Advertisement automatic recommendation system based on video acquisition information in elevator
By collecting information and historical data in the elevator video, combining Attention technology and PINN model, an advertisement allocation model is built, which solves the problems of content boredom and limited audience in the advertising recommendation system in the elevator, and achieves high coverage and targeted advertising Delivery improves the effectiveness of advertising and return on investment.
Patent Information
- Application Number
- CN202410877822.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-02
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing advertising recommendation system in the elevator has problems such as bored content, limited audience and poor quality of advertising content, resulting in low advertising promotion results.
An automatic advertising recommendation system based on video collection information in the elevator is adopted. Passenger data and store consumption information are analyzed through the data acquisition module, combined with the data of the historical analysis module, and the advertising allocation model is constructed using Attention technology and PINN model to accurately match advertising content.
It achieves high coverage and targeted advertising, improves the effectiveness and return on investment, and reduces initial investment and maintenance costs.
Smart Images

Figure CN119067728A_ABST
Abstract
Description
Technical Field
[0001] The invention discloses an automatic advertisement recommendation system based on video collection information in an elevator, and relates to the field of advertisement recommendation. Background Art
[0002] The existing systems for recommending advertisements in elevators have the following shortcomings:
[0003] Content boredom: If the advertising content in the elevator is not updated in a timely manner or the same advertisement is played repeatedly, passengers will feel bored and pay less attention to the advertisement.
[0004] Limited audience: The audience of elevator advertising is mainly limited to people who take elevators. The audience range of fixed or inappropriate elevator advertising is relatively limited and is not suitable for advertising promotion.
[0005] Inconsistent quality of advertising content: Poor advertising promotion using an advertising matching system or method based only on a single factor such as age, gender or advertising content will result in poor advertising that does not match the interests of passengers or is poorly produced, reducing passengers' acceptance and favorability of the ads. Summary of the invention
[0006] In view of the deficiencies in the prior art, the present invention aims to provide an automatic advertisement recommendation system based on video collection information in elevators, aiming to solve the problem of low efficiency of advertisement recommendation.
[0007] In order to achieve the above object, the present invention is implemented through the following technical solutions: An automatic advertising recommendation system based on video collection information in an elevator comprises:
[0008] Data acquisition module: used to acquire and analyze the elevator video of the target building, count the number of people entering and exiting each floor of the target building; identify the gender and age of passengers in the elevator video and obtain passenger data;
[0009] Obtain planning information for the target building, calculate the consumption proportion of each store on each floor of the target building, and the total consumption proportion of different consumption types in the target building;
[0010] Obtain the advertising information of the target building, and obtain the recommended product names and product consumption types;
[0011] Historical analysis module: used to obtain the historical elevator videos and historical advertising information of the target building in the past three months;
[0012] Obtain the historical business data of each store on the target building in the past three months;
[0013] Extract the historical number of people exiting and entering the elevator corresponding to the historical video, extract the historical recommended commodity names and historical commodity consumption types corresponding to the historical advertising information, and calculate the priority allocation coefficient group and the change coefficient group in combination with the historical business data;
[0014] Joint recommendation module: used to build an advertising allocation model based on the coefficient of variation group using Attention technology and a simple PINN model;
[0015] Substitute passenger data, recommended product names and product consumption types into the advertising allocation model for fitting to obtain model recommended advertisements; calculate the simulated consumption proportion corresponding to the model recommended advertisements and compare it with the consumption proportion;
[0016] If the comparison result between the simulated consumption ratio and the consumption ratio is similar, the advertisements are selected in the local advertisement library according to the total consumption ratio and the priority allocation coefficient group;
[0017] If the comparison result between the simulated consumption ratio and the consumption ratio is not similar, then the local advertising library will be supplemented with advertisements based on the total consumption ratio;
[0018] User interaction module: used to summarize the priority allocation coefficient group and the advertisements supplemented in the local advertisement library, and feed back to the user.
[0019] Furthermore, the workflow of the data acquisition module is as follows:
[0020] Process A1: Obtain elevator video and count the number of people entering and exiting each floor of the target building;
[0021] Process A11: Read the planning information of the target building from the database; the planning information of the target building includes: the total number of floors of the target building and the store information of each floor of the target building; the total number of floors is recorded as ln;
[0022] Database: stores the planning information of the target building;
[0023] Process A12: The elevator videos corresponding to the 1st to the lnth floors of the target building are obtained through the camera, and the original videos 1 to ln are obtained. The AdaBoost algorithm is used to count the number of people leaving and entering the elevator corresponding to the original videos 1 to ln, and they are recorded as o and i respectively to obtain:
[0024] The number of people leaving the 1st floor is o1, and the number of people entering is i1;
[0025] The number of people leaving the 2nd floor is o2, and the number of people entering is i2;
[0026] By analogy, the number of people coming out of the ln layer is o ln , number of people entering i ln ;
[0027] Process A2: Use the FairFace model to identify the gender and estimated age of passengers in original videos 1 to ln; divide the passengers into teenagers, young adults, middle-aged adults, and elderly people according to their estimated ages;
[0028] Summarize the data obtained from process A1 to process A2 as passenger data;
[0029] Process A3: According to the reference consumption type, the number of stores on each target floor is summarized to obtain:
[0030] The number of different shops on the first floor: a1, b1, c1, d1, e1, f1, g1;
[0031] The number of different shops on the 2nd floor: a2, b2, c2, d2, e2, f2, g2;
[0032] Similarly, the number of different stores on the ln layer is: a ln 、b ln 、c ln ,d ln 、e ln 、f ln , g ln ;
[0033] Among them, a, b, c, d, e, f and g represent: food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores respectively;
[0034] a1, b1, c1, d1, e1, f1, g1, respectively, represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores, and other stores on the first floor;
[0035] By analogy, a ln 、b ln 、c ln ,d ln 、e ln 、f ln , g ln , which respectively represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores corresponding to the ln layer.
[0036] Furthermore, the subsequent process of process A3 is as follows:
[0037] Process A4: Calculate the number of stores on each floor from 1st to lnth floor, denoted as n1, n2 to n ln ; Calculate n1~n ln The sum of is denoted as n;
[0038] Define calculation formula 11: b(x)=x / y; where x and y represent calculation parameters of calculation formula 11, and b(x) refers to the proportion corresponding to x;
[0039] Substitute a1, b1, c1, d1, e1, f1 and g1 into x in formula 11, substitute n1 into y in formula 11, calculate the consumption proportion corresponding to different types of consumption on the first floor, and obtain b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1);
[0040] By analogy, a ln 、b ln 、c ln d ln 、e ln 、f ln and g ln Substitute the x in equation 11 one by one, and change n ln Substitute y into equation 11 and calculate the consumption proportion corresponding to different consumption levels in the ln layer to obtain b(a ln )、b(b ln )、b(c ln )、b(d ln )、b(e ln )、b(f ln ) and b(g ln );
[0041] Process A5: Calculate the total number of different types of consumer stores on the 1st to lnth floors, and record them as aa, bb, cc, dd, ee, ff and gg respectively;
[0042] Substitute aa, bb, cc, dd, ee, ff and gg into x in formula 11, substitute n into y in formula 11, calculate the total consumption proportion corresponding to different types of consumption in commercial buildings, and obtain b(aa), b(bb), b(cc), b(dd), b(ee), b(ff) and b(gg);
[0043] Process A6: Using the KaeyBERT algorithm, extract the advertising information in the advertising area of the original video 1 to the original video ln, and obtain the recommended product name and product consumption type of the advertisement;
[0044] Process A7: Summarize the data obtained from processes A1 to A6 and enter the historical analysis module.
[0045] Furthermore, the workflow of the historical analysis module is as follows:
[0046] Process B1: Obtain the target building's elevator videos from the database for the past three months;
[0047] Database: store historical elevator videos and historical business data;
[0048] Process B2: Obtain the historical business data of each store on the target building in the past three months through the database, and obtain the historical operating profit and historical total sales volume corresponding to each floor (target building):
[0049] Historical operating profit corresponding to layer 1: h(a 1 1)~h(a 1 3) h(b 1 1)~h(b 1 3) h(c 1 1)~h(c 1 3) h(d 1 1)~h(d 1 3) h(e 1 1)~h(e 1 3) h(f 1 1)~h(f 1 3) and h(g 1 1)~h(g 1 3); The total historical sales volume corresponding to the first floor (target building): h(a 1 )、h(b 1 )、h(c 1 )、h(d 1 )、h(e 1 )、h(f 1 ) and h(g 1 );
[0050] By analogy, the historical operating profit corresponding to the ln floor (target building) is: h(a ln 1)~h(a ln 3) h(b ln 1)~h(b ln 3) h(c ln 1)~h(c ln 3) h(d ln 1)~h(d ln 3) h(e ln 1)~h(e ln 3) h(f ln 1)~h(f ln 3) and h(g ln 1)~h(g ln 3); The total historical sales volume corresponding to the ln floor (target building): h(a ln )、h(b ln )、h(c ln )、h(d ln )、h(e ln )、h(f ln ) and h(gln );
[0051] Process B3: Using the AdaBoost algorithm, count the number of people leaving and entering the elevator in the historical elevator videos, and record them as ho and hi respectively to get:
[0052] The number of people leaving the 1st floor is ho1, and the number of people entering is hi1; the number of people leaving the 2nd floor is ho2, and the number of people entering is hi2; and so on, the number of people leaving the lnth floor is ho ln , number of people entering ln ;
[0053] Process B4: Based on the data obtained from processes B1 to B3, a priority allocation coefficient group and a variation coefficient group are calculated;
[0054] Process B5: Summarize the priority allocation coefficient group and the variation coefficient group obtained in processes B1 to B4 and enter the joint recommendation module.
[0055] Furthermore, the specific process of process B4 is as follows:
[0056] Process B41: Calculate the priority allocation coefficient group;
[0057] Process B42: Calculate the priority allocation coefficients of all stores on the first floor;
[0058] Process B421: 1 layer h(a 1 1)~h(a 1 3) The corresponding store is recorded as the target store; define formula 21 to calculate the economic growth rate up (m 1-2 ) and up(m 2-3 );
[0059] Calculation 21: Where m represents the historical operating profit of the target store, x and y represent the subscripts of m, and up(m x-y ) represents the economic growth rate of the target store;
[0060] h(a 1 2) and h(a 1 1) Substitute into equation 21 in turn to obtain up(m 1-2 );
[0061] h(a 1 3) and h(a 1 2) Substitute into equation 21 one by one and get up(m 2-3 );
[0062] Process B422: Get the number of days in the past three months, recorded as α, β and γ; get the daily operating profit of the target store in the past three months:
[0063] In the past month, the daily operating profit h(a 1 11)~h(a 1 1α);
[0064] In the past two months, the daily operating profit h(a 1 21)~h(a 1 2β);
[0065] In the past three months, the daily operating profit h(a 1 31)~h(a 1 3γ);
[0066] Calculate the average daily operating profit of the target store in the past three months, and record it as:
[0067] In the past month, the average daily operating profit h(aa 1 1);
[0068] In the past two months, the average daily operating profit h(aa 1 2);
[0069] In the past three months, the average daily operating profit h(aa 1 3);
[0070] Process B423: Define relation 22: h(aa i j )≤h(a i j x); i and j both represent 1 to 3; x represents α, β and γ;
[0071] The corresponding numbers of target stores that satisfy equation 22 in the past three months are summarized and recorded as α1, β1 and γ1.
[0072] Furthermore, the subsequent process of process B423 is as follows:
[0073] Process B424: construct a state transfer matrix, recorded as matrix 23; the expression of matrix 23 is as follows:
[0074] Wherein, n represents the number of iterations of the matrix 23, and × represents matrix multiplication; n is a natural number, and the initial value of n is 1; each time the matrix 23 is iterated once, n is increased by 1 on the original basis;
[0075] Iterate the matrix 23 using the NumPy library function to obtain a steady-state transfer matrix, which is recorded as a matrix 24; read the outputs of the matrix 24 column by column, which are recorded as upa1 and upa2, upb1 and upb2, and upc1 and upc2 respectively;
[0076] Process B425: Define calculation formula 25: Wherein, tpn represents the transition parameter, x represents upa1 to upc1, and y represents upa2 to upc2;
[0077] Use formula 25 to calculate the transition parameters tpn1, tpn2 and tpn3 of the target store in the past three months; calculate the average value tpn(a 1 ), as the priority allocation coefficient of the target store;
[0078] Process B426: Repeat processes B421 to B425 to calculate h(a) of layer 1 1 1)~h(a 1 3) The same process as the store priority allocation coefficient is used to calculate the priority allocation coefficient tpn(b) of all stores on the first floor. 1 )~tpn(g 1 );
[0079] Process B43: Repeat the same process of calculating the priority allocation coefficients of all stores on the 1st floor to calculate the priority allocation coefficients of all stores on the 2nd to lnth floors:
[0080] Priority allocation coefficient for all shops on the second floor: tpn(a 2 )~tpn(g 2 );
[0081] Priority allocation coefficient for all stores on the 3rd floor: tpn(a 3 )~tpn(g 3 );
[0082] Similarly, the priority allocation coefficient of all stores on the ln layer is: tpn(a ln )~tpn(g ln );
[0083] Summarize the priority allocation coefficients of all stores on the 1st to lnth floors to obtain a priority allocation coefficient group;
[0084] Process B44: Calculate the coefficient of variation group.
[0085] Furthermore, the specific process of process B44 is as follows:
[0086] Process B441: Using the KaeyBERT algorithm, extract the information of the advertising area in the historical elevator video by layer to obtain historical advertising information 1 to historical advertising information ln;
[0087] The historical advertisement information includes: historical advertisement pictures and historical advertisement names;
[0088] Process B442: Based on historical advertising information 1, count the number and total number of different types of goods on the first floor (target building);
[0089] Process B4421: Record the historical advertisement pictures and historical advertisement names in the historical advertisement information 1 as target pictures and target names, respectively, and count the number of target names, recorded as nn1, that is, the total number of different types of commodities is nn1;
[0090] Process B4422: According to the target image and target name, extract the commodity consumption type corresponding to the target name and record it as the target type;
[0091] Process B4423: Count the number of commodities corresponding to different target types, denoted as fa1, fb1, fc1, fd1, fe1, ff1 and fg1;
[0092] Among them, fa1, fb1, fc1, fd1, fe1, ff1 and fg1 are all greater than or equal to 0, fa1+fb1+fc1+fd1+fe1+ff1+fg1=nn1;
[0093] Process B443: Repeat process B442 to count the number and total number of different types of goods on the 2nd to lnth floors (target building);
[0094] The number of different types of goods on the 2nd level: fa2, fb2, fc2, fd2, fe2, ff2 and fg2; the total number nn2;
[0095] Similarly, the number of different types of goods in the ln layer: fa ln 、fb ln 、fc ln 、fd ln 、 ln ,ff ln and fg ln ; Total nn ln ;
[0096] Process B444: Calculate the coefficient of variation of all stores on the first floor;
[0097] Process B4441: Get the number of people leaving ho1 and entering hi1 on level 1;
[0098] Process B4442: Define equation 26; substitute fa1, fb1, fc1, fd1, fe1, ff1, and fg1 into x in equation 26. j , calculate the coefficient of variation b(fa1)~b(fg1) of different types of commodities on the first floor; the calculation formula 26 is as follows:
[0099] Among them, x j represents the calculation parameter of formula 26; j represents the subscript of x, and the value range of j is: 1~ln; nno represents nn1; ho j andhij They represent the number of people leaving and entering the j-floor respectively;
[0100] Process B4443: Summarize the calculation results of process B4442 to obtain the coefficient of variation of layer 1;
[0101] Process B445: Obtain the number of people entering and exiting the 2nd to lnth floors, and expand the range of nno in formula 26 to nn2 to nn ln ; Repeat the same process of calculating the coefficient of variation of layer 1 to calculate the coefficient of variation of layers 2 to ln;
[0102] The variation coefficients of the 1 to ln layers are summarized to obtain a variation coefficient group.
[0103] Furthermore, the workflow of the joint recommendation module is as follows:
[0104] Process C1: Please refer to Figure 3 , import the simple PINN model; based on the simple PINN model, build an advertising allocation model;
[0105] Process C11: Define the input layer: used to receive the input data of the model, namely passenger data, recommended product names and product consumption types;
[0106] Embedding layer: calls the t-SNE algorithm to convert the input data into a low-dimensional vector to obtain the embedding vector;
[0107] Process C12: Add pooling function in the pooling layer and concatenation layer;
[0108] Define the loss function:
[0109] Where L represents the average loss value of the model; S represents the training set with a batch size of N, y∈{0,1} is the training set label; p(x) represents the output of the PINN model, that is, the predicted probability of the sample being clicked; x represents the input of the PINN model;
[0110] Process C13: Using the change coefficient group as the product information change condition of the first hidden layer and the second hidden layer to obtain an advertising allocation model;
[0111] Process C2: Substitute passenger data, recommended product names and product consumption types into the advertising allocation model for fitting to obtain model recommended advertisements;
[0112] Calculate the simulated consumption ratio corresponding to the advertisements recommended by the model and compare it with the consumption ratio;
[0113] Process C3: The number of advertisements recommended by the statistical model is recorded as mn; the product names corresponding to the advertisements recommended by the model are extracted and recorded as simulated names;
[0114] According to the simulation type, the statistical model recommends the number of different types of goods in the advertisement, which are denoted as ma, mb, mc, md, me, mf and mg respectively;
[0115] Among them, ma, mb, mc, md, me, mf and mg are all greater than or equal to 0, ma+mb+mc+md+me+mf+mg=mn;
[0116] Process C4: define calculation formula 31: b(m)=m / n; wherein m and n represent calculation parameters of calculation formula 31, and b(m) represents the ratio of m to n;
[0117] Substitute ma, mb, mc, md, me, mf and mg into m in equation 31, and substitute mn into n in equation 31, to calculate the simulated consumption proportion b(ma)~b(mg) corresponding to the advertisement recommended by the model;
[0118] Process C5: define judgment formula 32: p≥q; wherein p and q represent comparison parameters of judgment formula 32;
[0119] Get the consumption proportions b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1) corresponding to different consumption types at level 1;
[0120] Substitute b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1) into p in judgment formula 32, and substitute ma, mb, mc, md, me, mf and mg into q in judgment formula 32; count the number of consumption proportions of the first floor that satisfy judgment formula 32, recorded as gg1;
[0121] Process C6: Obtain the consumption proportions corresponding to different consumption types in the 2nd to lnth layers (process A4), repeat process C5, and count the number of consumption proportions in the 2nd to lnth layers that satisfy the judgment formula 32, and record them as gg2~gg ln .
[0122] Furthermore, the subsequent process of process C6 is as follows:
[0123] Process C7: Get the number of stores n1, n2~n on each floor from 1st to lnth floor ln ;
[0124] Comparison formula 33 is defined as: lim(x / y)→1; wherein x and y represent comparison parameters of comparison formula 33 respectively;
[0125] Change gg1, gg2 to ggln Substitute the x in comparison formula 33 one by one, and change n1, n2 to n ln Substitute y into the comparison formula 33 one by one; count the number of conditions that satisfy the comparison formula 33, and record it as on;
[0126] Substitute on and ln into x and y in comparison formula 33, respectively, to determine whether comparison formula 33 is satisfied; ln represents the total number of floors;
[0127] If it is satisfied, the advertisements are screened in the local advertisement library according to the total consumption ratio and the priority allocation coefficient group, and the process C8 is entered;
[0128] If it is not satisfied, then the local advertising library is supplemented with advertisements according to the total consumption ratio, and the process C9 is entered;
[0129] Process C8: Screening advertisements in the local advertisement library;
[0130] Process C81: Obtain the total consumption proportion corresponding to different consumption types in the commercial building (process A5);
[0131] Obtain the priority allocation coefficients of all stores on the 1st to lnth floors (process B43);
[0132] Process C82: Define calculation formula 34: Z(m i )=b(m)*tpn(m i );
[0133] Among them, i represents the floor, and the value range of i is: 1~ln; m represents the consumption type, that is, a~g; Z(m i ) represents the priority coefficient of type m consumption in layer i; b(m) represents the total consumption proportion of type m consumption; tpn(m i ) represents the priority allocation coefficient of the m-type consumer store on the i-th floor;
[0134] Process C83: Substitute the total consumption proportion and the priority allocation coefficient into the calculation formula 34 to obtain the priority coefficient Z (a 1 )~Z(g ln );
[0135] Scheme C84: Screening of Z(a 1 )~Z(g ln ) as the standard value; extract the floor and consumption type corresponding to the standard value as the priority floor and priority consumption type;
[0136] In the local advertising library, select the advertisements corresponding to the priority consumption type and play them on the priority floor;
[0137] Process C9: replenishing advertisements for the local advertisement library;
[0138] Repeat process C81 to process C83 to calculate Z(a 1 )~Z(g ln ) process, screening Z(a 1 )~Z(g ln ) as the standard value; extract the consumption type corresponding to the standard value as the supplementary type; and filter the advertisements in the Meta advertisement database and import them into the local advertisement library according to the supplementary type.
[0139] Compared with the prior art, the present invention has the following beneficial effects:
[0140] High coverage and targeting: The present invention can collect current and historical video information in the elevator and analyze passenger data, so as to accurately customize advertising content and improve the targeting and effect of advertising.
[0141] Cost-effectiveness: Compared with traditional advertising media, the initial investment and maintenance costs of the elevator advertising of the present invention are relatively low. At the same time, due to its high coverage and targeting, advertisers can obtain a higher return on investment.
[0142] Technological integration and innovation: This invention combines the latest digital technologies, such as AI intelligent recognition and big data analysis, to adjust advertising content according to the audience's gender, age and even economic growth, to achieve highly personalized advertising and improve the effectiveness of advertising. BRIEF DESCRIPTION OF THE DRAWINGS
[0143] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:
[0144] Figure 1 It is a schematic diagram of the system of the present invention;
[0145] Figure 2 This is a schematic diagram of the advertising area of the present invention;
[0146] Figure 3 It is a schematic diagram of a simple PINN model of the present invention. DETAILED DESCRIPTION
[0147] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0148] See also Figure 1 An automatic advertising recommendation system based on video collection information in an elevator includes: a data acquisition module, a history analysis module, a joint recommendation module, a user interaction module, a database and a server; wherein the data acquisition module, the history analysis module, the joint recommendation module and the user interaction module are connected to the database and the server respectively.
[0149] Database: used to store the planning information of the target building, historical elevator videos, and historical business data; historical business data includes: historical operating profits and historical total sales of different stores. Data acquisition module: used to acquire and analyze the elevator video of the target building, count the number of people leaving (elevator) and entering (elevator) on each floor of the target building; identify the gender and age of passengers in the elevator video to obtain passenger data;
[0150] Obtain planning information of the target building (commercial building), calculate the consumption proportion of each store on each floor of the target building and the total consumption proportion of different consumption types in the target building;
[0151] Obtain the advertising information of the target building, and obtain the recommended product names and product consumption types;
[0152] Process A: The workflow of the data acquisition module is as follows:
[0153] Process A1: Obtain elevator video and count the number of people entering and exiting the elevator on each floor of the target building;
[0154] Process A11: Read the planning information of the target building (commercial building) from the database; the planning information of the target building (commercial building) includes: the total number of floors of the target building and the information of the shops (selling goods) on each floor of the target building; the total number of floors is recorded as ln;
[0155] Process A12: The elevator videos corresponding to the 1st to the lnth floors of the target building are obtained through the camera, and the original videos 1 to ln are obtained. The AdaBoost algorithm is used to count the number of people leaving and entering the elevator corresponding to the original videos 1 to ln, and they are recorded as o and i respectively to obtain:
[0156] The number of people leaving (elevator) on the 1st floor (target building) is o1, and the number of people entering (elevator) is i1;
[0157] The number of people leaving (elevator) on the 2nd floor (target building) is o2, and the number of people entering (elevator) is i2;
[0158] By analogy, the number of people exiting (elevator) on the ln floor (target building) is o ln , the number of people entering (elevator) i ln ;
[0159] It should be noted that the "target building (i.e., service object)" in the present invention is a "commercial building" by default; if the service object of the present invention is not a "commercial building", the commercial building closest to the service object is selected as the standard building, and under the (data acquisition) authorization of the relevant person in charge of the standard building, the planning information of the standard building is read by default; if the relevant person in charge of the standard building is not authorized, the process of selecting the "standard building" is repeated until the relevant person in charge of the standard building authorizes it.
[0160] Process A2: Use the FairFace model (open source model) to identify the gender and estimated age of passengers in original videos 1 to ln (exiting and entering the elevator); divide the passengers into teenagers, young adults, middle-aged adults, and elderly people according to their estimated age (i.e., if the estimated age is ≤ 18, it is a teenager; if the age is 18 < estimated age ≤ 30, it is a young person; if the age is 30 < estimated age ≤ 60, it is an adult; and if the age is 60 < estimated age, it is an elderly person);
[0161] Summarize the data obtained from process A1 to process A2 as passenger data;
[0162] It should be noted that the FairFace model is an "open source model specifically designed for identifying the gender and estimating the age of people's images."
[0163] Process A3: According to the reference consumption types (food consumption, clothing consumption, daily necessities consumption, digital consumption, education consumption, entertainment consumption and other consumption), the number of stores on each target floor is summarized to obtain:
[0164] The number of different shops on the first floor (commercial building): a1, b1, c1, d1, e1, f1, g1;
[0165] Different number of shops on the 2nd floor (commercial building): a2, b2, c2, d2, e2, f2, g2;
[0166] Similarly, the number of different stores on the ln floor (commercial building): a ln 、b ln 、c ln ,d ln 、e ln 、f ln , g ln ;
[0167] Among them, a, b, c, d, e, f and g represent: food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores respectively;
[0168] a1, b1, c1, d1, e1, f1, g1, respectively, represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores, and other stores on the first floor (commercial building);
[0169] a2, b2, c2, d2, e2, f2, and g2 represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores, and other stores on the second floor (commercial building), respectively;
[0170] By analogy, a ln 、b ln 、c ln ,dln 、e ln 、f ln , g ln , respectively represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores corresponding to the lnth floor (commercial building);
[0171] It should be noted that the "reference consumption type" in the present invention refers to food consumption, clothing consumption, daily necessities consumption, digital consumption, education consumption, entertainment consumption and other consumption; users or relevant technical personnel can adjust the reference scope of the "reference consumption type" according to actual needs;
[0172] "Educational Stores" refers to stores such as educational institutions, tutoring institutions, training institutions, etc. that directly or indirectly improve personal learning ability or specialized skills;
[0173] "Other stores" refers to stores such as hardware stores, gyms, laundries, etc. that do not provide or do not fully provide "food consumption, clothing consumption, daily necessities consumption, digital consumption, educational consumption and entertainment consumption."
[0174] Process A4: Calculate the number of stores on each floor from 1st to lnth floor (commercial building), denoted as n1, n2 to n ln ; Calculate n1~n ln The sum of is denoted as n;
[0175] Define calculation formula 11: b(x)=x / y; where x and y represent calculation parameters of calculation formula 11, and b(x) refers to the proportion corresponding to x;
[0176] Substitute a1, b1, c1, d1, e1, f1 and g1 into x in formula 11, substitute n1 into y in formula 11, calculate the consumption proportions corresponding to different types of consumption (food consumption, clothing consumption, daily necessities consumption, digital consumption, education consumption, entertainment consumption and other consumption) on the first floor (commercial building), and obtain b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1);
[0177] Substitute a2, b2, c2, d2, e2, f2 and g2 into x in equation 22, substitute n2 into y in equation 11, calculate the corresponding consumption proportions of different types of consumption (food consumption, clothing consumption, daily necessities consumption, digital consumption, education consumption, entertainment consumption and other consumption) on the second floor (commercial building), and obtain b(a2), b(b2), b(c2), b(d2), b(e2), b(f2) and b(g2);
[0178] By analogy, a ln 、b ln 、cln ,d ln 、e ln 、f ln and g ln Substitute the x in equation 11 one by one, and change n ln Substitute y into equation 11 and calculate the corresponding consumption proportion of different types of consumption (food consumption, clothing consumption, daily necessities consumption, digital consumption, education consumption, entertainment consumption and other consumption) on the ln floor (commercial building), and get b(a ln )、b(b ln )、b(c ln )、b(d ln )、b(e ln )、b(f ln ) and b(g ln );
[0179] Process A5: Calculate the total number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores in the 1st to lnth floors (commercial buildings) in turn, and record them as aa, bb, cc, dd, ee, ff and gg respectively;
[0180] Substitute aa, bb, cc, dd, ee, ff and gg into x in formula 11, substitute n into y in formula 11, calculate the total consumption proportion corresponding to different types of consumption (food consumption, clothing consumption, daily necessities consumption, digital consumption, education consumption, entertainment consumption and other consumption) in commercial buildings, and obtain b(aa), b(bb), b(cc), b(dd), b(ee), b(ff) and b(gg);
[0181] Process A6: Please refer to Figure 2 , using the KaeyBERT algorithm, extract the advertising information in the advertising area of the original video 1 to the original video ln;
[0182] Advertising information includes: advertisement (product) pictures and advertisement (product) names;
[0183] Please refer to Table 1, the commodity classification table; in Table 1, use the SQL statement 1 of "select commodity consumption type from commodity classification table where advertisement (commodity) picture = commodity picture and advertisement (commodity) name = commodity name" to query the commodity consumption type corresponding to the advertisement (commodity) picture and the advertisement (commodity) name;
[0184] The advertisement (product) name is recorded as the recommended product name;
[0185]
[0186]
[0187] Table 1 (Commodity Classification Table)
[0188] It should be noted that the "commodity classification table" in the present invention is provided by users or relevant technical personnel.
[0189] Process A7: Summarize the data obtained from processes A1 to A6 and enter the historical analysis module.
[0190] Historical analysis module: used to obtain the historical elevator videos and historical advertising information of the target building in the past three months; obtain the historical business data of each store on the target building in the past three months;
[0191] Extract the historical number of people exiting and entering the elevator corresponding to the historical elevator video, extract the historical recommended commodity names and historical commodity consumption types corresponding to the historical advertising information, and calculate the priority allocation coefficient group and the change coefficient group in combination with the historical business data;
[0192] Process B: The workflow of the historical analysis module is as follows:
[0193] Process B1: Obtain the target building's elevator videos from the database for the past three months;
[0194] Database: store historical elevator videos and historical business data;
[0195] Process B2: Obtain the historical business data of each store on the target building in the past three months through the database, and obtain the historical operating profit and historical total sales volume corresponding to each floor (target building):
[0196] Historical operating profit corresponding to the first floor (target building): h(a 1 1)~h(a 1 3) h(b 1 1)~h(b 1 3) h(c 1 1)~h(c 1 3) h(d 1 1)~h(d 1 3) h(e 1 1)~h(e 1 3) h(f 1 1)~h(f 1 3) and h(g 1 1)~h(g 1 3); The total historical sales volume corresponding to the first floor (target building): h(a 1 )、h(b 1 )、h(c 1 )、h(d 1 )、h(e 1 )、h(f 1 ) and h(g 1);
[0197] Historical operating profit corresponding to the second floor (target building): h(a 2 1)~h(a 2 3) h(b 2 1)~h(b 2 3) h(c 2 1)~h(c 2 3) h(d 2 1)~h(d 2 3) h(e 2 1)~h(e 2 3) h(f 2 1)~h(f 2 3) and h(g 2 1)~h(g 2 3); The total historical sales volume corresponding to the 2nd floor (target building): h(a 2 )、h(b 2 )、h(c 2 )、h(d 2 )、h(e 2 )、h(f 2 ) and h(g 2 );
[0198] By analogy, the historical operating profit corresponding to the ln floor (target building) is: h(a ln 1)~h(a ln 3) h(b ln 1)~h(b ln 3) h(c ln 1)~h(c ln 3) h(d ln 1)~h(d ln 3) h(e ln 1)~h(e ln 3) h(f ln 1)~h(f ln 3) and h(g ln 1)~h(g ln 3); The total historical sales volume corresponding to the ln floor (target building): h(a ln )、h(b ln )、h(c ln )、h(d ln )、h(e ln )、h(f ln ) and h(g ln );
[0199] Among them, h(a 1 1)~h(a 1 3) h(b 11)~h(b 1 3) h(c 1 1)~h(c 1 3) h(d 1 1)~h(d 1 3) h(e 1 1)~h(e 1 3) h(f 1 1)~h(f 1 3) and h(g 1 1)~h(g 1 3) and h(a 1 )、h(b 1 )、h(c 1 )、h(d 1 )、h(e 1 )、h(f 1 ) and h(g 1 ), which respectively represent the operating profits and total sales of the food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores corresponding to the first floor (target building) in the past three months. The superscript represents the floor and the subscript represents the time. Similarly, h(a ln 1)~h(a ln 3) h(b ln 1)~h(b ln 3) h(c ln 1)~h(c ln 3) h(d ln 1)~h(d ln 3) h(e ln 1)~h(e ln 3) h(f ln 1)~h(f ln 3) and h(g ln 1)~h(g ln 3) and h(a ln )、h(b ln )、h(c ln )、h(d ln )、h(e ln )、h(f ln ) and h(g ln ), which represent the operating profits and total sales of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores corresponding to the ln floor (target building) in the past three months;
[0200] Process B3: Using the AdaBoost algorithm, count the number of people leaving and entering the elevator in the historical elevator videos, and record them as ho and hi respectively to get:
[0201] The number of people leaving (elevator) on the 1st floor (target building) is ho1, and the number of people entering (elevator) is hi1;
[0202] The number of people leaving (elevator) on the 2nd floor (target building) is ho2, and the number of people entering (elevator) is hi2;
[0203] By analogy, the number of people exiting (elevator) on the ln floor (target building) is ho ln , number of people entering (elevator)hi ln ;
[0204] Process B4: Based on the data obtained from processes B1 to B3, a priority allocation coefficient group and a variation coefficient group are calculated;
[0205] Process B41: Calculate the priority allocation coefficient group;
[0206] Process B42: Calculate the priority allocation coefficients of all stores on the first floor (target building);
[0207] Process B421: 1st floor (target building) h(a 1 1)~h(a 1 3) The corresponding store is recorded as the target store; define formula 21 to calculate the economic growth rate up (m 1-2 ) and up(m 2-3 );
[0208] Calculation 21: Where m represents the historical operating profit of the target store, x and y represent the subscripts of m, and up(m x-y ) represents the economic growth rate of the target store;
[0209] h(a 1 2) and h(a 1 1) Substitute into equation 21 in turn to obtain up(m 1-2 );
[0210] h(a 1 3) and h(a 1 2) Substitute into equation 21 one by one and get up(m 2-3 );
[0211] It should be noted that the above “h(a 1 1)~h(a 1 3)” means: h(a 1 1) to h(a 1 3), that is, h(a 1 1) h(a 1 1) and h(a 1 3); The subsequent data containing “~” all indicate the same type of parameters where “only the subscript increases in natural number order”.
[0212] Process B422: Get the number of days in the past three months, recorded as α, β and γ; get the daily operating profit of the target store in the past three months:
[0213] In January (the past month), daily operating profit h(a 1 11)~h(a 1 1α);
[0214] In February (the last two months), daily operating profit h(a 1 21)~h(a 1 2β);
[0215] In March (the past three months), daily operating profit h(a 1 31)~h(a 1 3γ);
[0216] Calculate the average daily operating profit of the target store in the past three months, and record it as:
[0217] In January (the past month), the average daily operating profit h(aa 1 1);
[0218] In February (the past two months), the average daily operating profit h(aa 1 2);
[0219] In March (three months in the past), the average daily operating profit h(aa 1 3)
[0220] Process B423: Define relation 22: h(aa i j )≤h(a i j x); i and j both represent 1 to 3; x represents α, β and γ;
[0221] Summarize the corresponding numbers of target stores that satisfy equation 22 in the past three months, and record them as α1, β1, and γ1;
[0222] Process B424: Summarize the data in processes B43 to B46 and construct a state transfer matrix, recorded as matrix 23; the expression of matrix 23 is as follows:
[0223] Wherein, n represents the number of iterations of the (state transfer) matrix 23 (the power operation on the left side of the matrix 23), and × represents matrix multiplication; n is a natural number, and the initial value of n is 1; each time the (state transfer) matrix 23 is iterated once, n is increased by 1 on the original basis;
[0224] Iterate the (state transfer) matrix 23 using the NumPy library function to obtain a steady-state transfer matrix, recorded as a matrix 24; read the output of the matrix 24 column by column, recorded as upa1 and upa2, upb1 and upb2, and upc1 and upc2 respectively;
[0225] Process B425: Define calculation formula 25: Wherein, tpn represents the transition parameter, x represents upa1 to upc1, and y represents upa2 to upc2;
[0226] Use formula 25 to calculate the transition parameters tpn1, tpn2 and tpn3 of the target store in the past three months; calculate the average value tpn(a 1 ), as the target store (h(a 1 1)~h(a 1 3) The priority allocation coefficient of the corresponding store;
[0227] Process B426: Repeat processes B421 to B425 to calculate the h(a) of the first floor (target building) 1 1)~h(a 1 3) The same process as the store priority allocation coefficient is used to calculate the priority allocation coefficient tpn (b 1 )~tpn(g 1 );
[0228] Process B43: Repeat the same process (process B2) of calculating the priority allocation coefficients of all stores on the 1st floor (target building) to calculate the priority allocation coefficients of all stores on the 2nd floor to the lnth floor (target building):
[0229] Priority allocation coefficient for all stores on the second floor (target building): tpn(a 2 )~tpn(g 2 );
[0230] Priority allocation coefficient for all stores on the 3rd floor (target building): tpn(a 3 )~tpn(g 3 );
[0231] Similarly, the priority allocation coefficient of all stores on the ln floor (target building) is: tpn(a ln )~tpn(g ln );
[0232] Summarize the priority allocation coefficients of all stores on the 1st to lnth floors (target building) to obtain a priority allocation coefficient group;
[0233] Process B44: Calculate the coefficient of variation group;
[0234] Process B441: using the KaeyBERT algorithm, extracting historical advertising information of the advertising area in the historical elevator video by layer, and obtaining historical advertising information 1 to historical advertising information ln;
[0235] Historical advertising information includes: historical advertising (product) pictures and historical advertising (product) names;
[0236] Process B442: Based on historical advertising information 1, count the number and total number of different types of goods on the first floor (target building);
[0237] Process B4421: Record the historical advertisement (product) image and the historical advertisement (product) name in the historical advertisement information 1 as the target image and the target name respectively; count the number of target names, record it as nn1, that is, the total number of products is nn1;
[0238] According to Table 1 (the above process A6), replace "advertisement (product) picture" in SQL statement 1 with "target picture" and "advertisement (product) name" with "target name"; repeat SQL statement 1 to extract the product consumption type corresponding to the target name, and record it as the target type;
[0239] Process B4422: According to the target type, count the number of different types of goods (food goods, clothing goods, daily necessities goods, digital goods, educational goods, entertainment goods and other goods), and record them as fa1, fb1, fc1, fd1, fe1, ff1 and fg1 respectively; fa1, fb1, fc1, fd1, fe1, ff1 and fg1 are all greater than or equal to 0, fa1+fb1+fc1+fd1+fe1+ff1+fg1=nn1;
[0240] Process B443: Repeat process B442 to count the number and total number of different types of goods on the 2nd to lnth floors (target building);
[0241] The number of different types of goods on the 2nd floor (destination building): fa2, fb2, fc2, fd2, fe2, ff2 and fg2; the total number nn2;
[0242] Similarly, the number of different types of goods on the ln floor (target building): fa ln 、fb ln 、fc ln 、fd ln 、 ln ,ff ln and fg ln ; Total nn ln ;
[0243] Process B444: Calculate the coefficient of variation of all stores on the 1st floor (target building);
[0244] Process B4441: Get the number of people leaving (elevator) ho1 and the number of people entering (elevator) hi1 on the 1st floor (target building);
[0245] Process B4442: Define equation 26; substitute fa1, fb1, fc1, fd1, fe1, ff1, and fg1 into x in equation 26. j , calculate the coefficients of variation b(fa1)~b(fg1) of different types of goods on the first floor (target building); calculation formula 26 is as follows:
[0246] Among them, x j represents the calculation parameter of formula 26; j represents the subscript of x, and the value range of j is: 1~ln; nno represents nn1; ho j andhi j They represent the number of people leaving (elevator) and entering (elevator) on the jth floor respectively;
[0247] Process B4443: Summarize the calculation results of process B4442 to obtain the coefficient of variation of layer 1;
[0248] Process B445: Get the number of people leaving and entering the elevator on the 2nd to lnth floors (target building), and expand the range of nno in formula 26 to nn2 to nn ln ; Repeat the same process of calculating the coefficient of variation of the 1st floor (target building) to calculate the coefficient of variation of the 2nd to lnth floors;
[0249] Summarize the variation coefficients of layers 1 to ln to obtain a variation coefficient group;
[0250] Process B5: Summarize the priority allocation coefficient group and the variation coefficient group obtained in processes B1 to B4 and enter the joint recommendation module.
[0251] Joint recommendation module: used to build an advertising allocation model based on the coefficient of variation group using Attention technology and a simple PINN model;
[0252] Substitute passenger data, recommended product names and product consumption types into the advertising allocation model for fitting to obtain model recommended advertisements; calculate the simulated consumption proportion corresponding to the model recommended advertisements and compare it with the consumption proportion;
[0253] If the comparison result between the simulated consumption ratio and the consumption ratio is similar, the advertisements are selected in the local advertisement library according to the total consumption ratio and the priority allocation coefficient group;
[0254] If the comparison result between the simulated consumption ratio and the consumption ratio is not similar, then the local advertising library will be supplemented with advertisements based on the total consumption ratio;
[0255] Process C: The workflow of the joint recommendation module is as follows:
[0256] Process C1: Please refer to Figure 3 , import the simple PINN model; based on the simple PINN model, build an advertising allocation model;
[0257] Process C11: Define the input layer: used to receive the input data of the model, namely passenger data, recommended product names and product consumption types;
[0258] Embedding layer: calls the t-SNE algorithm to convert the input data into a low-dimensional vector to obtain the embedding vector;
[0259] Process C12: Add pooling function in the pooling layer and concatenation layer;
[0260] Define the loss function:
[0261] Where L represents the average loss value of the model; S represents the training set with a batch size of N, y∈{0,1} is the training set label; p(x) represents the output of the PINN model, that is, the predicted probability of the sample being clicked; x represents the input of the PINN model;
[0262] Process C13: Using the change coefficient group as the product information change condition of the first hidden layer and the second hidden layer to obtain an advertising allocation model;
[0263] Process C2: Substitute passenger data, recommended product names and product consumption types into the advertising allocation model for fitting to obtain model recommended advertisements;
[0264] Calculate the simulated consumption ratio corresponding to the advertisements recommended by the model and compare it with the consumption ratio;
[0265] Process C3: The number of advertisements recommended by the statistical model is recorded as mn; the product names corresponding to the advertisements recommended by the model are extracted and recorded as simulated names;
[0266] Replace "advertisement (product) name" in SQL statement 1 with "simulation name"; repeat SQL statement 1 to extract the product consumption type corresponding to the simulation name, and record it as the simulation type;
[0267] According to the simulation type, the statistical model recommends the number of different types of goods (food goods, clothing goods, daily necessities goods, digital goods, educational goods, entertainment goods and other goods) in the advertisement, which are recorded as ma, mb, mc, md, me, mf and mg respectively;
[0268] Among them, ma, mb, mc, md, me, mf and mg are all greater than or equal to 0, ma+mb+mc+md+me+mf+mg=mn;
[0269] Process C4: define calculation formula 31: b(m)=m / n; wherein m and n represent calculation parameters of calculation formula 31, and b(m) represents the ratio of m to n;
[0270] Substitute ma, mb, mc, md, me, mf and mg into m in equation 31, and substitute mn into n in equation 31, to calculate the simulated consumption proportion b(ma)~b(mg) corresponding to the advertisement recommended by the model;
[0271] Process C5: define judgment formula 32: p≥q; wherein p and q represent comparison parameters of judgment formula 32;
[0272] Obtain the consumption proportions b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1) corresponding to different consumption types at level 1; (Process A4)
[0273] Substitute b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1) into p in judgment formula 32, and substitute ma, mb, mc, md, me, mf and mg into q in judgment formula 32; count the number of consumption proportions of the first floor that satisfy judgment formula 32, recorded as gg1;
[0274] Process C6: Obtain the consumption proportions corresponding to different consumption types in the 2nd to lnth layers (process A4), repeat process C5, and count the number of consumption proportions in the 2nd to lnth layers that satisfy the judgment formula 32, and record them as gg2~gg ln ;
[0275] Process C7: Get the number of stores n1, n2~n on each floor from 1st to lnth floor ln ;
[0276] Comparison formula 33 is defined as: lim(x / y)→1; wherein x and y represent comparison parameters of comparison formula 33 respectively;
[0277] Change gg1, gg2 to gg ln Substitute the x in comparison formula 33 one by one, and change n1, n2 to n ln Substitute y into the comparison formula 33 one by one; count the number of conditions that satisfy the comparison formula 33, and record it as on;
[0278] Substitute on and ln into x and y in comparison formula 33, and determine whether comparison formula 33 is satisfied.
[0279] If it is satisfied, it means that the comparison result of the simulated consumption ratio and the consumption ratio is similar. According to the total consumption ratio and the priority allocation coefficient group, the advertisements are selected in the local advertisement library and the process C8 is entered;
[0280] If not, it means that if the comparison result of the simulated consumption ratio and the consumption ratio is not similar, then the local advertising library is supplemented with advertisements according to the total consumption ratio, and the process C9 is entered;
[0281] Process C8: Screening advertisements in the local advertisement library;
[0282] Process C81: Obtain the total consumption proportion corresponding to different consumption types in the commercial building (process A5);
[0283] Obtain the priority allocation coefficients of all stores on the 1st to lnth floors (process B43);
[0284] Process C82: Define calculation formula 34: Z(m i )=b(m)*tpn(m i );
[0285] Among them, i represents the floor, and the value range of i is: 1~ln; m represents the consumption type, that is, a~g; Z(m i ) represents the priority coefficient of type m consumption in layer i; b(m) represents the total consumption proportion of type m consumption; tpn(m i ) represents the priority allocation coefficient of the m-type consumer store on the i-th floor;
[0286] Process C83: Substitute the total consumption proportion and the priority allocation coefficient into the calculation formula 34 to obtain the priority coefficient Z (a 1 )~Z(g ln );
[0287] Scheme C84: Screening of Z(a 1 )~Z(g ln ) as the standard value; extract the floor and consumption type corresponding to the standard value as the priority floor and priority consumption type;
[0288] In the local advertising library, select the advertisements corresponding to the priority consumption type and play them on the priority floor;
[0289] Process C9: replenishing advertisements for the local advertisement library;
[0290] Repeat process C81 to process C83 to calculate Z(a 1 )~Z(g ln ) process, screening Z(a 1 )~Z(g ln ) as the standard value; extract the consumption type corresponding to the standard value as the supplementary type; and filter the advertisements in the Meta advertisement database and import them into the local advertisement library according to the supplementary type.
[0291] It should be noted that the "local advertising library" in the present invention refers to the database for storing advertisements in commercial buildings; the number of filtered advertisements and the number of supplementary advertisements are both set to 100 by default, and users or relevant technical personnel can adjust the number of filtered advertisements and the number of supplementary advertisements according to actual needs.
[0292] User interaction module: used to summarize the priority allocation coefficient group and the advertisements supplemented in the local advertisement library, and feed back to the user.
[0293] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technical personnel in this field according to actual conditions. For example, if there are weight coefficients and proportional coefficients, their set sizes are to quantify each parameter to obtain a specific value, which is convenient for subsequent comparison. Regarding the size of the weight coefficient and the proportional coefficient, it is sufficient as long as it does not affect the proportional relationship between the parameter and the quantized value.
[0294] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. An automatic advertising recommendation system based on video collection information in elevators, characterized in that: The recommended system includes: Data acquisition module: used to acquire and analyze the elevator video of the target building, count the number of people entering and exiting each floor of the target building; identify the gender and age of passengers in the elevator video and obtain passenger data; Obtain planning information for the target building, calculate the consumption proportion of each store on each floor of the target building, and the total consumption proportion of different consumption types in the target building; Obtain the advertising information of the target building, and obtain the recommended product names and product consumption types; Historical analysis module: used to obtain the historical elevator videos and historical advertising information of the target building in the past three months; Obtain the historical business data of each store on the target building in the past three months; Extract the historical number of people exiting and entering the elevator corresponding to the historical video, extract the historical recommended commodity names and historical commodity consumption types corresponding to the historical advertising information, and calculate the priority allocation coefficient group and the change coefficient group in combination with the historical business data; Joint recommendation module: used to build an advertising allocation model based on the coefficient of variation group using Attention technology and a simple PINN model; Substitute passenger data, recommended product names and product consumption types into the advertising allocation model for fitting to obtain model recommended advertisements; calculate the simulated consumption proportion corresponding to the model recommended advertisements and compare it with the consumption proportion; If the comparison results of the simulated consumption ratio and the consumption ratio are similar, the advertisements are screened in the local advertisement library according to the total consumption ratio and the priority allocation coefficient group; if the comparison results of the simulated consumption ratio and the consumption ratio are not similar, the advertisements are supplemented to the local advertisement library according to the total consumption ratio; User interaction module: used to summarize the priority allocation coefficient group and the advertisements supplemented in the local advertisement library, and feed back to the user.
2. The automatic advertising recommendation system based on video collection information in elevators according to claim 1 is characterized in that: The workflow of the data acquisition module is as follows: Process A1: Obtain elevator video and count the number of people entering and exiting each floor of the target building; Process A11: Read the planning information of the target building through the database; The planning information of the target building includes: the total number of floors of the target building and the store information of each floor of the target building; the total number of floors is recorded as ln; Database: stores the planning information of the target building; Process A12: Obtain the elevator videos corresponding to the 1st to lnth floors of the target building, and obtain original videos 1 to ln; use the AdaBoost algorithm to count the number of people exiting and entering the elevator corresponding to original videos 1 to ln, and record them as o and i respectively to obtain: The number of people leaving the 1st floor is o1, and the number of people entering is i1; The number of people leaving the 2nd floor is o2, and the number of people entering is i2; By analogy, the number of people coming out of the ln layer is o ln , number of people entering i ln ; Process A2: Use the FairFace model to identify the gender and estimated age of passengers in original videos 1 to ln; Summarize the data obtained from process A1 to process A2 as passenger data; Process A3: According to the reference consumption type, the number of stores on each target floor is summarized to obtain: The number of different shops on the first floor: a1, b1, c1, d1, e1, f1, g1; The number of different shops on the 2nd floor: a2, b2, c2, d2, e2, f2, g2; Similarly, the number of different stores on the ln layer is: a ln 、b ln 、c ln ,d ln 、e ln 、f ln , g ln ; Among them, a, b, c, d, e, f and g represent: food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores respectively; a1, b1, c1, d1, e1, f1, g1, respectively, represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores, and other stores on the first floor; By analogy, a ln 、b ln 、c ln ,d ln 、e ln 、f ln , g ln , which respectively represent the number of food stores, clothing stores, daily necessities stores, digital stores, education stores, entertainment stores and other stores corresponding to the ln layer.
3. The automatic advertisement recommendation system based on video collection information in an elevator according to claim 2, characterized in that: The subsequent process of process A3 is as follows: Process A4: Calculate the number of stores on each floor from 1st to lnth floor, denoted as n1, n2 to n ln ; Calculate n1~n ln The sum of is denoted as n; Define calculation formula 11: b(x)=x / y; where x and y represent calculation parameters of calculation formula 11, and b(x) refers to the proportion corresponding to x; Substitute a1, b1, c1, d1, e1, f1 and g1 into x in formula 11, substitute n1 into y in formula 11, calculate the consumption proportion corresponding to different types of consumption on the first floor, and obtain b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1); By analogy, a ln 、b ln 、c ln d ln 、e ln 、f ln and g ln Substitute the x in equation 11 one by one, and change n ln Substitute y into equation 11 and calculate the consumption proportion corresponding to different consumption levels in the ln layer to obtain b(a ln )、b(b ln )、b(c ln )、b(d ln )、b(e ln )、b(f ln ) and b(g ln ); Process A5: Calculate the total number of different types of consumer stores on the 1st to lnth floors, and record them as aa, bb, cc, dd, ee, ff and gg respectively; Substitute aa, bb, cc, dd, ee, ff and gg into x in formula 11, substitute n into y in formula 11, calculate the total consumption proportion corresponding to different types of consumption in commercial buildings, and obtain b(aa), b(bb), b(cc), b(dd), b(ee), b(ff) and b(gg); Process A6: Using the KaeyBERT algorithm, extract the advertising information in the advertising area of the original video 1 to the original video ln, and obtain the recommended product name and product consumption type of the advertisement; Process A7: Summarize the data obtained from processes A1 to A6 and enter the historical analysis module.
4. The automatic advertisement recommendation system based on video collection information in elevators according to claim 3 is characterized in that: The workflow of the historical analysis module is as follows: Process B1: Obtain the historical elevator videos of the target building in the past three months; Process B2: Obtain the historical business data of each store on the target building in the past three months, and obtain the historical operating profit and historical total sales volume corresponding to each floor: Historical operating profit corresponding to layer 1: h(a 1 1)~h(a 1 3) h(b 1 1)~h(b 1 3) h(c 1 1)~h(c 1 3) h(d 1 1)~h(d 1 3) h(e 1 1)~h(e 1 3) h(f 1 1)~h(f 1 3) and h(g 1 1)~h(g 1 3); The total historical sales volume corresponding to the first layer: h(a 1 )、h(b 1 )、h(c 1 )、h(d 1 )、h(e 1 )、h(f 1 ) and h(g 1 ); By analogy, the historical operating profit corresponding to the ln layer is: h(a ln 1)~h(a ln 3) h(b ln 1)~h(b ln 3) h(c ln 1)~h(c ln 3) h(d ln 1)~h(d ln 3) h(e ln 1)~h(e ln 3) h(f ln 1)~h(f ln 3) and h(g ln 1)~h(g ln 3); The total historical sales volume corresponding to the ln layer: h(a ln )、h(b ln )、h(c ln )、h(d ln )、h(e ln )、h(f ln ) and h(g ln ); Process B3: Using the AdaBoost algorithm, count the number of people leaving and entering the elevator in the historical elevator videos, and record them as ho and hi respectively to get: The number of people leaving the 1st floor is ho1, and the number of people entering is hi1; the number of people leaving the 2nd floor is ho2, and the number of people entering is hi2; and so on, the number of people leaving the lnth floor is ho ln , number of people entering ln ; Process B4: Based on the data obtained from processes B1 to B3, a priority allocation coefficient group and a variation coefficient group are calculated; Process B5: Summarize the priority allocation coefficient group and the variation coefficient group obtained in processes B1 to B4 and enter the joint recommendation module.
5. The automatic advertisement recommendation system based on video collection information in elevators according to claim 4 is characterized in that: The specific process of process B4 is as follows: Process B41: Calculate the priority allocation coefficient group; Process B42: Calculate the priority allocation coefficients of all stores on the first floor; Process B421: 1 layer h(a 1 1)~h(a 1 3) The corresponding store is recorded as the target store; define formula 21 to calculate the economic growth rate up (m 1-2 ) and up(m 2-3 ); Calculation 21: Where m represents the historical operating profit of the target store, x and y represent the subscripts of m, and up(m x-y ) represents the economic growth rate of the target store; h(a 1 2) and h(a 1 1) Substitute into equation 21 in turn to obtain up(m 1-2 ); h(a 1 3) and h(a 1 2) Substitute into equation 21 one by one and get up(m 2-3 ); Process B422: Get the number of days in the past three months, recorded as α, β and γ; get the daily operating profit of the target store in the past three months: In the past month, the daily operating profit h(a 1 11)~h(a 1 1α); In the past two months, the daily operating profit h(a 1 21)~h(a 1 2β); In the past three months, the daily operating profit h(a 1 31)~h(a 1 3γ); Calculate the average daily operating profit of the target store in the past three months, and record it as: In the past month, the average daily operating profit h(aa 1 1); In the past two months, the average daily operating profit h(aa 1 2); In the past three months, the average daily operating profit h(aa 1 3); Process B423: Define relation 22: h(aa i j )≤h(a i j x); wherein i and j represent 1 to 3; x represents α, β and γ; The corresponding numbers of target stores that satisfy equation 22 in the past three months are summarized and recorded as α1, β1 and γ1.
6. The automatic advertisement recommendation system based on video collection information in elevators according to claim 5, characterized in that: The subsequent process of process B423 is as follows: Process B424: construct a state transfer matrix, recorded as matrix 23; the expression of matrix 23 is as follows: Wherein, n represents the number of iterations of the matrix 23, and × represents matrix multiplication; n is a natural number, and the initial value of n is 1; each time the matrix 23 is iterated once, n is increased by 1 on the original basis; Iterate the matrix 23 using the NumPy library function to obtain a steady-state transfer matrix, which is recorded as a matrix 24; read the outputs of the matrix 24 column by column, which are recorded as upa1 and upa2, upb1 and upb2, and upc1 and upc2 respectively; Process B425: Define calculation formula 25: Wherein, tpn represents the transition parameter, x represents upa1 to upc1, and y represents upa2 to upc2; Use formula 25 to calculate the transition parameters tpn1, tpn2 and tpn3 of the target store in the past three months; calculate the average value tpn(a 1 ), as the priority allocation coefficient of the target store; Process B426: Repeat processes B421 to B425 to calculate h(a) of layer 1 1 1)~h(a 1 3) The same process as the store priority allocation coefficient is used to calculate the priority allocation coefficient tpn(b) of all stores on the first floor. 1 )~tpn(g 1 ); Process B43: Repeat the same process of calculating the priority allocation coefficients of all stores on the 1st floor to calculate the priority allocation coefficients of all stores on the 2nd to lnth floors: Priority allocation coefficient for all shops on the second floor: tpn(a 2 )~tpn(g 2 ); Priority allocation coefficient for all stores on the 3rd floor: tpn(a 3 )~tpn(g 3 ); Similarly, the priority allocation coefficient of all stores on the ln layer is: tpn(a ln )~tpn(g ln ); Summarize the priority allocation coefficients of all stores on the 1st to lnth floors to obtain a priority allocation coefficient group; Process B44: Calculate the coefficient of variation group.
7. The automatic advertisement recommendation system based on video collection information in elevators according to claim 6, characterized in that: The specific process of process B44 is as follows: Process B441: Using the KaeyBERT algorithm, extract the information of the advertising area in the historical elevator video by layer to obtain historical advertising information 1 to historical advertising information ln; The historical advertisement information includes: historical advertisement pictures and historical advertisement names; Process B442: According to historical advertising information 1, count the number and total number of different types of goods at level 1; Process B4421: Record the historical advertisement pictures and historical advertisement names in the historical advertisement information 1 as target pictures and target names, respectively, and count the number of target names, recorded as nn1; Process B4422: According to the target image and target name, extract the commodity consumption type corresponding to the target name and record it as the target type; Process B4423: Count the number of commodities corresponding to different target types, denoted as fa1, fb1, fc1, fd1, fe1, ff1 and fg1; Among them, fa1, fb1, fc1, fd1, fe1, ff1 and fg1 are all greater than or equal to 0, fa1+fb1+fc1+fd1+fe1+ff1+fg1=nn1; Process B443: Repeat process B442 to count the number and total number of different types of goods on the 2nd to lnth layers; The number of different types of goods on the 2nd level: fa2, fb2, fc2, fd2, fe2, ff2 and fg2; the total number nn2; Similarly, the number of different types of goods in the ln layer: fa ln 、fb ln 、fc ln 、fd ln 、 ln ,ff ln and fg ln ; Total nn ln ; Process B444: Calculate the coefficient of variation of all stores on the first floor; Process B4441: Get the number of people leaving ho1 and entering hi1 on level 1; Process B4442: Define equation 26; substitute fa1, fb1, fc1, fd1, fe1, ff1, and fg1 into x in equation 26. j , calculate the coefficient of variation b(fa1)~b(fg1) of different types of commodities on the first floor; the calculation formula 26 is as follows: Among them, x j represents the calculation parameter of formula 26; j represents the subscript of x, and the value range of j is: 1~ln; nno represents nn1; ho j andhi j They represent the number of people leaving and entering the j-floor respectively; Process B4443: Summarize the calculation results of process B4442 to obtain the coefficient of variation of layer 1; Process B445: Obtain the number of people entering and exiting the 2nd to lnth floors, and expand the range of nno in formula 26 to nn2 to nn ln ; Repeat the same process of calculating the coefficient of variation of layer 1 to calculate the coefficient of variation of layers 2 to ln; The variation coefficients of the 1 to ln layers are summarized to obtain a variation coefficient group.
8. The automatic advertisement recommendation system based on video collection information in an elevator according to claim 4, characterized in that: The workflow of the joint recommendation module is as follows: Process C1: Import the simple PINN model; build an advertising allocation model based on the simple PINN model; Process C11: Define the input layer: used to receive the input data of the model, namely passenger data, recommended product names and product consumption types; Embedding layer: calls the t-SNE algorithm to convert the input data into a low-dimensional vector to obtain the embedding vector; Process C12: Add pooling function in the pooling layer and concatenation layer; Define the loss function: Where L represents the average loss value of the model; S represents the training set with a batch size of N, y∈{0,1} is the training set label; p(x) represents the output of the PINN model, that is, the predicted probability of the sample being clicked; x represents the input of the PINN model; Process C13: Using the change coefficient group as the product information change condition of the first hidden layer and the second hidden layer to obtain an advertising allocation model; Process C2: Substitute passenger data, recommended product names and product consumption types into the advertising allocation model for fitting to obtain model recommended advertisements; Calculate the simulated consumption ratio corresponding to the advertisements recommended by the model and compare it with the consumption ratio; Process C3: The number of advertisements recommended by the statistical model is recorded as mn; the product names corresponding to the advertisements recommended by the model are extracted and recorded as simulated names; According to the simulation type, the statistical model recommends the number of different types of goods in the advertisement, which are denoted as ma, mb, mc, md, me, mf and mg respectively; Among them, ma, mb, mc, md, me, mf and mg are all greater than or equal to 0, ma+mb+mc+md+me+mf+mg=mn; Process C4: define calculation formula 31: b(m)=m / n; wherein m and n represent calculation parameters of calculation formula 31, and b(m) represents the ratio of m to n; Substitute ma, mb, mc, md, me, mf and mg into m in equation 31, and substitute mn into n in equation 31, to calculate the simulated consumption proportion b(ma)~b(mg) corresponding to the advertisement recommended by the model; Process C5: define judgment formula 32: p≥q; wherein p and q represent comparison parameters of judgment formula 32; Get the consumption proportions b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1) corresponding to different consumption types at level 1; Substitute b(a1), b(b1), b(c1), b(d1), b(e1), b(f1) and b(g1) into p in judgment formula 32, and substitute ma, mb, mc, md, me, mf and mg into q in judgment formula 32; count the number of consumption proportions of the first floor that satisfy judgment formula 32, recorded as gg1; Process C6: Obtain the consumption proportions corresponding to different consumption types from layer 2 to layer ln, repeat process C5, and count the number of consumption proportions from layer 2 to layer ln that satisfy judgment formula 32, and record them as gg2~gg ln .
9. The automatic advertisement recommendation system based on video collection information in elevators according to claim 8, characterized in that: The subsequent process of process C6 is as follows: Process C7: Get the number of stores n1, n2~n on each floor from 1st to lnth floor ln ; Comparison formula 33 is defined as: lim(x / y)→1; wherein x and y represent comparison parameters of comparison formula 33 respectively; Change gg1, gg2 to gg ln Substitute the x in comparison formula 33 one by one, and change n1, n2 to n ln Substitute y into the comparison formula 33 one by one; count the number of conditions that satisfy the comparison formula 33, and record it as on; Substitute on and ln into x and y in comparison formula 33, respectively, to determine whether comparison formula 33 is satisfied; ln represents the total number of floors; If it is satisfied, the advertisements are screened in the local advertisement library according to the total consumption ratio and the priority allocation coefficient group, and the process C8 is entered; If it is not satisfied, then the local advertising library is supplemented with advertisements according to the total consumption ratio, and the process C9 is entered; Process C8: Screening advertisements in the local advertisement library; Process C81: Obtain the total consumption proportion corresponding to different consumption types in the commercial building; Get the priority allocation coefficients of all stores on the 1st to lnth floors; Process C82: Define calculation formula 34: Z(m i )=b(m)*tpn(m i ); Among them, i represents the floor, and the value range of i is: 1~ln; m represents the consumption type, that is, a~g; Z(m i ) represents the priority coefficient of type m consumption in layer i; b(m) represents the total consumption proportion of type m consumption; tpn(m i ) represents the priority allocation coefficient of the m-type consumer store on the i-th floor; Process C83: Substitute the total consumption proportion and the priority allocation coefficient into the calculation formula 34 to obtain the priority coefficient Z (a 1 )~Z(g ln ); Scheme C84: Screening of Z(a 1 )~Z(g ln ) as the standard value; extract the floor and consumption type corresponding to the standard value as the priority floor and priority consumption type; In the local advertising library, select the advertisements corresponding to the priority consumption type and play them on the priority floor; Process C9: replenishing advertisements for the local advertisement library; Repeat process C81 to process C83 to calculate Z(a 1 )~Z(g ln ) process, screening Z(a 1 )~Z(g ln ) as the standard value; extract the consumption type corresponding to the standard value as the supplementary type; and filter the advertisements in the Meta advertisement database and import them into the local advertisement library according to the supplementary type.
Citation Information
Patent Citations
Advertisement putting method and system, electronic equipment and storage medium
CN113393260A
Advertisement activity search outdoor advertisement system based on regional information collection
CN114493692A
Advertisement recommendation method and device based on big data
CN115034817A
Advertisement putting management system for elevator operation
CN116188083A
System and method for providing recommendations based on consumer location
US20200098010A1
