Commodity advertisement recommendation system based on e-commerce platform

Abstract

The invention relates to the technical field of commodity advertisement recommendation, and discloses a commodity advertisement recommendation system based on an e-commerce platform, and the system comprises the steps: obtaining an environment temperature, a relative humidity, and a screen back temperature of an advertisement display; the dew point temperature is calculated according to the environment temperature and the relative humidity; comparing the dew point temperature with the temperature behind the screen to obtain a fog increment prediction value; dividing a display area of the advertisement display into M * N sub-areas; playing the double-stripe reference frame on the advertisement display, and collecting a brightness difference value of the brightness values of the first stripe reference frame and the second stripe reference frame, so as to generate a scattering coefficient of each sub-region based on the brightness difference value; generating a readable weight matrix by combining the scattering coefficient of the kth sub-region, the fog increment prediction value and a preset readability coefficient; and based on the readable weight matrix and the urgency degree, performing matching optimization by adopting a Hungary algorithm to obtain a matching result of the commodity and the sub-region, and performing commodity advertisement recommendation processing based on the matching result.

Description

Technical Field

[0001] The present invention relates to the technical field of product advertisement recommendation, and more specifically, to a product advertisement recommendation system based on an e-commerce platform. Background Art

[0002] For fresh products, existing e-commerce platforms include Hema Fresh and others.

[0003] For example, product advertisements are rotated in Hema Fresh stores for two reasons: Since fresh products have a shelf life, fresh products nearing the end of their shelf life need to be rotated in stores through displays to encourage consumers to purchase the corresponding fresh products as soon as possible, thereby reducing product damage.

[0004] However, the cool air from the refrigerated or seafood counters in stores lowers the temperature of the glass in front of the display screen, while the surrounding relative humidity is high. This often causes the glass temperature to fall below the dew point, causing water vapor to condense into a fog film, which diffuses and scatters light, reducing the brightness of the display screen. This makes it difficult for consumers to see product advertisements clearly, resulting in increased damage to goods. Summary of the Invention

[0005] The present invention provides a product advertisement recommendation system based on an e-commerce platform to solve the technical problems raised in the background technology.

[0006] The present invention provides a product advertisement recommendation system based on an e-commerce platform, comprising:

[0007] Data acquisition module, used to obtain ambient temperature, relative humidity and the temperature behind the screen of the advertising display;

[0008] The first processing module is used to calculate the dew point temperature according to the ambient temperature and relative humidity;

[0009] The second processing module is used to compare the dew point temperature with the temperature behind the screen to obtain the fog increment prediction value;

[0010] A third processing module is used to divide the display area of ​​the advertising display into M×N sub-areas;

[0011] Playing a double-stripe reference frame on the advertising display, and collecting a brightness difference between the brightness values ​​of the first stripe reference frame and the second stripe reference frame to generate a scattering coefficient for each sub-region based on the brightness difference;

[0012] A fourth processing module is configured to generate a readability weight matrix by combining the scattering coefficient of the k-th sub-region, the fog increment prediction value, and the preset readability coefficient;

[0013] The fifth processing module is used to calculate the urgency of the i-th product based on the normalized scrap cost, remaining shelf life, inventory level, and predicted demand of the i-th product;

[0014] The advertising recommendation module is used to perform matching optimization based on the readable weight matrix and urgency, using the Hungarian algorithm to obtain matching results between products and sub-regions, and recommend product advertisements based on the matching results.

[0015] Furthermore, the dew point temperature is calculated based on the ambient temperature and relative humidity as follows:

[0016]

[0017] Where γ represents the temperature-humidity coupling coefficient, RH represents relative humidity, T represents ambient temperature, and 17.27 and 237.7 are both Magnus empirical constants.

[0018]

[0019] Among them, T d Indicates the dew point temperature.

[0020] Furthermore, the dew point temperature is compared with the temperature behind the screen to obtain the fog increment prediction value, including:

[0021] ΔT cond =T d -T s

[0022] Where, ΔT cond represents the condensation driving temperature difference, T s Indicates the temperature behind the screen;

[0023] Δζ pred =Norm(max(0,ΔT cond ))

[0024] Among them, Δζ pred Represents the fog increment prediction value, Norm represents the normalization process, and max represents the max function.

[0025] Furthermore, a double-stripe reference frame is played on the advertising display, and a brightness difference between the brightness values ​​of the first stripe reference frame and the second stripe reference frame is collected to generate a scattering coefficient of each sub-area based on the brightness difference, including:

[0026] Step 41: The advertisement player continuously plays two frames of a first fringe reference frame and a second fringe reference frame with the same frequency and a phase difference of π within a preset time period.

[0027] Step 42 , respectively collecting the grayscale matrix S1 and the grayscale matrix S2 corresponding to the first fringe reference frame and the second fringe reference frame;

[0028] Step 43: Divide the display area into M×N identical sub-areas;

[0029] Step 44: extract the first average grayscale and the second average grayscale of the grayscale matrix S1 and the grayscale matrix S2 corresponding to the kth sub-region respectively, and perform difference processing to obtain the brightness difference value ΔL of the kth sub-region. k ;

[0030] Step 45: When the relative humidity is 20%, repeat steps 41 to 44 to obtain the reference brightness difference ΔL. ref ;

[0031] Step 46: Scattering coefficient Δζ of the kth sub-region k , 1≤k≤K, k is a positive integer.

[0032] Furthermore, the readability weight matrix is ​​generated by combining the scattering coefficient of the k-th sub-region, the fog increment prediction value, and the preset readability coefficient, including:

[0033] Calculate the sum of the scattering coefficient of the kth sub-area and the fog increment prediction value;

[0034] If the sum is greater than 1, the composite scattering value ζ of the kth sub-region is k,syn Set to 1;

[0035] If the sum is less than 0, the composite scattering value ζ of the kth sub-region is k,syn Set to 0;

[0036] Otherwise, the sum is set to the composite scattering value ζ of the kth sub-region k,syn ;

[0037] Obtain P product advertisement templates, read the font size and weight of each advertisement template, perform clustering processing to form J readability clusters, and configure a readability coefficient for each readability cluster; the readability coefficient value range is: (0,1];

[0038] The readability weight is calculated based on the composite scattering of the kth sub-region and the jth readability cluster as follows:

[0039]

[0040] Among them, R j,k Represents readable weight, e represents the natural base, c represents the decay constant, A j represents the readability coefficient of the jth readability cluster, 1≤j≤J, j is a positive integer;

[0041] Based on readable weight R j,k Forming a readable weight matrix.

[0042] Furthermore, the normalized scrap cost, remaining shelf life, inventory, and predicted demand of the i-th product are obtained, and the urgency of the i-th product is calculated, including:

[0043] The predicted demand is obtained based on the output of the pre-trained LSTM neural network model;

[0044] The urgency of the i-th product is as follows:

[0045]

[0046] Among them, U i Indicates the urgency of the i-th product, Q i represents the scrapping cost of the i-th commodity, R i represents the remaining shelf life of the i-th product, S i represents the inventory of the i-th commodity, λ i represents the predicted demand for the i-th commodity.

[0047] Furthermore, based on the readability weight matrix and urgency, the Hungarian algorithm is used for matching optimization to obtain the matching results of products and sub-regions, including:

[0048] Determine the readability cluster to which the product advertisement of the i-th product belongs and the readability weight of the k-th sub-region;

[0049] The product of the urgency of the i-th product, the readability cluster to which the product advertisement of the i-th product belongs, and the readability weight of the k-th sub-region is used as the comprehensive weight W i,k ;

[0050] Based on the comprehensive weight W i,k , construct the cost matrix C; where the cost C in the i-th row and k-th column of the cost matrix C is i,k , C i,k =1-W i,k ;

[0051] The Hungarian algorithm is executed with the cost matrix C as input to obtain the optimal assignment set Π = {(i, k(i)) | i = 1, …, N}; where k(i) represents the sub-region matched by the i-th product, and N represents the number of products.

[0052] Furthermore, product advertisement recommendation processing is performed based on the matching result, including: playing the product advertisement of the i-th product in the matching sub-area.

[0053] The beneficial effects of the present invention are: by introducing a heat and moisture coupling perception mechanism of ambient temperature, relative humidity and the temperature behind the advertising display screen, the risk of atomization is dynamically predicted and the scattering coefficient of the display area is obtained, and the readability clustering characteristics of the advertising template are combined to construct a readability weight matrix; then, the scrap cost, inventory, shelf life and urgency obtained from demand forecast of the product are combined, and the Hungarian algorithm is used to achieve the optimal matching of product advertisements and sub-areas, thereby significantly improving the readability of advertising content and the timeliness of product recommendations in high humidity and low temperature environments, effectively reducing the damage rate of fresh products and enhancing consumers' visual acquisition efficiency and purchase conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a module diagram of the product advertising recommendation system based on the e-commerce platform of the present invention. DETAILED DESCRIPTION

[0055] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0056] like Figure 1 As shown in the figure, the product advertising recommendation system based on the e-commerce platform includes:

[0057] Data acquisition module, used to obtain ambient temperature, relative humidity and the temperature behind the screen of the advertising display;

[0058] The first processing module is used to calculate the dew point temperature according to the ambient temperature and relative humidity;

[0059] The second processing module is used to compare the dew point temperature with the temperature behind the screen to obtain the fog increment prediction value;

[0060] A third processing module is used to divide the display area of ​​the advertising display into M×N sub-areas;

[0061] Playing a double-stripe reference frame on the advertising display, and collecting a brightness difference between the brightness values ​​of the first stripe reference frame and the second stripe reference frame to generate a scattering coefficient for each sub-region based on the brightness difference;

[0062] A fourth processing module is configured to generate a readability weight matrix by combining the scattering coefficient of the k-th sub-region, the fog increment prediction value, and the preset readability coefficient;

[0063] The fifth processing module is used to calculate the urgency of the i-th product based on the normalized scrap cost, remaining shelf life, inventory level, and predicted demand of the i-th product;

[0064] The advertising recommendation module is used to perform matching optimization based on the readable weight matrix and urgency, using the Hungarian algorithm to obtain matching results between products and sub-regions, and recommend product advertisements based on the matching results.

[0065] In one embodiment of the present invention, the dew point temperature is calculated based on the ambient temperature and relative humidity as follows:

[0066]

[0067] Where γ represents the temperature-humidity coupling coefficient, RH represents relative humidity, T represents ambient temperature, and 17.27 and 237.7 are both Magnus empirical constants.

[0068] Ambient temperature and humidity jointly determine the saturated water vapor pressure of the air. Therefore, based on the Magnus model, the ambient temperature and humidity are coupled to obtain the temperature-humidity coupling coefficient.

[0069]

[0070] Among them, T d Indicates the dew point temperature.

[0071] The temperature-humidity coupling coefficient is mapped into dew point temperature through the Magnus model.

[0072] In one embodiment of the present invention, comparing the dew point temperature with the temperature behind the screen to obtain a fog increment prediction value includes:

[0073] ΔT cond =T d -T s

[0074] Where, ΔT cond represents the condensation driving temperature difference, T s Indicates the temperature behind the screen;

[0075] The difference between the dew point and the temperature behind the screen reflects the conditions for condensation of water vapor on the glass surface and the strength of the driving force. Fogging occurs only when the dew point exceeds the temperature behind the screen; otherwise, no additional fog is generated.

[0076] The condensation-driven temperature difference is used to directly quantify the driving force of heat and moisture coupling.

[0077] Δζ pred =Norm(max(0,ΔT cond ))

[0078] Among them, Δζ pred Represents the fog increment prediction value, Norm represents the normalization process, and max represents the max function.

[0079] The condensation-driven temperature difference has the dimension of degrees Celsius and is not suitable for direct coupling with the dimensionless spatial scattering coefficient or readable weights, and only positive values ​​should trigger fog amount prediction.

[0080] The normalization process performs normalization on the condensation-driven temperature difference based on a preset maximum value of the condensation-driven temperature difference.

[0081] In one embodiment of the present invention, a dual-stripe reference frame is played on an advertising display, and a brightness difference between brightness values ​​of a first stripe reference frame and a second stripe reference frame is collected to generate a scattering coefficient for each sub-region based on the brightness difference, including:

[0082] Step 41: The advertisement player continuously plays two frames of a first fringe reference frame and a second fringe reference frame with the same frequency and a phase difference of π within a preset time period.

[0083] The first fringe reference frame and the second fringe reference frame differ in phase by π, which makes the brightness of the two frames complementary to each other in an ideal fog-free state (i.e., relative humidity of 20%). When there is fog scattering on the glass surface, the brightness difference between the two frames becomes a sensitive indicator of the scattering intensity.

[0084] Step 42 , respectively collecting the grayscale matrix S1 and the grayscale matrix S2 corresponding to the first fringe reference frame and the second fringe reference frame;

[0085] Only by accurately obtaining the brightness distribution of each pixel on the screen of the two fringe patterns can the scattering effect be calculated based on the difference. The grayscale matrix S1 and grayscale matrix S2 are generated to provide raw data for the subsequent calculation of the average grayscale difference within the sub-region.

[0086] Step 43: Divide the display area into M×N identical sub-areas;

[0087] It should be noted that the aspect ratio of each sub-area should match the aspect ratio of the product advertisement.

[0088] Step 44: extract the first average grayscale and the second average grayscale of the grayscale matrix S1 and the grayscale matrix S2 corresponding to the kth sub-region respectively, and perform difference processing to obtain the brightness difference value ΔL of the kth sub-region. k ;

[0089] The average grayscale difference within the region can reflect the attenuation degree of the fringe contrast at that location, which is then mapped to the degree of fog film scattering.

[0090] Step 45: When the relative humidity is 20%, repeat steps 41 to 44 to obtain the reference brightness difference ΔL.ref ;

[0091] ΔL under normal reference environment (relative humidity 20%) ref It can be used as a haze-free contrast baseline to eliminate the non-ideal stripe contrast differences of the screen itself. This can be used to construct a normalized benchmark to ensure that the subsequent scattering coefficient is independent of the screen model and stripe contrast.

[0092] Step 46: Scattering coefficient Δζ of the kth sub-region k , 1≤k≤K, k is a positive integer.

[0093] Scattering coefficient Δζ of the kth subregion k , where 0 represents no scattering (clear) and 1 represents maximum scattering (completely foggy).

[0094] represents the brightness attenuation ratio, Used to map the ratio to a scattering coefficient, ensuring that the stronger the scattering, the larger the value.

[0095] In one embodiment of the present invention, a readability weight matrix is ​​generated by combining the scattering coefficient of the kth sub-region, the fog increment prediction value, and the preset readability coefficient, including:

[0096] Calculate the sum of the scattering coefficient of the kth sub-area and the fog increment prediction value;

[0097] If the sum is greater than 1, the composite scattering value ζ of the kth sub-region is k,syn Set to 1;

[0098] If the sum is less than 0, the composite scattering value ζ of the kth sub-region is k,syn Set to 0;

[0099] Otherwise, the sum is set to the composite scattering value ζ of the kth sub-region k,syn ;

[0100] The real-time measured spatial scattering intensity (scattering coefficient) and the short-term predicted fog increase (fog increment prediction value) are unified into a single indicator to comprehensively reflect the degree of fog in the kth sub-area.

[0101] Obtain P product advertisement templates, read the font size and weight of each advertisement template, perform clustering processing to form J readability clusters, and configure a readability coefficient for each readability cluster; the readability coefficient value range is: (0,1];

[0102] It should be noted that the first code is assigned according to the font size, for example, the first codes of 0.25 and 0.3 are assigned to the small four font size and the four font size respectively, and the second code is assigned according to the font thickness, for example, the second code of 0.8 is assigned to the bold font size, and the second code of 0.5 is assigned to the plain font size.

[0103] Calculate the product of the first code and the second code of each product advertisement;

[0104] Obtain J readability clusters through K-means clustering.

[0105] The readability weight is calculated based on the composite scattering of the kth sub-region and the jth readability cluster as follows:

[0106]

[0107] Among them, R j,k Represents readable weight, e represents the natural base, c represents the decay constant, A j represents the readability coefficient of the jth readability cluster, 1≤j≤J, j is a positive integer;

[0108] Combining the degree of atomization (synthetic scattering amount) with the inherent readability of the product advertisement (readability weight), the readability weight of each video advertisement in each sub-area is obtained.

[0109] in, Used to simulate the exponential attenuation of fog on readability naturally. Multiply by A j Take into account the inherent readability of product advertisements.

[0110] Based on readable weight R j,k Forming a readable weight matrix.

[0111] In one embodiment of the present invention, obtaining the normalized scrap cost, remaining shelf life, inventory level, and predicted demand of the i-th product and calculating the urgency of the i-th product includes:

[0112] The predicted demand is obtained based on the output of the pre-trained LSTM neural network model;

[0113] The future demand momentum of a product directly determines the speed of inventory digestion and must be accurately estimated based on historical sales trends. Using a pre-trained LSTM neural network model, inputting sequence data such as historical product sales, the model outputs the predicted demand for the i-th product, providing a dynamic sales rate for urgency calculations.

[0114] It should be noted that the pre-trained LSTM neural network model includes:

[0115] adjacent first sliding window and second sliding window;

[0116] The first sliding window is used to extract the sales volume of goods in the first time period, and the second sliding window is used to extract the sales volume of goods in the second time period;

[0117] The normalized vector constructed by the time series of product sales in the first time period is used as training data;

[0118] The vector constructed by normalizing the sales volume of the product in the second time period is used as the sample data;

[0119] The first sliding window is before the second sliding window, and the window size of the first sliding window is larger than that of the second sliding window.

[0120] The urgency of the i-th product is as follows:

[0121]

[0122] Among them, U i Indicates the urgency of the i-th product, Q i represents the scrapping cost of the i-th commodity, R i represents the remaining shelf life of the i-th product, S i represents the inventory of the i-th commodity, λ i represents the predicted demand for the i-th commodity.

[0123] The comparable dimensionless urgency U is formed by dividing the scrapping cost of each item by the remaining shelf life and multiplying it by the ratio of inventory to forecast demand. i , to get the urgency.

[0124] Urgency is used to comprehensively reflect the potential loss risk per unit time.

[0125] In one embodiment of the present invention, the Hungarian algorithm is used to perform matching optimization based on the readability weight matrix and the urgency, and the matching results of products and sub-regions are obtained, including:

[0126] Determine the readability cluster to which the product advertisement of the i-th product belongs and the readability weight of the k-th sub-region;

[0127] The product of the urgency of the i-th product, the readability cluster to which the product advertisement of the i-th product belongs, and the readability weight of the k-th sub-region is used as the comprehensive weight W i,k ;

[0128] The value of displaying a product advertisement in the kth sub-region depends on both the urgency of the product and the readability weight of the product advertisement in the kth sub-region.

[0129] Multiply the urgency of the i-th product by the readability weight of its product advertisement in the k-th sub-region to obtain the comprehensive weight.

[0130] Based on the comprehensive weight W i,k , construct the cost matrix C; where the cost C in the i-th row and k-th column of the cost matrix C is i,k , C i,k =1-W i,k ;

[0131] The Hungarian algorithm solves the assignment problem of minimizing cost, which requires converting the comprehensive weight to be maximized into the corresponding minimized cost. For the i-th row and k-th column of the cost matrix C, the comprehensive weight W i,k The linear mapping has a cost C i,k The cost matrix C, whose elements C i,k The smaller the value, the W i,k The bigger, the better.

[0132] The Hungarian algorithm is executed with the cost matrix C as input to obtain the optimal assignment set Π = {(i, k(i)) | i = 1, …, N}; where k(i) represents the sub-region that matches the i-th product, and N represents the number of products.

[0133] We need to assign N items to K sub-regions one by one, maximizing the total weight. Using the cost matrix C as input, we call the Hungarian algorithm to obtain the optimal assignment set.

[0134] In one embodiment of the present invention, the product advertisement recommendation process based on the matching result includes: playing the product advertisement of the i-th product in the matching sub-area.

[0135] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A product advertising recommendation system based on an e-commerce platform, characterized by: include: Data acquisition module, used to obtain ambient temperature, relative humidity and the temperature behind the screen of the advertising display; The first processing module is used to calculate the dew point temperature according to the ambient temperature and relative humidity; The second processing module is used to compare the dew point temperature with the temperature behind the screen to obtain the fog increment prediction value; A third processing module is used to divide the display area of ​​the advertising display into M×N sub-areas; Playing a double-stripe reference frame on the advertising display, and collecting a brightness difference between the brightness values ​​of the first stripe reference frame and the second stripe reference frame to generate a scattering coefficient for each sub-region based on the brightness difference; A fourth processing module is configured to generate a readability weight matrix by combining the scattering coefficient of the k-th sub-region, the fog increment prediction value, and the preset readability coefficient; The fifth processing module is used to calculate the urgency of the i-th product based on the normalized scrap cost, remaining shelf life, inventory level, and predicted demand of the i-th product; The advertising recommendation module is used to perform matching optimization based on the readable weight matrix and urgency, using the Hungarian algorithm to obtain matching results between products and sub-regions, and recommend product advertisements based on the matching results.

2. The product advertisement recommendation system based on the e-commerce platform according to claim 1, characterized in that: The dew point temperature is calculated based on the ambient temperature and relative humidity as follows: Where γ represents the temperature-humidity coupling coefficient, RH represents relative humidity, T represents ambient temperature, and 17.27 and 237.7 are both Magnus empirical constants. Among them, T d Indicates the dew point temperature.

3. The product advertisement recommendation system based on the e-commerce platform according to claim 2, characterized in that: Compare the dew point temperature with the temperature behind the screen to get the fog increment prediction value, including: ΔT cond =T d -T s Where, ΔT cond represents the condensation driving temperature difference, T s Indicates the temperature behind the screen; Δζ pred =Norm(max(0,ΔT cond )) Among them, Δζ pred Represents the fog increment prediction value, Norm represents the normalization process, and max represents the max function.

4. The product advertisement recommendation system based on the e-commerce platform according to claim 3, characterized in that: Playing a double-stripe reference frame on an advertising display and collecting a brightness difference between a brightness value of a first stripe reference frame and a brightness value of a second stripe reference frame to generate a scattering coefficient of each sub-region based on the brightness difference, including: Step 41: The advertisement player continuously plays two frames of a first fringe reference frame and a second fringe reference frame with the same frequency and a phase difference of π within a preset time period. Step 42 , respectively collecting the grayscale matrix S1 and the grayscale matrix S2 corresponding to the first fringe reference frame and the second fringe reference frame; Step 43: Divide the display area into M×N identical sub-areas; Step 44: extract the first average grayscale and the second average grayscale of the grayscale matrix S1 and the grayscale matrix S2 corresponding to the kth sub-region respectively, and perform difference processing to obtain the brightness difference value ΔL of the kth sub-region. k ; Step 45: When the relative humidity is 20%, repeat steps 41 to 44 to obtain the reference brightness difference ΔL. ref ; Step 46: Scattering coefficient Δζ of the kth sub-region k , k is a positive integer.

5. The product advertisement recommendation system based on the e-commerce platform according to claim 4 is characterized in that: Combining the scattering coefficient of the kth sub-region, the fog increment prediction value and the preset readability coefficient, a readability weight matrix is ​​generated, including: Calculate the sum of the scattering coefficient of the kth sub-area and the fog increment prediction value; If the sum is greater than 1, the composite scattering value ζ of the kth sub-region is k,syn Set to 1; If the sum is less than 0, the composite scattering value ζ of the kth sub-region is k,syn Set to 0; Otherwise, the sum is set to the composite scattering value ζ of the kth sub-region k,syn ; Obtain P product advertisement templates, read the font size and weight of each advertisement template, perform clustering processing to form J readability clusters, and configure a readability coefficient for each readability cluster; the readability coefficient value range is: (0,1]; The readability weight is calculated based on the composite scattering of the kth sub-region and the jth readability cluster as follows: Among them, R j,k Represents readable weight, e represents the natural base, c represents the decay constant, A j represents the readability coefficient of the jth readability cluster, 1≤j≤J, j is a positive integer; Based on readable weight R j,k Forming a readable weight matrix.

6. The product advertisement recommendation system based on the e-commerce platform according to claim 5, characterized in that: Obtain the normalized scrap cost, remaining shelf life, inventory, and predicted demand of the i-th item, and calculate the urgency of the i-th item, including: The predicted demand is obtained based on the output of the pre-trained LSTM neural network model; The urgency of the i-th product is as follows: Among them, U i Indicates the urgency of the i-th product, Q i represents the scrapping cost of the i-th commodity, R i represents the remaining shelf life of the i-th product, S i represents the inventory of the i-th commodity, λ i represents the predicted demand for the i-th commodity.

7. The product advertisement recommendation system based on the e-commerce platform according to claim 6, characterized in that: Based on the readability weight matrix and urgency, the Hungarian algorithm is used for matching optimization to obtain the matching results of products and sub-regions, including: Determine the readability cluster to which the product advertisement of the i-th product belongs and the readability weight of the k-th sub-region; The product of the urgency of the i-th product, the readability cluster to which the product advertisement of the i-th product belongs, and the readability weight of the k-th sub-region is used as the comprehensive weight W i,k ; Based on the comprehensive weight W i,k , construct the cost matrix C; where the cost C in the i-th row and k-th column of the cost matrix C is i,k , C i,k =1-W i,k ; The Hungarian algorithm is executed with the cost matrix C as input to obtain the optimal assignment set Π = {(i, k(i)) | i = 1, …, N}; where k(i) represents the sub-region that matches the i-th product, and N represents the number of products.

8. The product advertisement recommendation system based on the e-commerce platform according to claim 7, characterized in that: The product advertisement recommendation process is performed based on the matching result, including: playing the product advertisement of the i-th product in the matching sub-area.

Images