diversity). Preferably, the smart checkout display: When consumers check out at dual-screen kiosks or self-checkout machines, interactive recommendations are made based on recommendation records and already checked-out items. Especially recommended are easily accessible items placed near the checkout area. Simultaneously, different display methods are adopted based on consumers' screen interaction habits; for example, for consumers who frequently use touchscreens, more clickable entry points to detailed product information are provided. Fixed shopping guide displays within the supermarket: These displays, located in frequently used aisles or rest areas, showcase comprehensive product recommendations based on overall consumer behavior analysis, as well as information on popular promotions. The shopping guide displays use large images, videos, and other visually impactful formats to attract consumer attention. Through linkage with the smart shopping cart display, when consumers approach the display, brief information on the highlighted recommended items is simultaneously displayed on the shopping cart screen for easy comparison.

[0072] Feedback Optimization Module: This module collects consumer feedback on the recommendation results, including but not limited to whether they clicked on recommended products, whether they purchased them, and their reviews. Based on this feedback, it adjusts the weights of the hybrid recommendation algorithm in the recommendation algorithm module, optimizes the consumer behavior feature model, and continuously improves the accuracy and effectiveness of the recommendation system.

[0073] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0074] The equipment and method provided in this application are one-to-one correspondences. Therefore, the equipment also has similar beneficial technical effects as its corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the equipment will not be repeated here.

[0075] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0076] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0077] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A product recommendation method for offline retail scenarios, characterized in that, The method includes: Obtain consumers' basic personal data, which includes at least identity attribute data and historical consumption data; Consumer behavior data is collected using image recognition and target tracking technologies, and then the consumer behavior data is associated with the individual's basic data; Generate consumer behavior feature vectors based on the correlated data integration; Based on the behavioral feature vector, the consumer's neighboring users are determined, and a first recommendation list is generated using the neighboring users' consumption data within a preset time period. Also, based on the behavioral feature vector, the consumer's interested products are extracted, and a second recommendation list is generated based on the product attributes of the interested products. Based on the first recommendation list and the second recommendation list, a list of recommended products is generated using preset weights.

2. The product recommendation method for offline retail scenarios according to claim 1, characterized in that, Obtain consumers' basic personal data, including: Based on the consumer's facial recognition results, obtain or generate the consumer's corresponding identity tag; The identity attribute data and historical consumption data of the consumer are obtained through the identity tag. The identity attribute data includes at least one of gender, age, occupation and smoking history, and the historical consumption data includes at least one of the following: product name, consumption time, consumption location, consumption quantity and consumption price.

3. The product recommendation method for offline retail scenarios according to claim 2, characterized in that, Consumer behavior data is collected using image recognition and target tracking technologies, including: The face recognition result of the consumer is generated using image recognition technology, and the face target of the consumer is tracked using target tracking technology; When the consumer's consumption action is captured, the image recognition technology is used to identify the consumption target corresponding to the consumption action, and the target tracking technology is used to identify the number of times the consumption action occurs and the time of occurrence. Consumer behavior data is generated by analyzing the consumption goals, frequency, and timing of the consumption actions.

4. The product recommendation method for offline retail scenarios according to claim 3, characterized in that, Linking the consumption behavior data with the personal basic data includes: Add the consumer's identity tag to the consumption behavior data; After adding identity tags for association, the method further includes: For all data under the aforementioned identity label, perform at least one of the following processes: cleaning, noise reduction, missing value handling, and standardization.

5. The product recommendation method for offline retail scenarios according to claim 1, characterized in that, Based on the integrated data after association, a consumer behavior feature vector is generated, including: In the historical consumption data, a portion of the consumption data corresponding to the consumption target is extracted from the consumption behavior data, and consumption preference data corresponding to the consumption target is generated based on the portion of the consumption data; The consumption preference data is supplemented based on the frequency and timing of consumption actions in the consumption behavior data. The behavioral feature vector is generated based on the consumer preference data.

6. The product recommendation method for offline retail scenarios according to claim 1, characterized in that, Before generating the first recommendation list, the method further includes: Using a collaborative filtering algorithm, the similarity between the consumer and other consumers is calculated based on the consumer's behavioral feature vector, in order to identify neighboring users among the other consumers; The first recommendation list is generated based on the historical consumption data and consumption preference data of the neighboring users.

7. The product recommendation method for offline retail scenarios according to claim 5, characterized in that, Before generating the second recommendation list, the method further includes: The consumption goals in the aforementioned consumer behavior data are identified as products of interest. Extract the product attributes of the product of interest, and obtain other products with similar attributes to the product attributes; The second recommendation list is generated using the other products.

8. The product recommendation method for offline retail scenarios according to claim 1, characterized in that, After generating the recommended product list, the method further includes: The recommended product list is displayed on the checkout display screen and / or the shopping guide display screen, wherein the shopping guide display screen is the shopping guide display screen along the consumer's passageway.

9. The product recommendation method for offline retail scenarios according to claim 8, characterized in that, After generating the recommended product list, the method further includes: Based on the consumer's consumption feedback information regarding the recommended product list, the behavioral feature vector, the first recommendation list, and the second recommendation list are adjusted. The consumption feedback information includes the consumer's interaction information with the checkout display screen and / or the shopping guide display screen, as well as the consumer's consumption information regarding the recommended product list.

10. A product recommendation system for offline retail scenarios, characterized in that, The system includes: The data acquisition module obtains consumers' basic personal data, which includes at least identity attribute data and historical consumption data. The data acquisition module uses image recognition technology and target tracking technology to collect consumer behavior data and associates the consumer behavior data with the personal basic data; The behavior analysis module generates consumer behavior feature vectors based on the correlated data. The product recommendation module determines the neighboring users corresponding to the consumer based on the behavioral feature vector, generates a first recommendation list based on the consumption data of the neighboring users within a preset time period, and extracts the products of interest of the consumer based on the behavioral feature vector, and generates a second recommendation list based on the product attributes of the products of interest. The product recommendation module generates a list of recommended products based on the first recommendation list and the second recommendation list, using preset weights.

Images