An Interactive Elevator Advertising Intelligent Recommendation Method and System

Abstract

The present invention discloses an interactive elevator advertisement intelligent recommendation method and system. The recommendation method includes: collecting passenger information through a monitoring camera in the elevator and collecting passenger interaction scenario information through a touch advertisement screen; analyzing the passenger information based on a passenger attribute recognition algorithm; storing the interaction scenario information in a database module, and constructing an advertisement information set and an interaction scenario database according to the interaction scenario information; on the basis of the interaction scenario database, intelligently recommending advertisements to passengers based on a scenario-aware FP-growth recommendation algorithm; obtaining associated advertisements that target passengers in the entire elevator cluster are interested in, and locally correcting the results; finding the target passengers and recommending advertisements to the target passengers. Based on a touch-type advertisement display screen, combining passenger attributes and interaction information, the present invention designs an interactive elevator advertisement intelligent recommendation method and system with certain robustness and high accuracy, and realizes the maximization of advertisement placement benefits.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator advertising, and in particular to an interactive elevator advertising intelligent recommendation method and system. Background Art

[0002] With the increase in the number of domestic elevators, the value of data information generated in elevators has attracted more and more attention. As a new type of advertising medium, elevator advertising can directly reach community users and utilize the massive information generated in elevators to better serve users. It is necessary to design an elevator advertising intelligent recommendation system that meets the application scenarios.

[0003] Existing research on elevator advertising intelligent recommendation mainly focuses on the one-way analysis of the recommendation system, such as based on passenger attribute detection, etc., to achieve intelligent advertisement placement. However, it lacks the correlation analysis of the interaction behavior data between passengers and advertisements, and fails to maximize the advertising placement benefit. Summary of the Invention

[0004] The present invention provides an interactive elevator advertising intelligent recommendation method and system to solve the problem in the prior art that the correlation analysis of the interaction behavior data between passengers and advertisements is lacking and the advertising placement benefit is not maximized.

[0005] On the premise of having a touch-type advertising display screen, this application combines passenger attributes (system initiative analysis) and interaction information (interaction behaviors such as passengers swiping the display screen and clicking on advertisements, and passenger initiative feedback information), and further explores on the basis of existing research to design an interactive elevator advertising intelligent recommendation method and system with certain robustness and high accuracy, which is beneficial for users to obtain better advertising services and achieve the maximization of advertising placement benefits.

[0006] The technical solution adopted by the present invention is: to provide an interactive elevator advertising intelligent recommendation method, including the following steps:

[0007] Collect passenger information through a monitoring camera in the elevator, and collect passenger interaction scenario information through a touch advertising screen;

[0008] Analyze the passenger information based on a passenger attribute recognition algorithm;

[0009] Store the interaction scenario information in a database module, and construct an advertisement information set and an interaction scenario database according to the interaction scenario information;

[0010] Based on the interaction scenario database, intelligently recommend advertisements to passengers based on the scenario-aware FP-growth recommendation algorithm;

[0011] Obtain the associated advertisements that target passengers in the entire elevator cluster are interested in, and locally correct the results;

[0012] Find the target passengers and recommend advertisements to the target passengers.

[0013] As a preferred way of the intelligent recommendation method for interactive elevator advertisements, the method for constructing the advertisement information set and the interactive scenario database according to the interactive scenario information includes:

[0014] Construct the advertisement information set: S = {s1, s2,..., s m}, where s i = {k1, k2, k i ,....K m}, 1 ≤ i ≤ m, s i is the sequence of advertisements k i concerned by the target passenger within the time period t1. The start time of the time period t1 is when the passenger clicks on the advertisement interface, and the end time is when the passenger leaves the elevator;

[0015] Construct the interactive scenario database: D = {d1, d2,..., d n}, where d i = {[x, y]: s1, s2,...}, 1 ≤ i ≤ n, d i represents the specific interactive scenario data, that is, the thing database composed of the target passenger and the advertisement information set, where [x, y] represents the target passenger attributes, x is the gender attribute, and y is the age attribute.

[0016] As a preferred way of the intelligent recommendation method for interactive elevator advertisements, during the process of constructing the advertisement information set, time thresholds t2 and t3 are added; the continuous loss of passenger information by the camera within the time period t2 indicates that the passenger has left the elevator, and the advertisement screen automatically exits the interactive interface until the next click is received and then starts constructing the advertisement information set again; if the duration from the passenger clicking on advertisement k i to switching the interface is greater than t3, then k i is added to s i , otherwise it is not added.

[0017] As a preferred way of the intelligent recommendation method for interactive elevator advertisements, the method for intelligently recommending advertisements to passengers based on the scenario-aware FP-growth recommendation algorithm on the basis of the interactive scenario database includes:

[0018] Build the item header table, that is, the number of occurrences of each k i in the advertisement information set S, define the minimum support as T MS , T MS ∈ (0, 1); scan the advertisement information set S to obtain the count of all frequent 1-item sets (k1, k2, k i ,....) ( C k2, C ki .....); then delete the items with support lower than the threshold T MS Put the frequent 1-item sets into the item header table and sort them in descending order of support;

[0019] Construct the FP-tree, scan the advertisement information set, filter out the non-frequent 1-item sets from the read original data and sort them in descending order of support; read the sorted data set and insert it into the FP-tree in the sorted order. The nodes sorted earlier are the root nodes, and the later ones are the child nodes; if there are shared root nodes, the corresponding count is incremented by 1; if new nodes appear, the corresponding nodes in the item header table will link the new nodes through the node linked list; until all the data is inserted into the FP-tree;

[0020] Mine the FP-tree, find the conditional pattern bases corresponding to the item header table items from the bottom items of the item header table in turn; recursively mine the frequent item sets of the item header table items; define the number of items N in the frequent item set FP , and only return the frequent item sets that meet the item number requirements;

[0021] Define the global weighted parameter, perform batch normalization on the occurrence times of the specific items in the maximum frequent item set as the global weighted parameter corresponding to each specific item.

[0022] As a preferred way of the intelligent recommendation method for interactive elevator advertisements, the method for locally correcting the results includes:

[0023] Cluster the interactive data scenario information of specific points based on the K-means algorithm. The size of the cluster is equal to the passenger attribute category, and perform batch normalization on the number of times the advertisement appears in each cluster;

[0024] Weight the same-category products in the clustering results and re-sort them; define the advertisement attention threshold T P , and retain the advertisement items with scores higher than T P to generate a new advertisement catalog.

[0025] As a preferred way of the intelligent recommendation method for interactive elevator advertisements, the method for finding target passengers includes:

[0026] After the passenger enters the elevator, detect the passenger's position based on the deep learning algorithm, detect the position of the human eyes based on the opencv human eye detection model (haarcascade_eye.xml), perform template matching according to the human eye state library, and detect the position of the human eye's attention for the passenger. The human eye state library includes 5 types of attention to the front, left, right, up, and down. Since the camera is located above the advertisement screen, when the passenger faces the advertisement screen, the human eye state is considered to be the front, and establish a target passenger model:

[0027] F(x) = αR1(x) + βR2(x)

[0028] Where x is a specific passenger, R1(x) is the area of the passenger detection frame, R2(x) is the confidence level that the passenger is facing directly forward, α and β are the weight factors for the corresponding factors respectively, and F(x) is the final score. Based on this ranking, the passenger priority is confirmed. The passenger with a higher priority is the main target passenger, and the advertisement catalog is recommended based on the passenger attributes.

[0029] As a preferred way of the interactive elevator advertisement intelligent recommendation method, the method for recommending advertisements to the target passenger includes:

[0030] Generating an advertisement catalog for the target passenger based on the scenario-aware FP-growth recommendation algorithm

[0031] When the advertisement catalog for the target passenger x ends and the camera still detects the target passenger x, if there are no other target passengers, the content in the advertisement catalog is played in rounds. If there are other target passengers x+i, then the advertisement catalog corresponding to x+i is played.

[0032] When the advertisement catalog for the target passenger x has not finished playing and the target passenger x exits the elevator, if there are no other target passengers, the advertisement screen resumes the default playback. If there are other target passengers x+i, then the advertisement catalogs corresponding to x+i are played in sequence according to the priority.

[0033] The present invention also provides an interactive elevator advertisement intelligent recommendation system, including:

[0034] A data central control center, responsible for the calculation, storage, and transmission of the data information of the entire elevator cluster

[0035] An information collection module, used to collect the passenger attribute information and interaction information in the elevator

[0036] A passenger attribute analysis module, obtaining the passenger attribute information based on a deep learning algorithm

[0037] A database module, storing the scenario information of the interaction between the passenger and the touch advertisement screen in the form of data

[0038] A global recommendation module, analyzing the database information based on the scenario-aware FP-growth recommendation algorithm, and making an intelligent recommendation of advertisements to the passengers based on the data of the entire elevator cluster

[0039] A local correction module, making a comprehensive judgment based on further analysis of the specific point information based on K-means on the basis of the global recommendation

[0040] The beneficial effects of the present invention are as follows: on the premise of having a touch-type advertising display screen, the present invention combines passenger attributes and interaction information, further explores on the basis of existing research, and designs an interactive elevator advertising intelligent recommendation method and system with certain robustness and high accuracy, which is beneficial for users to obtain better advertising services and achieve the maximization of advertising placement benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flowchart of the interactive elevator advertising intelligent recommendation method disclosed by the present invention.

[0042] Figure 2 It is a structural block diagram of the interactive elevator advertising intelligent recommendation system disclosed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0044] Embodiment 1:

[0045] See Figure 1 , this embodiment discloses an interactive elevator advertising intelligent recommendation method, including the following steps:

[0046] S1: Collect passenger information through a monitoring camera in the elevator, and collect passenger interaction scenario information through a touch advertising screen.

[0047] Among them, the elevator environment is as follows: the elevator touch advertising screen is installed on the left or right side of the elevator and on the same side as the elevator door, avoiding the button side, facing the passengers getting out of the elevator; the monitoring camera is located at the top of the car and directly above the advertising screen, which can monitor the entire elevator area and is also conducive to accurately analyzing the passengers in front of the advertising screen.

[0048] The touch advertising screen is as follows: there is a main playback window and an advertisement catalog window. In the main playback window, advertisements can be played daily in a scrolling manner. In the advertisement catalog window, the scroll bar can be slid, and any advertisement of interest can be clicked. If the currently played advertisement is the last one in the advertisement catalog, the advertising screen will perform a carousel.

[0049] The information collection is divided into two parts. One part is to collect passenger information through a monitoring camera, and the other part is to collect passenger interaction scenario information through a touch advertising screen. The passenger information is fed back to S2, and the interaction scenario information is fed back to S3.

[0050] S2: Analyze the passenger information based on the passenger attribute recognition algorithm.

[0051] Based on the passenger attribute recognition algorithm, the passenger information is analyzed. In this embodiment, only the gender information and age information of the passengers are fed back and analyzed. In addition, other attribute information such as the height, clothing, and whether the passenger wears accessories can also be considered.

[0052] Passenger attribute recognition has received much attention in the field of video surveillance. To improve the accuracy of passenger attribute recognition, deep learning methods are adopted. In this embodiment, based on the Attribute Localization Module (ALM), the gender and age of passengers are recognized. This model does not require additional regional annotations (implemented by STN (Spatial Transformer Networks)), and can be trained end-to-end under multi-scale deep supervision, significantly outperforming most existing methods on three pedestrian attribute datasets, namely PETA, RAP, and PA-100K. Among them, in the RAP dataset, the accuracy of Female (gender: male, female) reaches 96%, and the accuracy of AgeLess16 reaches 88%, both of which are better than the baseline model of the RAP dataset. The RAP dataset has a total of 41,585 pedestrian samples, including 72 attributes, collected from 26 cameras for indoor surveillance, with a resolution range from 36×92 to 344×554, having a certain similarity to the elevator scenario.

[0053] S3: Store the interaction scenario information in the database module, and construct an advertisement information set and an interaction scenario database according to the interaction scenario information.

[0054] The passenger interacts with the advertisement screen, selects the interested advertisement through the catalog window, and clicks to jump to obtain the detailed advertisement information. At this time, the recommendation system stores the passenger attributes and interaction scenario information in the database module. The interaction scenario information refers to the sequence of advertisements that the target passenger is interested in within a period of time. The specific process is as follows.

[0055] Construct the advertisement information set: S = {s1, s2,..., s m}, where s i = {k1, k2, k i ,....K m}, 1 ≤ i ≤ m, s i is the sequence of advertisements k i that the target passenger is interested in during the t1 time period. The start time of the t1 time period is when the passenger clicks on the advertisement interface, and the end time is when the passenger leaves the elevator; add time thresholds t2 and t3; the continuous loss of passenger information by the camera within the t2 time period indicates that the passenger has left the elevator, and the advertisement screen automatically exits the interaction interface until the next click is received and then starts constructing the advertisement information set again; if the duration from the passenger clicks on the advertisement k i to switching the interface is greater than t3, then k iAdd s i , otherwise do not add.

[0056] Construct an interactive scenario database: D = {d1, d2,..., d n}, where d i = {[x, y]: s1, s2,...}, 1 ≤ i ≤ n, d i represents specific interactive scenario data, that is, the thing database composed of target passengers and advertising information sets. Among them, [x, y] represents the attributes of the target passenger, x is the gender attribute, and y is the age attribute. In addition, other attributes such as height, clothing, and whether wearing ornaments can also be added.

[0057] S4: Based on the interactive scenario database, use the FP-growth recommendation algorithm based on scenario awareness to intelligently recommend advertisements to passengers.

[0058] Among them, FP-growth is an algorithm for mining data association rules, which effectively compresses the database into a data structure with a small storage space, overcomes the defect of multiple scans of the database in the classical algorithm Aprior, only needs to scan the transaction database twice, transforms the problem of finding long frequent patterns into a recursive pattern growth strategy, avoids generating a large number of candidate sets, and greatly reduces the time complexity of the algorithm.

[0059] The specific process is as follows:

[0060] Build an item header table, that is, the number of occurrences of each k in the advertising information set S i , define the minimum support as T MS , T MS ∈(0, 1); Scan the advertising information set S to obtain the counts ([C i , C C k2 , C ki .....) of all frequent 1-itemsets (k1, k2, k i ....); Then delete the items with support lower than the threshold T MS

[0061]

[0062] Mine the FP-tree, and find the conditional pattern bases corresponding to the item header table items from the bottom item of the item header table upwards in sequence; recursively mine the frequent item sets of the item header table items from the conditional pattern bases; define the number of items N of the frequent item sets FP , and only return the frequent item sets that meet the item number requirements;

[0063] Define the global weighting parameter, and perform batch normalization on the occurrence times of the specific items in the maximum frequent item set as the global weighting parameter corresponding to each specific item.

[0064] S5: Obtain the associated advertisements that the target passengers in the entire elevator cluster are interested in, and perform local correction on the results.

[0065] Obtain the associated advertisements that the target passengers in the entire elevator cluster are interested in from the global recommendations in S4, but the relative independence of each specific location should also be considered during the recommendation. Cluster the interactive data scenario information of the specific locations based on the K-means algorithm, where the size of the cluster is equal to the passenger attribute category, and perform batch normalization on the number of times the advertisements appear in each cluster. Combine the global recommendations in S4, weight the same type of products in the clustering results, and reorder them. Define the advertisement attention threshold T P , and retain the advertisement items with scores higher than T P to generate a new advertisement catalog.

[0066] S6: Locate the target passengers and recommend advertisements to the target passengers.

[0067] Among them, the method for locating the target passengers includes: after the passengers enter the elevator, detect the passenger positions based on the deep learning algorithm, detect the eye positions based on the opencv human eye detection model (haarcascade_eye.xml), perform template matching according to the eye state library, and detect the eye attention positions of the passengers. The eye state library includes 5 types of attention, namely straight ahead, left, right, up, and down. Since the camera is located above the advertisement screen, when the passengers face the advertisement screen, the eye state is considered to be straight ahead, and a target passenger model is established:

[0068] F(x) = αR1(x) + βR2(x)

[0069] Among them, x is the specific passenger, R1(x) is the area of the passenger detection frame, R2(x) is the confidence level of the passenger facing straight ahead, α and β are the weight factors of the corresponding factors respectively, F(x) is the final score, and the passengers are sorted based on this to confirm the passenger priorities. The passengers with higher priorities are the main target passengers, and advertisements are recommended based on the passenger attributes.

[0070] The method for recommending advertisements to the target passengers includes:

[0071] The FP-growth recommendation algorithm based on scenario awareness generates an advertisement catalog for the target passenger. When the advertisement catalog playback for the target passenger x ends and the camera still detects the target passenger x, if there are no other target passengers, the content in the advertisement catalog is played in rounds. If there are other target passengers x+i, the advertisement catalog corresponding to x+i is played. If the advertisement catalog for the target passenger x has not finished playing and the target passenger x exits the elevator, if there are no other target passengers, the advertisement screen resumes the default playback. If there are other target passengers x+i, the advertisement catalogs corresponding to x+i are played in order of priority.

[0072] Embodiment 2

[0073] See Figure 2 , this embodiment discloses an interactive elevator advertisement intelligent recommendation system, including:

[0074] The data central control center is responsible for the calculation, storage, and transmission of data information for the entire elevator cluster;

[0075] The information collection module is used to collect the passenger attribute information and interaction information in the elevator;

[0076] The passenger attribute analysis module obtains the passenger attribute information based on the deep learning algorithm;

[0077] The database module stores the scenario information of the interaction between the passenger and the touch advertisement screen in the form of data;

[0078] The global recommendation module analyzes the database information based on the FP-growth recommendation algorithm based on scenario awareness, and based on the data of the entire elevator cluster, makes an intelligent advertisement recommendation for the passengers;

[0079] The local correction module, on the basis of the global recommendation, further analyzes the specific location information based on K-means to make a comprehensive judgment, making the advertisement recommendation more accurate and efficient.

[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An interactive elevator advertisement intelligent recommendation method, characterized in that, It includes the following steps: Collect passenger information through a monitoring camera in the elevator, and collect passenger interaction scenario information through a touch advertising screen; Analyze the passenger information based on a passenger attribute recognition algorithm; Store the interaction scenario information in a database module, and construct an advertising information set and an interaction scenario database according to the interaction scenario information; The method for constructing the advertising information set and the interaction scenario database according to the interaction scenario information includes: Construct an advertisement information set: , where , , is the advertisement concerned by the target passengers in a sequence, The start time of the time period is when the passenger clicks on the advertisement interface, and the end time is when the passenger leaves the elevator; Construct an interactive scenario database: , where , , d l represents specific interactive scenario information, that is, a thing database composed of target passengers and advertisement information sets, where represents the target passenger attribute, is the gender attribute, is the age attribute; Based on the scenario-aware FP-growth recommendation algorithm on the basis of the interaction scenario database, intelligently recommend advertisements to passengers; Obtain the associated advertisements that target passengers in the entire elevator cluster are interested in, and perform local correction on the results; Locate the target passengers and recommend advertisements to the target passengers; Increase the time threshold during the process of constructing the advertisement information set and ; If the camera continuously loses passenger information within a time period, it indicates that the passenger has left the elevator, and the advertisement screen automatically exits the interactive interface until the next click is received, then starts constructing the advertisement information set again; if the duration from the passenger clicks on the advertisement to switching the interface is greater than , then is added to , otherwise it is not added; The method for intelligently recommending advertisements to passengers based on the scenario-aware FP-growth recommendation algorithm on the basis of the interaction scenario database includes: Create a header table, i.e., an advertisement information set Each The number of occurrences, the minimum support is defined as , ; Scan the advertising information set , get all frequent 1-item sets Count ; Then delete the support values ​​below the threshold , put the frequent 1-item set into the item header table and arrange them in descending order of support; Construct an FP-tree, scan the advertising information set, remove the non-frequent 1-item sets from the read original data, and sort them in descending order of support; read the sorted data set, and insert it into the FP-tree in the sorted order. The nodes sorted earlier are the root nodes, and the later ones are the child nodes; if there are shared root nodes, the corresponding count is incremented by 1; if new nodes appear, the corresponding nodes in the item header table will be linked to the new nodes through a node linked list; until all the data is inserted into the FP-tree; Mine the FP-tree, find the conditional pattern bases corresponding to the item header table items from the bottom item of the item header table upwards in sequence; recursively mine the frequent item sets of the item header table items from the conditional pattern bases; define the number of items in the frequent item sets , and only return the frequent item sets that meet the requirement of the number of items; Define a global weighting parameter, perform batch normalization on the occurrence times of the specific items in the maximum frequent item set, and use it as the global weighting parameter corresponding to each specific item; The method for performing local correction on the results includes: Cluster the interaction scenario information at specific locations based on the K-means algorithm. The size of the cluster is equal to the passenger attribute category, and batch normalization is performed on the number of times advertisements appear in each cluster; Weight the same-category advertisements in the clustering results and reorder them; define an advertisement attention threshold , and retain the advertisement items with scores higher than to generate a new advertisement catalog.

2. The interactive elevator advertisement intelligent recommendation method according to claim 1, characterized in that, The method for locating the target passengers includes: After the passenger enters the elevator, detect the passenger's position based on a deep learning algorithm, detect the position of the human eyes based on the opencv human eye detection model haarcascade_eye.xml, perform template matching according to the human eye state library, and detect the position where the passenger's eyes are focused. The human eye state library includes 5 types of focusing forward, left, right, up, and down. Since the camera is located above the advertising screen, when the passenger faces the advertising screen, the human eye state is considered to be forward, and a target passenger model is established: Among them, is a specific passenger, is the area of the passenger detection box, is the confidence that the passenger is facing directly forward, are the weight factors of the corresponding factors respectively, is the final score. Based on this ranking, the passenger priority is confirmed. The passenger with a higher priority is the main target passenger, and the advertisement catalog is recommended based on the passenger attributes.

3. The interactive elevator advertisement intelligent recommendation method according to claim 1, characterized in that, The method for recommending advertisements to the target passengers includes: Generate a target passenger advertisement catalog based on the scenario-aware FP-growth recommendation algorithm; Target passenger The playback of the advertisement catalog for the target passenger ends, but the camera still detects the target passenger , if there are no other target passengers, the content in the advertisement catalog is played in rounds. If there are other target passengers , then play the corresponding advertisement catalog; Target passenger The advertisement catalog has not finished playing for the target passenger gets out of the elevator. If there are no other target passengers, the advertisement screen resumes default playback. If there are other target passengers , the corresponding advertisement catalogs will be played in order according to the priority .

4. An interactive elevator advertisement intelligent recommendation system, the system is used to implement the interactive elevator advertisement intelligent recommendation method described in any one of claims 1-3, characterized in that, It includes: A data central control center, responsible for the calculation, storage, and transmission of data information of the entire elevator cluster; An information collection module, used to collect passenger attribute information and interaction information in the elevator; A passenger attribute analysis module, which obtains passenger attribute information based on a deep learning algorithm; A database module, which stores the scenario information of the interaction between passengers and the touch advertising screen in the form of data; A global recommendation module, which analyzes the database information based on the scenario-aware FP-growth recommendation algorithm, and intelligently recommends advertisements to passengers based on the data of the entire elevator cluster; The local correction module makes a comprehensive judgment based on further analysis of specific point information using K-means on the basis of global recommendation.

Images