Articulated naturality web client intelligent analysis and management method based on artificial intelligence technology
By constructing a customer behavior map based on multimodal data, abnormal areas in retail stores can be identified and customized strategies can be triggered. This solves the problems of accuracy in customer behavior analysis and lack of targeted management strategies in the traditional retail industry, and achieves efficient intelligent management and operational improvement.
Patent Information
- Application Number
- CN202610052825.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-15
- Publication Date
- 2026-04-28
AI Technical Summary
Traditional retail customer behavior analysis lacks multimodal data fusion, cannot dynamically reflect changes in customer interests, lacks targeted management strategies, and existing technologies are difficult to adapt to dynamic differences in different time periods and scenarios.
By integrating multimodal data such as customer movement trajectories, dwell time, and micro-expressions, a customer behavior map with temporal correlation is constructed to identify abnormal areas. Differentiated management strategies are triggered through preset anomaly judgment rules, forming an intelligent management closed loop.
It enables progressive analysis from basic behavioral tags to advanced behavioral patterns, improving the accuracy and operational efficiency of customer behavior analysis, increasing conversion rates, and supporting data-driven management model upgrades.
Smart Images
Figure CN121937159A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of customer analysis and management technology, and relates to a visual network customer intelligent analysis and management method based on artificial intelligence technology. Background Technology
[0002] Traditional retail customer behavior analysis and regional management largely rely on manual observation, basic video surveillance, or discrete data statistics, which has significant limitations. First, existing technologies capture customer behavior from a single dimension, lacking the fusion and analysis of multimodal data such as movement trajectories, dwell time, and micro-expressions, thus failing to dynamically reflect changes in customer interests. Second, existing technologies typically rely on preset static thresholds for judgment, making it difficult to adapt to dynamic differences in different time periods and scenarios, and unable to accurately attribute the causes of anomalies, resulting in a lack of targeted management strategies. Summary of the Invention
[0003] In view of this, in order to solve the problems mentioned in the background technology, a visual network customer intelligent analysis and management method based on artificial intelligence technology is proposed.
[0004] The objective of this invention can be achieved through the following technical solution: This invention provides a video network customer intelligent analysis and management method based on artificial intelligence technology, including: Step 1, extracting customer behavior information, customer transaction information and shopping environment information based on multi-dimensional video stream data of target shopping stores within a set monitoring time period.
[0005] Step 2: Based on the customer behavior information, determine the structured behavior tags and dynamic interest tags of all customers within the set monitoring period, and construct a customer behavior map with time-series correlation accordingly.
[0006] Step 3: Identify abnormal areas in the target retail store based on customer behavior mapping, including high-traffic, low-conversion areas and high-touch, low-conversion areas.
[0007] Step 4: Analyze the causes of abnormal areas according to the preset abnormal judgment rules, and trigger differentiated management strategies to achieve intelligent management.
[0008] Compared with existing technologies, the beneficial effects of the present invention are as follows: 1. By integrating multimodal data such as customer movement trajectory, dwell time, and micro-expressions, the present invention constructs a customer behavior map with temporal correlation, realizing progressive analysis from basic behavior tags to higher-order behavior patterns, and achieving a leap from "passive recording" to "active understanding" of customer needs, providing underlying data support for precise operation, and significantly improving the granularity and accuracy of customer behavior analysis.
[0009] 2. This invention identifies high-traffic, low-conversion areas and high-touch, low-conversion areas in real time based on customer behavior mapping, and accurately locates the causes of anomalies based on multi-dimensional data analysis through preset anomaly judgment rules, realizing automated analysis from "discovering problems" to "locating causes".
[0010] 3. This invention automatically triggers customized strategies for different causes of anomalies, forming an intelligent management closed loop of "monitoring-analysis-execution-feedback", which effectively improves operational efficiency and conversion rate, and provides data accumulation for long-term operation strategies, supports periodic display adjustments and product selection optimization, and realizes the upgrade from "experience-driven" to "data-driven" management model. Attached Figure Description
[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 This is a schematic diagram illustrating the implementation steps of the method of the present invention. Detailed Implementation
[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0014] Please see Figure 1 As shown, the present invention provides a video network customer intelligent analysis and management method based on artificial intelligence technology. The specific steps are as follows: Step 1: Extract customer behavior information, customer transaction information and shopping environment information based on the multi-dimensional video stream data of the target shopping store within the set monitoring time period.
[0015] It should be noted that the set monitoring time period can be the daily operating hours of the target shopping store.
[0016] In a specific example, the customer behavior information includes the movement trajectory of all customers during their movement in the merchandise display area, the types of shelves they stopped at (e.g., "Shelf A - Snack Area" and "Shelf B - Daily Necessities Area"), the duration of stay at each shelf (e.g., 2 minutes in the snack area), the product categories touched by each shelf (associated with the product category by the location of hand contact, e.g., "touched potato chips and chocolate in the snack area"), the frequency of contact with each product, and the micro-expressions when stopping at shelves and touching products. The customer transaction information includes the purchase volume of each shelf area, and the shopping environment information includes the customer traffic of each shelf area.
[0017] One specific example is that the micro-expressions include, but are not limited to: (1) positive: happiness, surprise, astonishment, etc.; (2) neutral: no obvious emotion, hesitation, etc.; (3) negative: frowning, shaking head, disgust, etc.
[0018] It should be further explained that the specific methods for obtaining the customer behavior information, customer transaction information, and shopping environment information are as follows: A pose estimation model is used to capture the customer's movement trajectory in the merchandise display area; a time-series analysis algorithm is combined to calculate the dwell time on each shelf; a micro-expression recognition model is used to extract facial muscle motion unit codes; and simultaneously, target detection technology is used to statistically analyze the frequency and duration of merchandise contact actions; POS system data is integrated to obtain merchandise purchase records, thus obtaining the purchase volume for each shelf area; and a crowd density estimation algorithm is used to statistically analyze real-time customer flow, thereby obtaining the customer flow for each shelf area. The above methods are relatively mature in existing technologies and will not be elaborated upon here.
[0019] Step 2: Based on the customer behavior information, determine the structured behavior tags and dynamic interest tags of all customers within the set monitoring period, and construct a customer behavior map with time-series correlation accordingly.
[0020] As a preferred feasible embodiment, the behavioral tags are specifically: quick browsing, in-depth selection, hesitant comparison, goal-oriented, and random browsing.
[0021] It should be explained that the "quick browsing" refers to a short dwell time (e.g., <30 seconds) + no contact behavior; the "deep selection" refers to a long dwell time (e.g., >2 minutes) + contact with the product (≥3 times); the "hesitant comparison" refers to a long dwell time (e.g., >3 minutes) + multiple contact with the same type of product (e.g., take-put-take-take again) + micro-expression showing "hesitation"; the "goal-oriented" refers to quickly moving to a specific shelf + direct contact with the target product + short dwell time (e.g., <1 minute); and the "random wandering" refers to moving without a fixed route + frequent switching of areas + few contact behaviors (<3 times).
[0022] The specific process for determining the structured behavioral tags of all customers within the set monitoring period includes: A1. Using customer ID as the unit, and according to the time sequence of their behavior during their movement in the merchandise display area, the behavioral tags corresponding to each action are chained together to generate a behavioral tag sequence.
[0023] In a specific example, the sequence of actions of customer ID C003 during movement in the merchandise display area is as follows: entering the snack area (10:05) → stopping to observe (10:05-10:07) → touching and selecting potato chips (10:07-10:09) → moving to the beverage area (10:09) → deeply selecting juice (10:09-10:12) → leaving the beverage area (10:12). Thus, the behavior label sequence of customer ID C003 is: [entering the area, stopping to observe, touching and selecting, moving to switch, deeply selecting, leaving the area].
[0024] A2. If the same behavior tag appears consecutively for a customer (such as "contact to select" multiple times in a row), it will be merged into one behavior tag and the duration will be recorded.
[0025] In a specific example, if a customer continuously touches different varieties of potato chips in the snack section for 5 minutes, then this behavior is tagged as: Contact and selection (lasting 5 minutes).
[0026] A3. Based on the combination of tags in the behavior tag sequence and in conjunction with predefined behavior pattern rules, determine higher-order behavior patterns.
[0027] In a specific example, the predefined behavior pattern rules include, but are not limited to: if the behavior pattern [entering an area → staying to observe → moving to switch] occurs, then the behavior pattern is marked as "initial browsing, no interest generated".
[0028] If the following behavior pattern occurs: [Entering the area → Touching to select → Hesitating to compare → Moving to switch → Re-entering the original area → Touching to select], then mark the behavior pattern as "Repeatedly comparing, may return to purchase".
[0029] A4. Based on the customer's behavioral tag sequence, associate it with its higher-level behavioral patterns, collectively referred to as the customer's structured behavioral tags, and then obtain the structured behavioral tags of all customers within the set monitoring period.
[0030] As a preferred feasible embodiment, the dynamic interest tags are specifically categorized as no clear interest, slight interest, moderate preference, and high interest.
[0031] It should be explained that "no clear interest" means quick browsing without stopping (stopping for <1 minute + visual browsing only (no contact)); "light attention" means initial interest without in-depth interaction (stopping for 1-3 minutes + visual browsing only (no contact)); "moderate preference" means active exploration with clear interaction (stopping for 3-5 minutes + contact ≥1 time + no purchase); and "high interest" means in-depth interaction, close to a purchase decision (stopping for >5 minutes + multiple contact (≥3 times) + micro-expression "interested").
[0032] The specific process for determining the dynamic interest tags of all customers within the set monitoring period includes: obtaining the interest scores of each shelf category that all customers stopped at and each product category that they came into contact with during their movement in the merchandise display area based on the customer interest tag determination logic, and performing preliminary screening to determine the shelf categories and product categories of interest for all customers during their movement in the merchandise display area.
[0033] As a preferred feasible embodiment, the customer interest tag determination logic specifically includes: extracting from customer behavior information the duration of each customer's stay on each shelf during their movement in the merchandise display area, the frequency of contact with each product, and the micro-expressions when staying on the shelf and touching the product.
[0034] By comparing the dwell time of each shelf with the dwell time-interest tag mapping table stored in the database, the interest tags corresponding to the dwell time of each customer on each shelf during the movement of all customers in the merchandise display area can be obtained. Similarly, the contact frequency of each product touched by each customer during the movement of all customers in the merchandise display area, as well as the interest tags corresponding to the micro-expressions when stopping at the shelf and touching the product, can be obtained.
[0035] As a specific example, the dwell time-interest tag mapping is represented as shown in Table 1.
[0036] Table 1. Examples of Dwell Time-Interest Tag Mapping
[0037] It needs further explanation that the specific process of obtaining the interest scores of each shelf category that all customers stopped at and each product category that they came into contact with during their movement in the merchandise display area based on the customer interest tag determination logic is as follows: The interest scores corresponding to no clear interest, slight attention, moderate preference, and high interest are recorded as 0 points, 1 point, 2 points, and 3 points, respectively. Based on this, the interest tags corresponding to the duration of each customer's stay at each shelf and the interest scores corresponding to the micro-expressions during the stay at each shelf can be obtained. The interest scores corresponding to the duration of each customer's stay at each shelf and the interest scores corresponding to the micro-expressions during the stay are added together, and the result is recorded as the total interest score. The total interest scores of the same category of shelves are then accumulated to obtain the interest scores of each shelf category that all customers stopped at during their movement in the merchandise display area. Similarly, the interest scores of each product category that all customers came into contact with during their movement in the merchandise display area can be obtained.
[0038] As a customer moves through the merchandise display area, the categories of shelves of interest are sorted from highest to lowest according to interest rating. The top three categories of shelves of interest are selected as the customer's main categories of interest during the movement of the merchandise display area.
[0039] Similarly, we can obtain the main interest shelf categories and main interest product categories of all customers during their movement in the merchandise display area, and record them as the dynamic interest tags of all customers within the set monitoring period.
[0040] A specific example, customer ID C003: (1) Behavioral data: if the customer stays in the beverage area for 50 seconds (<60 seconds, no basic attention is triggered) and has no contact behavior, then the dynamic interest label of the customer in the beverage area is no clear interest; if the customer stays in the daily necessities area for 90 seconds (60 seconds ≤ duration < 120 seconds) and has no contact behavior, and has no obvious change in micro-expression, then the dynamic interest label of the customer in the beverage area is slight attention.
[0041] (2) Interest rating: Beverage area: 50 seconds stay → no clear interest; Daily necessities area: 90 seconds stay → slight attention (base score 1 point), no contact behavior → no clear interest, total score 1 point.
[0042] (3) Dynamic interest tags: Main interest shelf category: Daily necessities area (light interest); Main product categories of interest: None.
[0043] As a preferred feasible embodiment, the specific construction process of the customer behavior map with temporal correlation includes: aligning the behavioral tag sequence of each customer with the dynamic interest tag changes along the time axis to form a "behavioral stage-interest point" correlation graph, which is recorded as a customer behavior map with temporal correlation.
[0044] This invention integrates multimodal data such as customer movement trajectories, dwell time, and micro-expressions to construct a customer behavior map with temporal correlation. It achieves progressive analysis from basic behavior tags to higher-order behavior patterns, and realizes the leap from "passive recording" to "proactive understanding" of customer needs. This provides underlying data support for precise operations and significantly improves the granularity and accuracy of customer behavior analysis.
[0045] Step 3: Identify abnormal areas in the target retail store based on customer behavior mapping, including high-traffic, low-conversion areas and high-touch, low-conversion areas.
[0046] As a preferred feasible embodiment, the high-traffic, low-conversion area specifically refers to the shelf area within the target retail store where customer traffic (number of people passing by or staying) is high but the actual purchase volume is low.
[0047] The high-touch, low-conversion area specifically refers to the shelf areas within the target retail store where customers frequently interact with products (e.g., touch, try, repeatedly examine) but the actual purchase rate is relatively low.
[0048] As a preferred feasible embodiment, the specific identification process of abnormal areas in the target retail shopping store includes: extracting the customer flow of each shelf area from the shopping environment information, filtering out each shelf area with a customer flow greater than a preset customer flow threshold, and recording them as high flow areas; extracting the purchase volume of each high flow area from the customer transaction information, filtering out each high flow area with a purchase volume less than a preset purchase volume threshold, and recording them as high flow low conversion areas.
[0049] It should be noted that the specific operation of extracting the purchase volume of each high-traffic area from the customer transaction information is as follows: extract the purchase volume of each shelf area from the customer transaction information, and further filter the purchase volume of each high-traffic area.
[0050] Extract the contact volume of each shelf area from the shopping environment information, and filter out each shelf area with a contact volume greater than a preset contact volume threshold. These areas are then recorded as high-contact areas. Extract the purchase volume of each high-contact area from the customer transaction information, and filter out each high-traffic area with a purchase volume less than a preset purchase volume threshold. These high-contact low-conversion areas are then recorded as high-contact low-conversion areas.
[0051] It should be noted that the specific operation of extracting the purchase volume of each high-touch area from the customer transaction information is as follows: extract the purchase volume of each shelf area from the customer transaction information, and further filter the purchase volume of each high-touch area.
[0052] Step 4: Analyze the causes of abnormal areas according to the preset abnormal judgment rules, and trigger differentiated management strategies to achieve intelligent management.
[0053] As a preferred feasible embodiment, the specific content of the preset anomaly judgment rule includes: if the average dwell time of customers in a high-traffic, low-conversion area is less than or equal to one-third of the average dwell time of the whole store, it means that the customer is only "passing by" rather than "actively staying", and the cause of this is recorded as aisle-type invalid traffic.
[0054] If over 80% of customers in a high-traffic, low-conversion area move in a "fast, straight line" (without turning back or stopping), the cause is recorded as a lack of attractive product displays.
[0055] If a high-traffic, low-conversion area is located on the main passage, the only way to the entrance, or next to the cashier, then the reason for its occurrence is recorded as high natural traffic but not a target area.
[0056] As a preferred feasible embodiment, the specific content of the preset anomaly judgment rule also includes: if the proportion of customers who show negative micro-expressions such as "frowning" or "shaking their heads" when touching the product in a high contact low conversion area is greater than or equal to 30%, then the cause of this is recorded as poor product experience (such as damaged packaging, insufficient trial packs, or questionable quality).
[0057] If a customer in a high-touch, low-conversion area comes into contact with the same product multiple times (≥3 times) but does not make a purchase, the reason for this is recorded as decision hesitation (e.g., lack of comparative information, unclear promotional information).
[0058] If the inventory of frequently touched goods in a high-contact, low-conversion area is less than or equal to half of the safety stock, the cause of this is recorded as a shortage of goods.
[0059] This invention identifies high-traffic, low-conversion areas and high-touch, low-conversion areas in real time based on customer behavior mapping, and accurately locates the causes of anomalies based on multi-dimensional data analysis through preset anomaly judgment rules, realizing automated analysis from "discovering problems" to "locating causes".
[0060] As a preferred feasible embodiment, the specific content of the differentiated management strategy includes: if the cause of the abnormal area is invalid traffic of the passageway, then a strategy of inserting along the way to purchase goods, dynamic display guidance, or functional area segmentation management can be implemented.
[0061] It should be noted that the specific content of the "buy-on-the-way" product placement management strategy includes: setting up shelves for high-frequency, low-value products (such as chewing gum, bottled water, and charging cables), using the psychology of "buying on the way" to convert traffic, and using historical traffic data to calculate the critical time for customers to stay in this area (such as <15 seconds), and selectively choosing products that can be decided within 3 seconds.
[0062] The specific content of the dynamic display guidance management strategy includes: using installed electronic price tags to play dynamic information such as "limited-time offer" and "hot-selling recommendations" in a loop to forcefully attract attention (such as "Buy one get one free on potato chips 5 meters ahead").
[0063] The specific content of the functional area segmentation management strategy includes: dividing one side of the wide aisle into a "temporary experience area" (such as a sample trial table or pop-up display stand), and using physical props (such as low shelves or floor stickers) to force customers to slow down.
[0064] One concrete example is the "10-second quick makeup test station" set up in the cosmetics area aisle, along with self-service makeup remover wipes, to lower the barrier to entry for the experience.
[0065] If the cause of the abnormal area is a lack of attractive product display, then a product display structure adjustment management strategy can be implemented.
[0066] It should be noted that the specific contents of the product display structure adjustment management strategy include: adjusting the display according to the principle of "sight height = sales height": (1) 1.2-1.8 meters (adult eye level area): place high-margin, new products or brightly colored products (such as popular snacks and imported cosmetics); bottom shelf: use floor stickers to mark "surprise price zone" and match it with interesting copywriting that "can be seen when you squat down" (such as "surprise when you bend over, second item half price").
[0067] If the abnormal area is caused by high natural traffic but not the target area, then customer diversion and guidance or transitional product buffer zone management strategies can be implemented.
[0068] It should be noted that the specific content of the customer flow diversion and guidance management strategy includes: setting up smart guide screens at the entrance of the area and pushing routes based on customers' dynamic interest tags.
[0069] A specific example is that (1) for customers with children, the message "Parent-child goods area → Toy area → Cashier" is pushed; (2) for young women, the message "Beauty area → Clothing area → Rest area" is pushed.
[0070] The specific content of the transitional goods buffer zone management strategy includes: setting up "cross-category related goods" shelves at the edge of non-target areas as a traffic filter: In a concrete example, the entrance to the senior health products area displays food products "suitable for middle-aged and elderly people" (such as sugar-free biscuits) to attract the target customer group to go deeper, while non-target customers complete their purchases at a low decision cost here.
[0071] If the cause of the abnormal area is a poor product experience, then real-time inspection and rapid repair, transparent experience upgrades, and instant collection and management strategies for customer feedback can be implemented.
[0072] It should be noted that the specific content of the real-time inspection and rapid repair management strategy includes: deploying an AI visual inspection system to scan the shelves every 30 minutes, identify problems such as damaged packaging, empty bottles, and misaligned labels, and automatically generate work orders to push to the stock clerks; when trial packs are insufficient (e.g., less than 10% remaining), triggering the intelligent replenishment cabinet to automatically replenish them.
[0073] Problematic products must be processed within 30 minutes; failure to do so within this timeframe will escalate the issue to the supervisor.
[0074] The specific content of the transparent experience upgrade management strategy includes: adding an initial "experience commitment letter" for high-touch, low-conversion products; supporting "instant exchange" without going to the service desk; and providing a "7-day no-reason return and exchange" QR code, which can be scanned to apply directly.
[0075] One specific example is a "Trial Safe Card" in the skincare product section, which states, "This product has been sterilized, so please try it with confidence. If you are not satisfied, you can return or exchange it."
[0076] The specific content of the customer feedback collection and management strategy includes: setting up touch screen feedback machines in areas with poor experience, where customers can select tags such as "packaging problem", "no trial pack", and "quality question", which are synchronized to the management backend in real time.
[0077] Among them, three consecutive similar feedbacks trigger automatic replenishment / repair, avoiding missed inspections during manual inspections.
[0078] If the cause of the abnormal area is decision-making hesitation, a rapid decision-making information enhancement management strategy can be implemented.
[0079] It should be noted that the specific content of the rapid decision-making information enhancement management strategy includes: setting up "comparison information boards" next to the products to quickly convey the core differences using iconographic language.
[0080] A specific example is: (1) Price comparison: "15% cheaper than similar products" (with price comparison table); (2) Function comparison: "This product contains ingredient X, which improves the moisturizing effect by 30%" (with ingredient radar chart).
[0081] If the cause of an anomaly in a certain area is a shortage of goods, a three-level intelligent replenishment response management strategy can be implemented.
[0082] It should be noted that the specific content of the intelligent replenishment three-level response management strategy includes: Level 1 response (inventory 20%-50%): The system automatically sends a replenishment request to the warehouse, and the estimated arrival time is displayed on the electronic price tag on the shelf (e.g., "Replenishment at 10:00 tomorrow"); Level 2 response (inventory < 20%): Triggers "similar product recommendation" (e.g., if brand A milk powder is out of stock, a recommendation screen for brand B will automatically pop up on the display shelf); Level 3 response (inventory = 0%): Empty shelf spaces display "out-of-stock registration QR code", and customers can scan the code to schedule an arrival notification, with priority delivery upon arrival.
[0083] This invention automatically triggers customized strategies for different causes of anomalies, forming an intelligent management closed loop of "monitoring-analysis-execution-feedback", which effectively improves operational efficiency and conversion rate, and provides data accumulation for long-term operation strategies, supports periodic display adjustments and product selection optimization, and realizes the upgrade from "experience-driven" to "data-driven" management model.
[0084] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.
Claims
1. A method for intelligent analysis and management of customers based on artificial intelligence technology in the field of visual networks, characterized in that: include: Step 1: Based on the multi-dimensional video stream data of the target shopping store within the set monitoring time period, extract customer behavior information, customer transaction information, and shopping environment information; Step 2: Based on the customer behavior information, determine the structured behavior tags and dynamic interest tags of all customers within the set monitoring period, and construct a customer behavior map with time-series correlation accordingly; Step 3: Identify abnormal areas in the target retail store based on customer behavior mapping, including high-traffic, low-conversion areas and high-touch, low-conversion areas; Step 4: Analyze the causes of abnormal areas according to the preset abnormal judgment rules, and trigger differentiated management strategies to achieve intelligent management.
2. The video network customer intelligent analysis and management method based on artificial intelligence technology according to claim 1, characterized in that: The behavioral tags are specifically: quick browsing, in-depth selection, long dwell time, hesitant comparison, goal-oriented, and random browsing; The specific process for determining the structured behavior tags of all customers within the set monitoring period includes: A1. Using customer ID as the unit, the behavior tags corresponding to each action are chained together according to the time sequence of their behavior during their movement in the merchandise display area to generate a behavior tag sequence; A2. If the same behavior tag appears consecutively for a customer, it will be merged into one behavior tag and the duration will be recorded. A3. Based on the combination of tags in the behavior tag sequence and in conjunction with predefined behavior pattern rules, determine higher-order behavior patterns; A4. Based on the customer's behavioral tag sequence, associate it with its higher-level behavioral patterns, collectively referred to as the customer's structured behavioral tags, and then obtain the structured behavioral tags of all customers within the set monitoring period.
3. The method for intelligent analysis and management of video network customers based on artificial intelligence technology according to claim 2, characterized in that: The dynamic interest tags are specifically categorized as mild interest, moderate preference, and high interest. The specific process for determining the dynamic interest tags of all customers within the set monitoring period includes: obtaining the interest scores of each shelf category that all customers stopped at and each product category that they came into contact with during their movement in the merchandise display area based on the customer interest tag determination logic, and conducting preliminary screening to determine the shelf categories and product categories of interest of all customers during their movement in the merchandise display area. As a customer moves through the merchandise display area, the categories of shelves of interest are sorted from highest to lowest according to interest rating. The top three categories of shelves of interest are taken as the main categories of shelves of interest for the customer during the movement of the merchandise display area. Similarly, we can obtain the main interest shelf categories and main interest product categories of all customers during their movement in the merchandise display area, and record them as the dynamic interest tags of all customers within the set monitoring period.
4. The video network customer intelligent analysis and management method based on artificial intelligence technology according to claim 3, characterized in that: The specific content of the customer interest tag determination logic includes: Extract customer behavior information to show the duration of each customer's stay on each shelf, the frequency of each product contact, and the micro-expressions when stopping at shelves and touching products. By comparing the dwell time of each shelf with the dwell time-interest tag mapping table stored in the database, the interest tags corresponding to the dwell time of each customer on each shelf during the movement of all customers in the merchandise display area can be obtained. Similarly, the contact frequency of each product touched by each customer during the movement of all customers in the merchandise display area, as well as the interest tags corresponding to the micro-expressions when stopping at the shelf and touching the product, can be obtained.
5. The video network customer intelligent analysis and management method based on artificial intelligence technology according to claim 3, characterized in that: The specific construction process of the customer behavior graph with temporal correlation includes: Align each customer's behavioral tag sequence with the changes in dynamic interest tags along the timeline to form a "behavioral stage-interest point" relationship diagram, which is recorded as a customer behavior map with temporal correlation.
6. The method for intelligent analysis and management of video network customers based on artificial intelligence technology according to claim 1, characterized in that: The high-traffic, low-conversion area specifically refers to the shelf area within the target retail store where customer traffic is high but the actual purchase rate is low. The high-touch, low-conversion area specifically refers to the shelf areas within the target retail store where customers frequently interact with products but the actual purchase rate is relatively low.
7. The intelligent analysis and management method for video network customers based on artificial intelligence technology according to claim 1, characterized in that: The specific process for identifying abnormal areas in the target retail store includes: Extract customer traffic from shopping environment information for each shelf area, filter out shelf areas with customer traffic exceeding a preset customer traffic threshold and mark them as high-traffic areas. Extract purchase volume from customer transaction information for each high-traffic area, filter out high-traffic areas with purchase volume less than a preset purchase volume threshold and mark them as high-traffic low-conversion areas. Extract the contact volume of each shelf area from the shopping environment information, and filter out each shelf area with a contact volume greater than a preset contact volume threshold. These areas are then recorded as high-contact areas. Extract the purchase volume of each high-contact area from the customer transaction information, and filter out each high-traffic area with a purchase volume less than a preset purchase volume threshold. These high-contact low-conversion areas are then recorded as high-contact low-conversion areas.
8. The intelligent analysis and management method for video network customers based on artificial intelligence technology according to claim 1, characterized in that: The specific content of the preset anomaly determination rules includes: If the average dwell time of customers in a high-traffic, low-conversion area is less than or equal to one-third of the average dwell time of the entire store, it means that the customers are only "passing by" rather than "actively staying". The cause of this is recorded as invalid traffic from the passageway. If over 80% of customers in a high-traffic, low-conversion area move in a "fast, straight line" manner, then the cause is attributed to unattractive product displays. If a high-traffic, low-conversion area is located on the main passage, the only way to the entrance, or next to the cashier, then the reason for its occurrence is recorded as high natural traffic but not a target area.
9. The video network customer intelligent analysis and management method based on artificial intelligence technology according to claim 8, characterized in that: The specific content of the preset anomaly determination rules also includes: If the proportion of customers who exhibit negative micro-expressions such as "frowning" or "shaking their heads" when interacting with products in a high-contact, low-conversion area is greater than or equal to 30%, then the cause of this is recorded as poor product experience. If the same customer in a high-contact, low-conversion area comes into contact with the same product multiple times but does not purchase it, the reason for this is recorded as decision hesitation. If the inventory of frequently touched goods in a high-contact, low-conversion area is less than or equal to half of the safety stock, the cause of this is recorded as a shortage of goods.
10. The method for intelligent analysis and management of visual network customers based on artificial intelligence technology according to claim 9, characterized in that: The specific content of the differentiated management strategy includes: If the abnormal area is caused by invalid traffic in the aisle, then you can implement strategies such as inserting products along the way, dynamic display guidance, or functional area segmentation management. If the cause of the abnormal area is a lack of attractive product display, then a product display structure adjustment management strategy can be implemented. If the abnormal area is caused by high natural traffic but not the target area, then customer diversion and guidance or transitional goods buffer zone management strategies can be implemented. If the cause of the abnormal area is a poor product experience, then real-time inspection and rapid repair, transparent experience upgrade, and real-time collection and management of customer feedback can be implemented. If the cause of the abnormal area is decision-making hesitation, a rapid decision-making information enhancement management strategy can be implemented. If the cause of an anomaly in a certain area is a shortage of goods, a three-level intelligent replenishment response management strategy can be implemented.