A machine learning based deep analysis system for user behavior

Abstract

The application discloses a kind of user behavior depth analysis systems based on machine learning, belong to user behavior analysis technical field.The system includes data acquisition module obtains user position coordinate set, user purchase data and user take commodity frequency set;Trajectory generation module processes user position coordinate set and generates user trajectory set;Commodity comprehensive score module carries out user preference analysis to user purchase data and user take commodity frequency set and obtains commodity comprehensive score set;Space matching module is modeled in combination with GIS and CAD technology accordingly, and carries out heat analysis and space matching and obtains commodity preliminary layout scheme;Commodity layout optimization module optimizes commodity preliminary layout scheme by simulated annealing algorithm, and visual display module realizes commodity layout scheme visualization by 3D modeling software.The application cooperates by multiple modules, utilizes machine learning technology, realizes depth analysis to user behavior, and then optimizes commodity layout, improves commercial operation efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of user behavior analysis, and particularly relates to a user behavior deep analysis system based on machine learning. BACKGROUND

[0002] In today's business operation field, user behavior analysis plays a crucial role in improving business operation efficiency and optimizing business decisions. With the rapid development of information technology, how to deeply analyze user behavior with the help of advanced technical means and optimize commodity layout has become the focus of the industry.

[0003] The current traditional user behavior analysis method has obvious limitations: at the data collection level, it is difficult to comprehensively and real-time obtain the moving track of users in commercial places and the dynamic behavior data of commodity taking, resulting in lack of reliable support for analysis; when processing the track, the missing values and noise in the position coordinate data are not handled well, and the generated user track cannot truly reflect the behavior pattern; in commodity preference degree analysis, it is difficult to effectively integrate the taking frequency and purchase data, and it is difficult to accurately evaluate user preference, affecting the rationality of layout; layout optimization lacks effective algorithm model, and cannot associate user behavior with layout design, making it difficult to generate a scientific scheme, so that the commodity layout cannot meet the demand, affecting the sales efficiency; in addition, the layout display method is single, and lacks intuitive and dynamic visualization means, which is not conducive to the evaluation of the scheme effect by the management personnel.

[0004] In summary, the existing technology has problems of incomplete data collection, inaccurate track processing, insufficient data integration, imperfect optimization algorithm and non-intuitive visualization display in user behavior deep analysis and commodity layout optimization, and an urgent need exists for a user behavior deep analysis system based on machine learning to solve these problems, so as to realize deep analysis of user behavior, optimize commodity layout and improve business operation efficiency. SUMMARY

[0005] The application aims to provide a user behavior deep analysis system based on machine learning to solve the problems raised in the background.

[0006] To solve the above technical problems, the application provides the following technical solutions:

[0007] A user behavior deep analysis system based on machine learning, comprising a data collection module, a track generation module, a commodity comprehensive scoring module, a space matching module, a commodity layout optimization module and a visualization display module;

[0008] The data collection module is used to obtain a user position coordinate set, user purchase data and a user commodity taking frequency set;

[0009] a trajectory generation module for performing coordinate processing on the user position coordinate set to generate a user trajectory set;

[0010] a commodity comprehensive score module for performing user preference analysis on the user purchase data and the user commodity pickup frequency set to obtain a commodity comprehensive score set;

[0011] a space matching module for constructing a store commodity location digital map based on CAD and GIS spatial digital modeling technology, and sequentially performing user trajectory heat analysis and commodity space matching processing on the store commodity location digital map according to the commodity comprehensive score set and the user trajectory set to obtain a commodity preliminary layout scheme;

[0012] a commodity layout optimization module for processing the commodity preliminary layout scheme through a simulated annealing algorithm to obtain a commodity layout scheme;

[0013] a visual display module for realizing 3D visual display of the commodity layout scheme through a 3D modeling software.

[0014] Preferably, the process of obtaining the user position coordinate set, the user purchase data and the user commodity pickup frequency set comprises:

[0015] By deploying Wi-Fi probe devices and combining with IEEE 802.11 protocol framework, the MAC addresses of user mobile devices entering the coverage area are scanned and captured in real time, for each MAC address, the mobile device position coordinate data corresponding to each MAC address is recorded synchronously, the mobile device position coordinate data includes user mobile device MAC address, time stamp and mobile device position coordinate point, the user mobile device MAC address is used for uniquely marking the user mobile device to ensure the individual correlation of the trajectory data, the time stamp adopts UTC time format and is used for subsequent trajectory time sequence analysis, the mobile device position coordinate point is obtained through Wi-Fi signal strength fingerprint positioning algorithm, the mobile device position coordinate data of the same user mobile device MAC address in the time dimension is aggregated into a user position coordinate subset, and all user position coordinate subsets are integrated through a distributed storage architecture to form a user position coordinate set containing space-time dimensions;

[0016] The UHF frequency band RFID reader is embedded in the shelf board, the antenna array coverage range is designed as 0.8 m*0.5 m, which is used to ensure the effective identification of the commodity taking action, when the commodity attached with the EPC Class 1 Gen 2 standard label is taken by the user, the RFID reader analyzes the commodity ID through the anti-collision algorithm, and accumulates the corresponding frequency to form a user-taken commodity frequency subset with the commodity ID as the key, and the user-taken commodity frequency set is obtained accordingly, and the sales data is read in real time through the API interface of the POS system, the user purchase rate corresponding to each commodity ID is extracted, and the user purchase data is obtained, wherein the purchase rate is the ratio of the number of commodity purchases in a time period to the total inventory of the commodity;

[0017] The user position coordinate set, the user purchase data and the user-taken commodity frequency set are only used for legal scenarios.

[0018] Preferably, the process of generating the user trajectory set by coordinate processing the user position coordinate set comprises:

[0019] The missing mobile device position coordinate data in each user position coordinate subset in the user position coordinate set is identified through a time interval analysis method, the time interval analysis method adopts a sliding window mechanism for each user position coordinate subset, calculates the time interval of the mobile device position coordinate points of adjacent positions, when the time interval exceeds a preset interval threshold, it is determined that there is a discontinuity in data collection, and is marked as a data missing segment, and is filled through a linear interpolation method, to obtain the data-cleaning user position coordinate set, wherein the preset interval threshold is set based on the standard for judging the continuity of user behavior in GB / T 35626-2017 “Performance Requirements and Test Methods for Indoor Positioning System”;

[0020] Each user trajectory of each user position coordinate subset in the data-cleaning user position coordinate set is obtained through Kalman filtering algorithm for smoothing processing, and each user trajectory of each user position coordinate subset is taken as a user trajectory subset, and the user trajectory set is obtained accordingly.

[0021] Preferably, the process of analyzing the user preference degree of the user purchase data and the user-taken commodity frequency set comprises:

[0022] The user-taken commodity frequency corresponding to each commodity ID is extracted from the user purchase data and the user-taken commodity frequency set and the user purchase rate The user-taken commodity frequency corresponding to each commodity ID is extracted from the user purchase data and the user-taken commodity frequency set and the user purchase rate The comprehensive score corresponding to each commodity ID is obtained through a user preference degree weighting formula , and the commodity comprehensive score set is obtained accordingly, wherein the commodity ID is used To express distinction;

[0023] The user preference weighted formula is:

[0024] ;

[0025] in, is the weight coefficient corresponding to each product ID, and They are the maximum and minimum frequencies of taking goods for all goods;

[0026] The weight coefficient corresponding to each product ID Calculated by the product weight multi-attribute calculation formula;

[0027] The formula for calculating the product weight multi-attribute is:

[0028] ;

[0029] in, is the commodity price weight, is the product category weight, is the product volume weight, is the price mapping function, is the category mapping function, is the volume mapping function, and ;

[0030] The price mapping function is:

[0031] ,and ;

[0032] in, is the average price of all products. is the price standard deviation;

[0033] The category mapping function is:

[0034] ;

[0035] in, For the The preset weight of the category, is the first hot encoding vector of the product category Dimension value;

[0036] The volume mapping function is:

[0037] ;

[0038] in, is the volume of the product, and It is the maximum volume of all goods.

[0039] Preferably, the digital map of the store's merchandise locations is composed of The grid is composed of Assign unique spatial coordinates and base weight .

[0040] Preferably, the process of performing user trajectory thermal analysis on the digital map of store merchandise locations is as follows:

[0041] The coordinate points of the mobile device location in each user trajectory subset in the user trajectory set are mapped to the nearest grid of the digital map of the store product location according to the coordinates, and the kernel density estimation algorithm is used to calculate the density of each grid. The trajectory point density , the trajectory point density Substitute the thermal weight formula to calculate each grid Thermal weight , based on which user trajectory thermal analysis can be performed on the digital map of store merchandise locations;

[0042] The thermal weight formula is:

[0043] ;

[0044] in, is the dynamic adjustment coefficient, is the density sensitivity factor.

[0045] Preferably, the process of performing product space matching processing on the digital map of store product locations is as follows:

[0046] For each grid Thermal weight and base weight , calculated by the grid value formula, each grid is obtained Grid value , according to the grid value The size of each grid Arrange in descending order to obtain a grid descending sequence;

[0047] The grid value formula is:

[0048] ;

[0049] in, is the impact factor of the adjacent grid value;

[0050] The adjacent grid value influencing factor is:

[0051] ;

[0052] in, For Grid The grid value of the adjacent grid, and when the grid is located at the boundary, the missing adjacent grid is counted as 0 in the sum;

[0053] From the first grid in the descending grid sequence, sort by grid value Select 15% of the total number of grids in descending order to form a high-value grid area, and then start from the last grid in the descending order of grids. Select grids that account for 50% of the total number of grids in descending order to form a low-value grid area, and the remaining grids in descending order to form a medium-value grid area;

[0054] The comprehensive score corresponding to each product ID in the product comprehensive score set Normalize the comprehensive score Scale to the [0,1] interval to get the product value corresponding to each product ID , for each product ID corresponding to the product value Make commodity value judgments;

[0055] If the value of the product If it is greater than or equal to 0.8, the product represented by the product ID is a high-rated product;

[0056] If the value of the product If the value is less than 0.8 and greater than 0.5, the product represented by the product ID has a medium rating; otherwise, it has a low rating.

[0057] Randomly assign each high-scoring product to an unassigned grid in the high-value grid area, randomly assign each medium-scoring product to an unassigned grid in the medium-value grid area, and randomly assign each low-scoring product to an unassigned grid in the low-value grid area. Record the grid corresponding to each product ID accordingly. The spatial coordinates of , forming a preliminary product layout plan;

[0058] The preliminary product layout plan, namely the solution space encoding plan, includes a layout vector, a category grouping constraint, and a grid occupancy matrix. The layout vector uses a series of coordinates to describe the positions of all products and is the basic object of the simulated annealing algorithm. The category grouping constraint is used when the simulated annealing algorithm generates a new layout. The category grouping constraint is used to determine whether similar products are dispersed according to the category grouping constraint, thereby deciding whether to accept the new solution. The grid occupancy matrix is ​​used to quickly detect whether there is any product overlap after the simulated annealing algorithm generates a new layout. Similar products are classified as high-rated products, medium-rated products, or low-rated products.

[0059] The mathematical representation of the Layout vector is: ,in Indicates that the product ID is The grid corresponding to the product The spatial coordinates of ;

[0060] The mathematical expression of the category grouping constraint is: , which is represented as a set of products that are high-rated products, medium-rated products, or low-rated products, where is 011, 012 or 013, 011 represents high-rated products. 012 represents a medium-rated product. 013 represents low-rated products. Represents the product ID.

[0061] Preferably, the preliminary product layout plan is processed by a simulated annealing algorithm to obtain the product layout plan:

[0062] Set the constraints and process the preliminary product layout plan through the simulated annealing algorithm to obtain a product layout plan that meets the constraints. The constraints include the spatial region constraint Region and the display space constraint SpaceQuota. The spatial region constraint Region records the grids covered by the high-value grid area, low-value grid area and medium-value grid area respectively. The spatial coordinates of , the display space constraint SpaceQuota records the number of grids that each product ID needs to occupy, that is, The number of grid cells to occupy.

[0063] Preferably, a 3D virtual model of the store is constructed by 3D modeling software, and the product layout plan is imported into it to achieve 3D visual display. The 3D modeling software is SketchUp or 3ds Max.

[0064] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0065] This invention achieves in-depth analysis of user behavior and scientific optimization of product layout through multi-source data fusion collection, intelligent trajectory processing, multi-attribute modeling of product preferences, spatial value assessment based on GIS and CAD, simulated annealing algorithm optimization layout and 3D visualization display. Compared with traditional solutions, it solves the problems of data fragmentation, trajectory distortion, single evaluation and layout experience, improves the accuracy of user behavior analysis, the exposure rate of high-scoring products, the rationality of layout and sales efficiency, and has the ability to dynamically adapt to changes in business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0067] Figure 1 Schematic diagram of the system function modules of the present invention. DETAILED DESCRIPTION

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0069] Examples, such as Figure 1 The user behavior in-depth analysis system based on machine learning includes a data collection module, a trajectory generation module, a product comprehensive scoring module, a space matching module, a product layout optimization module and a visualization display module, which work together to complete the in-depth analysis of user behavior and obtain a product layout plan.

[0070] The data collection module is used to obtain the user's location coordinates, user purchase data, and the user's frequency of picking up goods;

[0071] A trajectory generation module, configured to perform coordinate processing on a user position coordinate set to generate a user trajectory set;

[0072] The product comprehensive rating module is used to analyze user preferences based on user purchase data and the frequency of user product acquisition to obtain a comprehensive product rating set;

[0073] The space matching module is configured to construct a store commodity position digital map based on CAD and GIS spatial digital modeling technology, and sequentially perform user trajectory heat analysis and commodity space matching processing on the store commodity position digital map according to a set of commodity comprehensive scores and a set of user trajectories to obtain a commodity preliminary layout scheme.

[0074] The commodity layout optimization module is configured to process the commodity preliminary layout scheme through a simulated annealing algorithm to obtain a commodity layout scheme.

[0075] The visual display module is configured to realize 3D visual display of the commodity layout scheme through a 3D modeling software.

[0076] Further, the working principle of the application is described below through an embodiment.

[0077] Twelve Wi-Fi probes are deployed in a 2000m2 supermarket, 50 UHF RFID readers are embedded in shelves, an IBM SurePOS 700 series is used for a POS system, a Spark cluster is built for distributed data processing, and an HBase 2.4.10 distributed database is used to store trajectory data and commodity scores, and the data retention period is 12 months.

[0078] The Wi-Fi probe scans the MAC address of a user's mobile phone at an interval of 100ms, generates a coordinate point through a signal strength fingerprint positioning algorithm, the positioning accuracy is 1.2m, for example, the MAC address of a certain user is 00:1A:2B:3C:4D:5E, the coordinate at 10:00:01 on July 1, 2025 is (12.5, 8.3), the timestamp record is 1688272801000, when stored in a distributed manner, each user position coordinate subset is stored in the format of {MAC: [(t1, x1, y1), (t2, x2, y2),...]} in the location_table table of HBase, and the RowKey is the MAC+timestamp; the RFID reader reads the commodity label through the ISO 18000-6C protocol, such as the taking action of the commodity ID G001234 on the milk shelf, the taking trigger frequency is accumulated each time, and is stored in the frequency_table table, the fields include the commodity ID, the taking time and the frequency; the POS system pushes the purchase data in real time through the REST API, for example, the purchase number of the commodity ID G001234 on July 1, 2025 is 23 times, the total inventory is 100, the purchase rate is 23%, and is stored in the purchase_table table.

[0079] A 5-second sliding window is used for each user coordinate subset to calculate the time interval between adjacent points. For example, in a user trajectory point sequence, the interval between t1=10:00:01 and t2=10:00:20 is 19 seconds, which is greater than the preset interval threshold of 15 seconds. This segment is considered missing and linear interpolation is performed to fill the missing segment. Given (t1, 12.5, 8.3) and (t2, 13.2, 8.5), the coordinates of the missing segment t=10:00:10 are calculated as:

[0080] , based on which the user location coordinate set after data cleaning is obtained; each user location coordinate subset in the user location coordinate set after data cleaning is smoothed by the Kalman filter algorithm, and the initial value of the state vector is set to the first mobile device location coordinate point in the user location coordinate subset [12.5, 8.3, 0, 0], the process noise covariance is diag([0.1, 0.1, 0.5, 0.5]), and the observation noise covariance is diag([2, 2]). Based on this, the trajectory of each user location coordinate subset is smoothed to obtain the user trajectory set.

[0081] Take a product IDG005678 as an example. The price of the product IDG00567 is 35 yuan. The average price of all products is is 25, the price standard deviation The value is 10, the category belongs to fresh products (the default weight of fresh products is 0.8), the hot encoding vector is [1,0,0], and the volume is 500cm 3 The maximum value of the volume of all goods is 1000, and the minimum value of the volume of all goods is 100. After standardization is 1, the price mapping function output , category mapping value , the volume mapping function is ; Based on this, the product weight multi-attribute calculation formula The calculated weight coefficient of product IDG005678 is 0.76, and the user's frequency of taking product IDG005678 is 85, user purchase rate 32% is the highest frequency of taking goods from all products With minimum value are 150 and 10 respectively, and the weighted formula of user preference is used. Calculate and get the comprehensive score of product IDG005678 About 0.49, the normalized commodity value The score is 0.49, which is determined to be a medium-rated product. Other products are treated in the same way as product IDG005678 to determine whether they are high-rated, medium-rated, or low-rated products.

[0082] A grid in a digital map of store merchandise locations Thermal weight is 0.7, and the grid is obtained by the kernel density estimation algorithm The trajectory point density 8 points / m 2 , substituted into the thermal weight formula to calculate , adjacent grid value is [0.6, 0.55, 0.62, 0], where the boundary grid is filled with 0, and the adjacent influence factor is calculated It is about 0.44, based on which the grid value formula is further used to calculate the grid Grid value , the grid The grid value of the grid ranks first 15% in the descending order of the grid, and is classified into the high-value grid area. The grids in the digital map of the store's product location are compared with the grids. The same process is used to determine whether it is a high-value grid area, a medium-value grid area, or a low-value grid area; each high-scoring product in the supermarket is randomly assigned to an unassigned grid in the high-value grid area, each medium-scoring product is randomly assigned to an unassigned grid in the medium-value grid area, and each low-scoring product is randomly assigned to an unassigned grid in the low-value grid area. Based on this, the grid corresponding to each product ID is recorded. The spatial coordinates of , and form a preliminary product layout plan.

[0083] The initial temperature of the simulated annealing algorithm is set to 100, the cooling coefficient is 0.95, and the number of iterations is 100. The constraints are that high-scoring products must be placed in high-value grid areas, medium-scoring products must be placed in medium-value grid areas, and low-scoring products must be placed in low-value grid areas. The volume of a shampoo product occupies 2 grids, and the SpaceQuota is 2. Based on this constraint, the preliminary product layout plan is optimized using the simulated annealing algorithm. The positions of two products are randomly exchanged, and the objective function is calculated, such as the sum of the grid values ​​of the high-scoring products. If the new solution is better, it is accepted, otherwise it is replaced by probability.

[0084] Accept and finally get the product layout plan.

[0085] Import the optimized product layout plan into SketchUp, place the product models according to the grid coordinates, mark high-scoring products in red, medium-scoring products in yellow, and low-scoring products in blue, overlay the trajectory heat map, and generate an interactive 3D model.

[0086] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A user behavior in-depth analysis system based on machine learning, characterized by: include: The data collection module is used to obtain the user's location coordinates, user purchase data, and the user's frequency of picking up goods; A trajectory generation module, configured to perform coordinate processing on a user position coordinate set to generate a user trajectory set; The product comprehensive rating module is used to analyze user preferences based on user purchase data and the frequency of user product acquisition to obtain a comprehensive product rating set; The spatial matching module is used to construct a digital map of store product locations based on spatial digital modeling technology using CAD and GIS. It also performs user trajectory thermal analysis on the digital map of store product locations based on a set of user trajectories, and calculates the thermal weight of each grid in the digital map of store product locations. Based on this, the modules perform spatial matching of products with each grid in the digital map of store product locations based on the thermal weight of each grid and the set of comprehensive product scores, thus obtaining a preliminary product layout plan. The product layout optimization module is used to process the preliminary product layout plan through the simulated annealing algorithm to obtain the product layout plan; Visual display module, used to realize 3D visualization of product layout plan through 3D modeling software; The method for obtaining the user location coordinate set, user purchase data, and user product picking frequency set is as follows: Capturing the MAC address of each user's mobile device through a Wi-Fi probe, recording in real time the mobile device location coordinate data corresponding to each user's mobile device MAC address, including the user's mobile device MAC address, timestamp, and mobile device location coordinate point, and treating the mobile device location coordinate data corresponding to each user's mobile device MAC address as a user location coordinate subset to form a user location coordinate set; The RFID reader embedded in the shelf records the frequency of product pickup corresponding to each product ID, and uses the product pickup frequency corresponding to each product ID as a subset of the user's product pickup frequency to form a user product pickup frequency set. At the same time, user purchase data is read from the POS system, where the user purchase data includes the user purchase rate corresponding to each product ID.

2. The user behavior in-depth analysis system based on machine learning according to claim 1 is characterized in that: The process of performing coordinate processing on the user position coordinate set to generate the user trajectory set: For each user location coordinate subset in the user location coordinate set, the time interval analysis method is used to identify missing mobile device location coordinate data, and the missing mobile device location coordinate data is filled by the linear interpolation method to obtain the user location coordinate set after data cleaning. Each user location coordinate subset in the user location coordinate set after data cleaning is smoothed by the Kalman filter algorithm to obtain the user trajectory of each user location coordinate subset. The user trajectory of each user location coordinate subset is regarded as a user trajectory subset, and the user trajectory set is obtained accordingly.

3. The user behavior in-depth analysis system based on machine learning according to claim 2 is characterized in that: The process of analyzing user preference based on user purchase data and user product-taking frequency: The user's product acquisition frequency and user purchase rate corresponding to each product ID are extracted from the user purchase data and the user's product acquisition frequency set. The user's product acquisition frequency and user purchase rate corresponding to each product ID are calculated using the user preference weighted formula to obtain the comprehensive score corresponding to each product ID, and based on this, a comprehensive product score set is obtained.

4. The user behavior in-depth analysis system based on machine learning according to claim 3 is characterized in that: The digital map of the store's merchandise locations is composed of a The grid is composed of Assign unique spatial coordinates and base weight .

5. The user behavior in-depth analysis system based on machine learning according to claim 4 is characterized in that: The process of performing user trajectory thermal analysis on the digital map of store product locations: The coordinate points of the mobile device location in each user trajectory subset in the user trajectory set are mapped to the nearest grid of the digital map of the store product location according to the coordinates, and the kernel density estimation algorithm is used to calculate the density of each grid. The trajectory point density , the trajectory point density Substitute the thermal weight formula to calculate each grid Thermal weight Based on this, user trajectory thermal analysis can be performed on the digital map of store product locations.

6. The user behavior in-depth analysis system based on machine learning according to claim 5, characterized in that: The process of commodity space matching processing: For each grid Thermal weight and base weight , calculated by the grid value formula, each grid is obtained Grid value , according to the grid value The size of each grid Arrange in descending order to obtain a grid descending sequence; From the first grid in the descending grid sequence, sort by grid value Select 15% of the total number of grids in descending order to form a high-value grid area, and then start from the last grid in the descending order of grids. Select grids that account for 50% of the total number of grids in descending order to form a low-value grid area, and the remaining grids in descending order to form a medium-value grid area; Normalize the comprehensive score corresponding to each product ID in the product comprehensive score set, scale the comprehensive score to the [0,1] interval, and obtain the product value corresponding to each product ID , for each product ID corresponding to the product value Make commodity value judgments; If the value of the product If it is greater than or equal to 0.8, the product represented by the product ID is a high-rated product; If the value of the product If the value is less than 0.8 and greater than 0.5, the product represented by the product ID has a medium rating; otherwise, it has a low rating. Randomly assign each high-scoring product to an unassigned grid in the high-value grid area, randomly assign each medium-scoring product to an unassigned grid in the medium-value grid area, and randomly assign each low-scoring product to an unassigned grid in the low-value grid area. Record the grid corresponding to each product ID accordingly. The spatial coordinates of , and form a preliminary product layout plan.

7. The user behavior in-depth analysis system based on machine learning according to claim 6, characterized in that: The process of processing the preliminary product layout plan through the simulated annealing algorithm to obtain the product layout plan: Set constraints and process the preliminary product layout plan through the simulated annealing algorithm to obtain a product layout plan that meets the constraints, where the constraints include spatial area constraints and display space constraints.

8. The user behavior in-depth analysis system based on machine learning according to claim 7 is characterized in that: The 3D modeling software is SketchUp or 3ds Max.

Images