Advertisement accurate pushing method and system based on take-out platform user data

Abstract

The application discloses an advertisement accurate pushing method and system based on take-out platform user data, relates to the technical field of data processing, and comprises the following steps: constructing a scoring model by a scoring model construction method, combining obtained take-out platform big data, user historical order data and associated influence factor data, generating a popular scoring model, a comprehensive scoring model and a personalized scoring model based on a set determination method, and outputting a pushing score of a current user through the scoring models to accurately determine whether the current user needs to push a commodity advertisement. The advertisement pushing method of the application scheme takes into account the advertisement pushing needs of new and old users, does not need to perform targeted data analysis and image on all users during pushing, greatly improves the pushing efficiency and also guarantees basic pushing accuracy, takes into account the personalized needs of old users during the pushing process, maximally guarantees the accuracy of advertisement pushing, and improves the advertisement click rate and order completion rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an advertisement accurate pushing method and system based on user data of a takeout platform. BACKGROUND

[0002] More and more users are now accustomed to purchasing food, fruits, desserts, beverages and other commodities through online takeout platforms. Various merchants also try every means to push commodity advertisements to users through the platform, so as to increase commodity sales. Obviously, how to determine the commodities that users are interested in and push commodity advertisements to users at the right time is the key to increasing commodity sales.

[0003] In the prior art, in order to accurately learn the needs and preferences of each user, the platform will profile the user according to user information data, including but not limited to understanding the user's takeout ordering habits, taste preferences, single price and other information, and then develop relevant advertisement pushing strategies according to the above information. In specific practice, the inventors have found that if the information data of each user is analyzed to generate an advertisement pushing strategy and achieve the effect of "one strategy for one person", a huge amount of data processing resources will be consumed, and high-frequency and efficient advertisement pushing on a large scale cannot be achieved in a short time. However, if the takeout platform big data is directly used to generate a generalized user profile, although the pushing efficiency can be improved, the advertisement click-through rate and order completion rate ultimately pushed to the user end will be significantly reduced. In summary, how to balance the accuracy, timeliness and economy of advertisement pushing is a problem that needs to be solved in the field of App advertisement pushing. SUMMARY

[0004] In order to solve the problem that it is difficult to balance accuracy and pushing efficiency when pushing App advertisements, the present application aims to provide an advertisement accurate pushing method based on user data of a takeout platform, which first uses takeout platform big data to generate a generalized pushing scoring model, realizes preliminary screening of target users based on the judgment result of the scoring model, and then generates an individualized scoring model based on the generalized scoring model combined with user historical order data, realizes accurate screening of target users and accurate pushing of commodity advertisements, and finally improves the click-through rate of advertisements and the order completion rate of commodities. In order to realize the advertisement accurate pushing method based on user data of a takeout platform, the present application also provides an advertisement accurate pushing system based on user data of a takeout platform, and the specific scheme is as follows:

[0005] An advertisement accurate pushing method based on user data of a takeout platform, comprising:

[0006] Setting a scoring model construction method, the scoring model is used to reflect the correlation between commodity sales and specific influence factor data;

[0007] Obtaining the to-be-pushed commodity information, constructing a popularization scoring model reflecting the corresponding relationship between the specific influence factor data score and the to-be-pushed commodity sales based on the takeout platform big data and according to the above scoring model construction method;

[0008] Obtaining the historical order data of the user and the historical influence factor data associated therewith;

[0009] Obtaining real-time influence factor data related to the current user and the to-be-pushed commodity;

[0010] Querying whether the historical order data contains order data of the same or similar category as the to-be-pushed commodity:

[0011] If not, generating a first reference score based on the real-time influence factor data and according to the popularization scoring model;

[0012] If yes, determining whether the number of orders exceeds a set value:

[0013] If yes, constructing a personalized scoring model based on the historical order data of the current user and the historical influence factor data associated therewith through the scoring model construction method, generating a second reference score according to the personalized scoring model combined with the real-time influence factor data;

[0014] If no, analyzing and confirming personalized influence factors associated with commodity sales based on the historical order data of the current user and the historical influence factor data associated therewith, revising the score value of the personalized influence factors in the popularization scoring model according to the correlation degree of each personalized influence factor and commodity sales, constructing a comprehensive scoring model, and generating a third reference score according to the comprehensive scoring model combined with the real-time influence factor data;

[0015] Comparing the first reference score, the second reference score, or the third reference score with a set push trigger threshold value, if the push trigger threshold value is exceeded, the current user is determined as a target user and the commodity advertisement data is pushed to the user end;

[0016] The influence factors include user gender, user age, order time, location, season, weather, commodity unit price, price discount, commodity delivery time, commodity type, commodity taste, and commodity color style.

[0017] By the technical scheme, firstly, the take-out platform big data is used to generate a popular scoring model reflecting the corresponding relationship between the commodity sales and the specific influence factor data, then it is detected whether the current user's historical order data contains the same or similar order data of the to-be-pushed commodity, if the current user has not purchased the current to-be-pushed commodity before, the scoring model is used to judge the scores of each influence factor related to the current user, and finally it is determined whether to push the commodity advertisement according to the scores; if the current user's historical order data contains the same or similar order data of the to-be-pushed commodity, it is determined whether a reliable scoring model can be generated according to the data amount of the order data, if the data amount is sufficient, a personalized scoring model is generated based on the relevant data of the current user, then it is determined whether to push the commodity advertisement according to the personalized scoring model and the real-time influence factor data related to the user, if the data amount is insufficient to generate a reliable scoring model, the influence factors having significant correlation with the commodity sales are analyzed and obtained based on the small amount of historical order data and the associated historical influence factor data, the scores of the influence factors in the popular scoring model are adjusted according to the correlation degree, thereby changing the overall weight of the influence factors, a comprehensive scoring model is generated, and finally it is determined whether to push the commodity advertisement according to the comprehensive scoring model and the real-time influence factor data related to the user. The advertisement pushing method takes into account the advertisement pushing needs of new and old users, and does not need to perform targeted data analysis and portrait on all users during pushing, thereby greatly improving the pushing efficiency and ensuring the basic pushing accuracy; meanwhile, the personalized needs of the old users are taken into account during pushing, thereby ensuring the accuracy of the advertisement pushing to the greatest extent and improving the advertisement click rate and the commodity order rate.

[0018] Further, the scoring model construction method comprises:

[0019] Converting the commodity sales and the influence factor data related thereto into numerical values;

[0020] Analyzing the influence factors influencing the commodity sales and the correlation coefficients between the commodity sales and the corresponding influence factors based on the source data;

[0021] Selecting the influence factors with the correlation coefficients greater than a set value as the correlation factors and storing the correlation factors in association with the commodity names;

[0022] Dividing the numerical values of the correlation factors into multiple intervals, obtaining and storing the corresponding reference scores of the numerical value intervals of the correlation factors in association with the names of the correlation factors according to the corresponding relationship between the numerical values of the correlation factors and the commodity sales;

[0023] Obtaining the correlation factors and the corresponding reference scores, configuring weights for the correlation factors and summing up the weights, and storing the sum as the scoring model:

[0024] The scoring output mode of the scoring model is: ;

[0025] The S c is the sum of the reference scores of the associated factors output by the scoring model, n is the number of associated factors, X i is the reference score of the i-th associated factor, and ω is the weight coefficient of the associated factor.

[0026] The source data includes to-be-pushed commodity information, take-out platform big data, current users, real-time influence factor data related to to-be-pushed commodities, user historical order data, and historical influence factor data associated therewith.

[0027] Through the above technical solution, the influence factors related to commodity sales can be accurately obtained from the source data, and the associated factors, i.e., specific influence factors, can be determined. Then, reference scores are assigned to each value segment of the associated factors according to the corresponding relationship between the value size of each associated factor and the high and low of commodity sales. The scoring model finally generated can output the score value size. Through comparison and judgment of the above score value size, it can be determined whether the advertisement needs to be pushed, thereby improving the click rate and order rate of the advertisement.

[0028] Further, the scoring output mode of the scoring model further includes a numerical characteristic curve.

[0029] Generating the numerical characteristic curve includes: obtaining and calculating the characteristic scores of each associated factor according to the reference scores and weight coefficients of each associated factor, and generating a numerical characteristic curve for reflecting the relative numerical relationship and numerical change trend among the associated factors based on the characteristic scores corresponding to each associated factor in a set arrangement order.

[0030] In the method for accurately pushing advertisements, the determination method for triggering the advertisement pushing action further includes:

[0031] Obtaining the corresponding relationship between each numerical characteristic curve shape and commodity sales, and setting a reference curve according to the above corresponding relationship.

[0032] Obtaining the numerical characteristic curve corresponding to the current user based on the scoring model, and comparing it with the reference curve. If the similarity exceeds a set value, the commodity advertisement data is pushed to the user end.

[0033] The Y-axis component of the numerical characteristic curve is represented as: M atrix =[ω1·X1,ω2·X2,···,ω n ·X n ];

[0034] The n is the number of associated factors, X nReference score for the nth correlation factor, ω n Weight coefficient for the nth correlation factor.

[0035] By the above technical solution, the willingness of the current user to click on the advertisement and purchase the product is determined by using the numerical relationship and numerical change trend between the correlation factors, which is simple and efficient, and can avoid the decisive influence of the numerical error of a single correlation factor on the overall determination result, thereby improving the reliability of the determination result.

[0036] Further, the score model construction method further includes:

[0037] Obtaining each correlation factor and its data associated with the product name;

[0038] Based on deep neural network analysis, the data characteristics of each correlation factor are extracted, and then the clustering algorithm is used to analyze and obtain the correlation factors with significant correlation changes in each correlation factor, which are stored as a correlation feature group;

[0039] The correlation feature group is taken as a whole and the sum of its numerical values is calculated, the sum of the numerical values of the correlation feature group is divided into multiple intervals, the corresponding relationship between the above numerical value interval of the correlation feature group and the product sales is obtained, the corresponding reference score is configured for each numerical value interval of the correlation feature group, and the correlation feature group name is associated and stored.

[0040] By the above technical solution, the correlation factors with significant correlation changes can be valued as a whole, which can improve the efficiency of scoring and facilitate the improvement of the efficiency of advertisement pushing.

[0041] Further, the advertisement precise pushing method further includes:

[0042] Divide the time of advertisement pushing into multiple pushing periods, and count the click rate and order rate corresponding to at least three pushing periods in which advertisement pushing has been completed, and generate a first trend line and a second trend line respectively;

[0043] Adjust the pushing trigger threshold corresponding to the subsequent pushing period according to the first trend line;

[0044] Adjust the numerical value of the specific correlation factor according to the second trend line;

[0045] The click rate and the pushing trigger threshold are inversely related.

[0046] The specific correlation factor includes product unit price or price discount.

[0047] By the above technical solution, the range of users covered by the advertisement and the influencing factors such as product price can be flexibly adjusted according to the response actions of the user end to the product advertisement, which helps to improve the click rate of the advertisement and the order rate of the product, and improves the effect of the advertisement.

[0048] Further, the advertisement accurate pushing method further comprises:

[0049] establishing and storing the association between the order sharing discount and the concentration degree of the personnel;

[0050] Before pushing the commodity advertisement data to the user end, the method further comprises:

[0051] obtaining the current location of the target user, and detecting whether there are other target users within a set range near the location coordinates;

[0052] If there are other target users and the number exceeds a set value, order sharing discount information of the current commodity is pushed according to the association.

[0053] Through the above technical solution, the order sharing discount promotion activities can be carried out according to the concentration degree of the target user, which helps to improve the click rate of the advertisement and the sales volume of the commodity.

[0054] Further, after determining the current user as the target user and pushing the commodity advertisement data to the user end, the method further comprises: marking the target user who has completed the commodity advertisement pushing, and storing the pushing time of this commodity advertisement pushing, the name of the pushed commodity, and whether the user clicks the advertisement or purchases the commodity information in association with the user information;

[0055] Before obtaining the historical order data of the user, the method further comprises: detecting the user information and confirming whether there is an advertisement pushing record in the user information:

[0056] If there is a commodity advertisement pushing record, the time interval between the current time and the last pushing time is detected: if the time interval does not exceed a set value, the pushing determination action of pushing the commodity advertisement to the current user is terminated; if the time interval exceeds the set value, the historical order data of the user is continuously obtained and the subsequent pushing determination steps are executed.

[0057] Through the above technical solution, the repeated pushing of the commodity advertisement can be effectively avoided, the user experience is improved, and the waste of system data processing resources is reduced.

[0058] An advertisement accurate pushing system based on user data of a takeout platform, comprising:

[0059] A scoring model construction unit configured to construct a scoring model reflecting the association between the commodity sales volume and the specific influence factor data according to the input source data and output;

[0060] A mass scoring model construction unit configured to obtain the big data of the takeout platform and construct a mass scoring model reflecting the corresponding relationship between the score of the specific influence factor data and the sales volume of the to-be-pushed commodity according to the scoring model construction method.

[0061] a source data acquisition unit configured to acquire to-be-pushed commodity information, food delivery platform big data, real-time influence factor data related to a current user and a to-be-pushed commodity, historical order data of the user and historical influence factor data associated therewith;

[0062] a push determination unit configured to query whether the current user historical order data contains order data identical to or similar in category to the to-be-pushed commodity;

[0063] If not, a first reference score is generated based on the real-time influence factor data and according to the popular scoring model;

[0064] If yes, it is determined whether the number of orders exceeds a set value;

[0065] If yes, a personalized scoring model is constructed based on the current user historical order data and the historical influence factor data associated therewith by using the scoring model construction method, and a second reference score is generated according to the personalized scoring model in combination with the real-time influence factor data;

[0066] If not, a personalized influence factor associated with commodity sales is analyzed and confirmed based on the current user historical order data and the historical influence factor data associated therewith, a score value corresponding to the personalized influence factor in the popular scoring model is revised according to the degree of association between each personalized influence factor and commodity sales, a comprehensive scoring model is constructed, and a third reference score is generated according to the comprehensive scoring model in combination with the real-time influence factor data;

[0067] an advertisement push unit configured to compare the first reference score, the second reference score or the third reference score with a set push trigger threshold value, and if the push trigger threshold value is exceeded, the current user is determined as a target user and commodity advertisement data is pushed to the user end;

[0068] The influence factors include user gender, user age, order time, location, season, weather, commodity unit price, price discount, commodity delivery time, commodity type, commodity taste and commodity color style.

[0069] Further, the scoring model construction unit comprises:

[0070] a data feature extraction subunit configured to convert commodity sales and influence factor data associated therewith into numerical values;

[0071] The correlation factor obtaining subunit is configured to obtain each influence factor affecting the commodity sales and a correlation coefficient between the commodity sales and the corresponding influence factor based on the source data analysis, and select the influence factor with a correlation coefficient greater than a set value as the correlation factor and store it in association with the commodity name;

[0072] The reference score subunit is configured to divide the values of each correlation factor into multiple intervals, obtain and configure a corresponding reference score for each value interval of the correlation factor according to the corresponding relationship between the value size of each correlation factor and the high and low of the commodity sales, and store it in association with the name of each correlation factor;

[0073] The score model generating subunit is configured to obtain each correlation factor and the corresponding reference score, configure a weight for each correlation factor and sum it up, and store it as the score model;

[0074] The score model generating subunit is configured to obtain each correlation factor and the corresponding reference score, configure a weight for each correlation factor and sum it up, and store it as the score model;

[0075] The S c The sum of the reference scores of each correlation factor output by the score model, n is the number of correlation factors, X i The reference score of the i-th correlation factor, and ω is the weight coefficient of the correlation factor;

[0076] The source data includes real-time influence factor data related to the to-be-pushed commodity information, the take-out platform big data, the current user, and the to-be-pushed commodity, user historical order data, and historical influence factor data associated therewith.

[0077] Through the above technical solution, the corresponding score model can be generated based on different source data, which is convenient for determining whether the current user is suitable for pushing the advertisement.

[0078] Further, the advertisement accurate pushing system further includes a parameter optimization unit, which includes:

[0079] The push period dividing subunit is configured to divide the time of the advertisement pushing into multiple push periods;

[0080] The advertisement click rate statistical subunit is configured to count the click rate corresponding to at least three push periods in which the advertisement pushing has been completed, and generate a first trend line;

[0081] The commodity order completion rate statistical subunit is configured to count the order completion rate corresponding to at least three push periods in which the advertisement pushing has been completed, and generate a second trend line;

[0082] The trigger threshold adjusting subunit is configured to adjust the push trigger threshold corresponding to the subsequent push period according to the first trend line;

[0083] ​The correlation factor value optimization subunit is configured to adjust the value of a specific correlation factor according to the second trend line, wherein the specific correlation factor includes a commodity unit price or a price discount.

[0084] Through the above technical solution, the push range of the advertisement and the corresponding commodity preferential price and other parameters can be adjusted and optimized according to the feedback of the user end after the early advertisement push, and the advertisement click rate and the commodity order rate are improved.

[0085] The present application at least includes the following one beneficial effect:

[0086] (1) The advertisement push method of the present application considers the advertisement push needs of new and old users, and does not need to perform targeted data analysis portrait on all users during push, greatly improving the push efficiency while ensuring the basic push accuracy;

[0087] (2) The present application considers the personalized needs of old users during push, maximizes the accuracy of advertisement push, and improves the advertisement click rate and the commodity order rate. BRIEF DESCRIPTION OF DRAWINGS

[0088] Figure 1 The present application is a schematic diagram of the overall process steps of the method;

[0089] Figure 2 The present application is a schematic diagram of the method steps of constructing the scoring model;

[0090] Figure 3 The present application is a schematic diagram of the functional module structure of the system. Reference signs: 100, scoring model construction unit; 200, popular scoring model construction unit; 300, source data acquisition unit; 400, push determination unit; 500, advertisement push unit; 110, data feature extraction subunit; 120, correlation factor acquisition subunit; 130, reference score subunit; 140, scoring model generation subunit. DETAILED DESCRIPTION

[0091] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0092] In the description of the present specification, the description of the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the described embodiments or examples are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0093] Before the description of the specific embodiments is described, it should be noted that in the embodiments of the present application, the to-be-pushed commodities include but are not limited to food and beverage, dessert and fruit, and are especially for commodities such as milk tea and bread. The take-out platform data includes data of mainstream take-out platforms such as Meituan and Ele.me, and also includes data of specific take-out channels of merchants such as WeChat applets and Douyin stores.

[0094] A method for accurately pushing advertisements based on user data of a take-out platform, as shown in Figure 1 , mainly includes the following steps:

[0095] S100, set a scoring model construction method, the scoring model is used to reflect the correlation between the sales of commodities and specific influence factor data;

[0096] S200, obtain the to-be-pushed commodity information, based on the big data of the take-out platform and according to the scoring model construction method, a popular scoring model is constructed to reflect the corresponding relationship between the scores of specific influence factor data and the sales of to-be-pushed commodities;

[0097] S300, obtain the historical order data of the user and the historical influence factor data associated therewith;

[0098] S400, obtain real-time influence factor data related to the current user and the to-be-pushed commodity;

[0099] S500, query whether the historical order data contains order data of the same or similar category as the to-be-pushed commodity:

[0100] S510, if not, generate a first reference score based on the real-time influence factor data and according to the popular scoring model;

[0101] S520, if yes, determine whether the number of orders exceeds a set value:

[0102] S521, if yes, construct a personalized scoring model based on the current user's historical order data and the historical influence factor data associated therewith through the scoring model construction method, and generate a second reference score according to the personalized scoring model combined with the real-time influence factor data;

[0103] S522, if not exceeded, based on the current user historical order data and the historical influence factor data associated therewith, analyzing and confirming the personalized influence factors associated with the commodity sales, revising the score values of the above-mentioned personalized influence factors in the popularization scoring model according to the correlation degree of each personalized influence factor and the commodity sales, constructing and generating a comprehensive scoring model, and generating a third reference score according to the comprehensive scoring model combined with the real-time influence factor data;

[0104] S600, comparing the first reference score, the second reference score or the third reference score with a set push trigger threshold value:

[0105] S610, if the push trigger threshold value is exceeded, the current user is determined as a target user and the commodity advertisement data is pushed to the user end.

[0106] In the embodiments of the present application, the above-mentioned influence factors include but are not limited to user gender, user age, order time, location, season, weather, commodity unit price, price discount, commodity delivery time, and commodity characteristics (such as type, taste, color style and other information data).

[0107] As shown in the detailed description, Figure 2 The scoring model construction method specifically includes:

[0108] S110, converting the commodity sales and the influence factor data related thereto into numerical quantities. The conversion of the above-mentioned numerical quantities can be processed based on the set rules, for example, the season is numerically coded: spring = 1, summer = 2, autumn = 3, winter = 4; for example, the weather type is coded: sunny = 1, cloudy = 2, rainy = 3, snowy = 4, foggy = 5, etc., and for the commodity category, hot encoding can be used, such as beverage = [0, 0, 1], etc., and for the commodity attributes, such as sweetness, it can be divided into 0-5 levels, after the above-mentioned conversion operation, each correlation factor is converted into a numerical quantity, which is convenient for data analysis and machine learning.

[0109] S120, based on the source data analysis, obtaining each influence factor affecting the commodity sales and the correlation coefficient between the commodity sales and the corresponding influence factor. In the embodiments of the present application, the above-mentioned source data mainly includes the big data of the takeout platform, the user historical order data and the historical influence factor data associated therewith. The above-mentioned data is usually stored in the server of each platform, such as the user historical order data which includes the order time, location, commodity unit price, commodity characteristics and delivery time, etc. At the same time, the user information of the takeout platform also records the user's gender, age and other information, and the weather information corresponding to the order time can be obtained through an external meteorological data platform.

[0110] In step S120, the correlation coefficient between the data is obtained by a correlation coefficient analysis method, and the correlation coefficient is used to measure the close degree of the relationship between two influence factors and between the influence factor and the commodity sales.

[0111] In step S130, the influence factor with a correlation coefficient greater than a set value is selected as a correlation factor and is stored in association with the commodity name. The main purpose of step S130 is to remove interference terms in the influence factor. For example, when the commodity to be pushed is fresh vegetables, the relationship between the user age and the commodity sales is not close, and the influence factor can be removed. When the commodity to be pushed is milk tea, the user age has a close relationship with the commodity sales, and the influence factor can be retained.

[0112] In step S140, the values of the correlation factors are divided into multiple intervals, the corresponding relationship between the value of each correlation factor and the high and low of the commodity sales is obtained, a corresponding reference score is configured for each value interval of the correlation factor, and the reference score is stored in association with the name of the correlation factor. The multiple intervals can be equally spaced and averaged, or can be divided according to the concentration of the values. For example, it is analyzed that when the temperature is higher than 30°C in summer, the sales of ice milk tea and other drinks will increase by 5%-8% for every 1°C increase in temperature. Therefore, the temperature above 30°C can be divided into intervals with a span of 1°C, and each value interval corresponds to a score of 0.5. For example, 33°C corresponds to a score of 4, and 35°C corresponds to a score of 5. The temperature from 15°C to 25°C is divided into intervals with a span of 5°C.

[0113] In step S150, the reference scores of the correlation factors and the corresponding reference scores are obtained, the weights of the correlation factors are configured, and the sum is stored as the score model.

[0114] In the embodiment of the application, the score output mode of the score model is:

[0115] ;

[0116] In step S140, the values of the correlation factors are divided into multiple intervals, the corresponding relationship between the value of each correlation factor and the high and low of the commodity sales is obtained, a corresponding reference score is configured for each value interval of the correlation factor, and the reference score is stored in association with the name of the correlation factor. The multiple intervals can be equally spaced and averaged, or can be divided according to the concentration of the values. For example, it is analyzed that when the temperature is higher than 30°C in summer, the sales of ice milk tea and other drinks will increase by 5%-8% for every 1°C increase in temperature. Therefore, the temperature above 30°C can be divided into intervals with a span of 1°C, and each value interval corresponds to a score of 0.5. For example, 33°C corresponds to a score of 4, and 35°C corresponds to a score of 5. The temperature from 15°C to 25°C is divided into intervals with a span of 5°C. c The sum of the reference scores of the correlation factors output by the score model, n is the number of correlation factors, X i is the reference score of the i-th correlation factor, and ω is the weight coefficient of the correlation factor.

[0117] Based on the above technical scheme, the influence factors related to the commodity sales can be accurately obtained from the source data, and the correlation factors, i.e., specific influence factors, can be determined. Then, reference scores are assigned to each value section of the correlation factors according to the corresponding relationship between the value size of each correlation factor and the high and low of the commodity sales. Finally, the generated scoring model can output the score size. Through comparison and judgment of the above score size, it can be determined whether the advertisement needs to be pushed, and the click rate and order rate of the advertisement are improved.

[0118] In specific practice, in order to avoid that the value error of a single correlation factor has a decisive influence on the overall determination result and improve the reliability of the determination result, the scoring output mode of the scoring model in the embodiments of the present application further includes a value characteristic curve. The data basis of the value characteristic curve is a one-dimensional matrix. The elements in the matrix are the products of the values of the correlation factors and their weight coefficients, i.e., characteristic scores.

[0119] Generating the above value characteristic curve includes: obtaining and calculating the characteristic scores of the correlation factors according to the reference scores of the correlation factors and the weight coefficients, and generating a value characteristic curve for reflecting the relative value relationship and value change trend among the correlation factors based on the characteristic scores corresponding to the correlation factors in a set arrangement order.

[0120] In the advertisement accurate pushing method described in the present application, the determination method of triggering the advertisement pushing action further includes:

[0121] S620, obtaining the corresponding relationship between the shape of each value characteristic curve and the commodity sales, and setting a reference curve according to the above corresponding relationship;

[0122] S630, obtaining the value characteristic curve corresponding to the current user based on the scoring model, and comparing it with the reference curve. If the similarity exceeds the set value, the commodity advertisement data is pushed to the user end.

[0123] The Y-axis component of the value characteristic curve is represented as: Matrix=[ω1·X1,ω2·X2,···,ω n ·X n ];

[0124] The above n is the number of correlation factors, X n is the reference score of the nth correlation factor, and ω n is the weight coefficient of the nth correlation factor. The Y-axis component of the value characteristic curve is stored as a one-dimensional matrix. Each element in the matrix, i.e., the characteristic score, is defined as the value of each point on the Y-axis in a two-dimensional plane coordinate system. Each characteristic score is set at equal intervals on the X-axis. Connecting the above characteristic scores constitutes the value characteristic curve.

[0125] The technical solution determines the willingness of the current user to click on the advertisement and purchase the product by using the numerical relationship and numerical change trend between the correlation factors, which is simple and efficient. For example, the Y-axis component of the reference curve for triggering the push of the ice milk tea advertisement is 8-4-6, where 8 is the product of time and its weight coefficient, 4 is the product of temperature and its weight coefficient, and 6 is the product of the location classification and its weight coefficient. If the Y-axis component of the numerical characteristic curve corresponding to the current user is 7.8-4.3-6.1, the push action of the product advertisement will also be triggered.

[0126] Based on the above optimization sub-solution, further, the scoring model construction method according to the embodiments of the application further comprises:

[0127] S131, obtaining each correlation factor and its data associated with the product name;

[0128] S132, extracting the data characteristics of each correlation factor based on deep neural network analysis, and then obtaining the correlation factors with significant correlation changes in each correlation factor through clustering algorithm analysis and storing them as a correlation feature group;

[0129] S133, taking the correlation feature group as a whole and calculating the sum of its values, dividing the sum of the values of the correlation feature group into multiple intervals, obtaining the corresponding relationship between the sum of the values of the correlation feature group and the high and low of the product sales, and configuring corresponding reference scores for each value interval of the correlation feature group and storing them in association with the name of the correlation feature group. For example, taking time, weather, and location as a correlation feature group, it can be taken as a whole. The above technical solution can assign values to correlation factors with significant correlation changes as a whole, which can improve the scoring efficiency and facilitate the efficiency of advertisement push.

[0130] The product information to be pushed in the step S200 includes product name, product sales, product characteristics, such as type, taste, color style, and other information data.

[0131] In order to flexibly adjust the range of users covered by the advertisement and the influence factors such as product price according to the response action of the user end to the product advertisement, the advertisement precise push method in the embodiments of the application further comprises a parameter optimization step, which specifically comprises:

[0132] S710, dividing the time of advertisement push into multiple push periods, counting the click rate and order rate corresponding to at least three push periods that have completed advertisement push, and generating a first trend line and a second trend line respectively. In actual application, the target user can be selected and the product advertisement can be pushed every 30 minutes.

[0133] S720, adjusting a push trigger threshold corresponding to a subsequent push period according to the first trend line, wherein the click rate is inversely related to the push trigger threshold, that is, the higher the user end advertisement click rate, the lower the push trigger threshold, until a set click rate value is reached, at which time the push trigger threshold no longer decreases, so that the pushed commodity advertisement can be clicked by more users.

[0134] S730, adjusting the value of a specific correlation factor according to the second trend line, wherein the specific correlation factor includes commodity unit price or price discount. The second trend line reflects the proportion of actual purchase after the user clicks the advertisement, and in general, when the user clicks into the commodity advertisement, for example, the recommended advertisement of a specific taste milk tea, the main influencing factors affecting whether the user ultimately purchases the commodity include commodity unit price and discount strength. Therefore, when the order rate is lower than the set value, the above influencing factors are adjusted, which can effectively improve the order rate after clicking the advertisement and is beneficial to improving the marketing effect of the advertisement.

[0135] In a specific embodiment, for sharing-friendly commodities such as dessert drinks, the advertisement accurate push method further comprises:

[0136] S810, establishing and storing the correlation between the sharing discount and the personnel concentration degree;

[0137] Before pushing the commodity advertisement data to the user end, it further comprises:

[0138] S820, obtaining the current location of the target user, and detecting whether there are other target users within a set range near the location coordinates;

[0139] S821, if there are other target users and the number exceeds a set value, then according to the above correlation relationship, the sharing discount information of the current commodity is pushed;

[0140] S822, if not, only push the commodity advertisement data.

[0141] The method for obtaining the current location of the target user is to use the user end, such as the built-in GPS positioning system in the smart phone, to locate the position coordinates of the target user. The above scheme can promote the sharing discount according to the concentration degree of the target user, which is helpful to improve the click rate of the advertisement and the sales volume of the commodity.

[0142] In the embodiment of the application, after the step S610 or S630 is executed and the push of the commodity advertisement is completed, the target user to which the commodity advertisement has been pushed is marked, and the push time of this commodity advertisement, the commodity name, and whether the user clicks the advertisement or purchases the commodity and other information are stored in association with the user information.

[0143] Before step S300, further comprising: detecting user information and confirming whether there is an advertisement pushing record in the user information:

[0144] If there is a commodity advertisement pushing record, detecting the time interval between the current time and the last pushing time, if the time interval does not exceed the set value, terminating the execution of the judgment action of pushing the commodity advertisement to the current user, that is, no longer executing the subsequent steps of obtaining user historical order data and the like; if the time interval exceeds the set value, continuing to execute step S300.

[0145] The above scheme can effectively avoid the repeated pushing of commodity advertisements, improve user experience, and reduce the waste of system data processing resources.

[0146] It should be pointed out that the above step numbers and orders are only preferred embodiments, and are not a limitation of the method. In specific applications, one or more steps can be selectively executed, or executed in a specific order.

[0147] To implement the above-mentioned advertisement accurate pushing method based on takeout platform user data, the application also discloses an advertisement accurate pushing system based on takeout platform user data, as shown in Figure 3 According to the function, it mainly includes: score model construction unit 100, popular score model construction unit 200, source data acquisition unit 300, pushing judgment unit 400 and advertisement pushing unit 500. The above-mentioned functional units are configured in the platform server, which is convenient for the calling, processing and storage of big data, and also convenient for the pushing of commodity advertisements.

[0148] The score model construction unit 100 is configured to construct a score model for reflecting the correlation between commodity sales and specific influence factor data according to the input source data and output. In detail, the score model construction unit 100 specifically includes: data feature extraction subunit 110, correlation factor acquisition subunit 120, reference score subunit 130, and score model generation subunit 140.

[0149] The data feature extraction subunit 110 is configured to convert the commodity sales and the influence factor data related thereto into numerical quantities, and is specifically configured as a conversion program module to convert different types of data into numerical data convenient for analysis and processing. The correlation factor acquisition subunit 120 is configured to acquire, based on the source data analysis, each influence factor affecting the commodity sales and the correlation coefficient between the commodity sales and the corresponding influence factor, and select the influence factor with a correlation coefficient greater than a set value as a correlation factor and store it in association with the commodity name. In specific applications, the correlation coefficient analysis program module is configured to input the large data in the form of numerical quantities into the above-mentioned correlation coefficient analysis program module for processing according to the data name, such as weather temperature. The reference score subunit 130 is configured to divide the numerical values of each correlation factor into multiple intervals, acquire and according to the corresponding relationship between the numerical value of each correlation factor and the high and low of the commodity sales, configure the corresponding reference score for each numerical interval of the correlation factor and store it in association with the name of each correlation factor. The score model generation subunit 140 is configured to acquire each correlation factor and the corresponding reference score thereof, configure the weight for each correlation factor and sum it up, and store it as the score model. The score output mode of the score model is: c The sum of the reference scores of each correlation factor output by the score model, n is the number of correlation factors, X i is the reference score of the i-th correlation factor, and ω is the weight coefficient of the correlation factor.

[0150] The popular score model construction unit 200 is configured to be data-connected with the score model construction unit 100 and the source data acquisition unit 300, and is used to acquire the take-out platform big data and construct the popular score model reflecting the corresponding relationship between the score of a specific influence factor data and the sales of a to-be-pushed commodity according to the above-mentioned score model construction method.

[0151] The source data acquisition unit 300 is configured to acquire the to-be-pushed commodity information, the take-out platform big data, the real-time influence factor data related to the current user and the to-be-pushed commodity, the historical order data of the user and the historical influence factor data associated therewith. In actual applications, the source data acquisition unit 300 is configured as a special data interface module in the server, and has the functions of data initial conversion and temporary storage.

[0152] ​The push determination unit 400 is configured to query whether the current user historical order data contains order data similar to the to-be-pushed commodity or the category of the to-be-pushed commodity. If the current user historical order data does not contain order data similar to the to-be-pushed commodity or the category of the to-be-pushed commodity, a first reference score is generated based on the real-time influence factor data and according to the popularization scoring model. If the current user historical order data contains order data similar to the to-be-pushed commodity or the category of the to-be-pushed commodity, it is determined whether the number of orders exceeds a set value. If yes, a personalized scoring model is constructed based on the current user historical order data and the historical influence factor data associated therewith by using the scoring model construction method, and a second reference score is generated according to the personalized scoring model in combination with the real-time influence factor data. If no, a personalized influence factor associated with the commodity sales is analyzed and confirmed based on the current user historical order data and the historical influence factor data associated therewith, the score value of the personalized influence factor in the popularization scoring model is revised according to the correlation degree of each personalized influence factor and the commodity sales, a comprehensive scoring model is constructed, and a third reference score is generated according to the comprehensive scoring model in combination with the real-time influence factor data.

[0153] The advertisement push unit 500 includes a numerical comparison subunit and a data sending subunit. The numerical comparison subunit is configured to compare the first reference score, the second reference score, or the third reference score with a set push trigger threshold value. If the push trigger threshold value is exceeded, the current user is determined as a target user by using the data sending subunit, and the commodity advertisement data is pushed to the user end. The data sending subunit includes an Internet communication module, which is used to send commodity advertisement data to the user end, such as a smart phone, a tablet computer, or the like, through a 4G / 5G network or other equivalent communication link.

[0154] In the embodiments of the present application, the related source data and the processing intermediate data are stored in a database connected to the server.

[0155] In the embodiments of the present application, the advertisement accurate push system further includes a parameter optimization unit, specifically including: a push time period division subunit, an advertisement click rate statistical subunit, a commodity order rate statistical subunit, a trigger threshold adjustment subunit, and an associated factor value optimization subunit.

[0156] The push period division sub-unit is configured to divide the time for pushing the advertisement into multiple push periods. The advertisement click rate statistics sub-unit is configured to count the click rate corresponding to at least three push periods for which the advertisement pushing has been completed, and generate a first trend line. The commodity order rate statistics sub-unit is configured to count the order rate corresponding to at least three push periods for which the advertisement pushing has been completed, and generate a second trend line. The trigger threshold adjustment sub-unit is configured to adjust the push trigger threshold corresponding to the subsequent push period according to the first trend line. The correlation factor value optimization sub-unit is configured to adjust the value of a specific correlation factor according to the second trend line, wherein the specific correlation factor includes the commodity unit price or the price discount. By configuring the above parameter optimization unit, the push range of the advertisement and the corresponding commodity preferential price and other parameters can be adjusted and optimized according to the feedback of the user end after the early advertisement pushing, so as to improve the advertisement click rate and the commodity order rate.

[0157] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those of ordinary skill in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for accurately pushing advertisements based on user data of a take-out platform, characterized in that, The method comprises the following steps: Setting a scoring model construction method, which reflects the correlation between commodity sales and specific influencing factor data; Obtaining the information of the commodity to be pushed, and constructing a popular scoring model based on the big data of the takeout platform and according to the scoring model construction method, which reflects the corresponding relationship between the score of the specific influencing factor data and the sales of the commodity to be pushed; Obtaining the historical order data of the user and the historical influencing factor data associated therewith; Obtaining real-time influencing factor data related to the current user and the commodity to be pushed; Querying whether the historical order data contains order data of the same or similar category as the commodity to be pushed: If not, generate a first reference score based on the real-time influencing factor data and according to the popular scoring model; If yes, determine whether the number of orders exceeds a certain value: If yes, construct a personalized scoring model based on the current user's historical order data and the historical influencing factor data associated therewith by using the scoring model construction method, generate a second reference score according to the personalized scoring model combined with the real-time influencing factor data; If no, analyze and confirm the personalized influencing factors that are associated with commodity sales based on the current user's historical order data and the historical influencing factor data associated therewith, revise the score values of the personalized influencing factors in the popular scoring model according to the correlation degree of each personalized influencing factor with commodity sales, construct a comprehensive scoring model, and generate a third reference score according to the comprehensive scoring model combined with the real-time influencing factor data; Compare the first reference score, the second reference score or the third reference score with a set push trigger threshold value, if it exceeds the push trigger threshold value, determine the current user as a target user and push the commodity advertisement data to the user end; The influencing factors include user gender, user age, order time, location, season, weather, commodity unit price, price discount, commodity delivery time, commodity type, commodity taste, and commodity color style. The scoring model construction method comprises: Converting commodity sales and influencing factor data related thereto into numerical values; Based on source data analysis, obtain each influencing factor that affects commodity sales and the correlation coefficient between commodity sales and its corresponding influencing factor; Select the influencing factors with correlation coefficients greater than a certain value as correlation factors and store them in association with the commodity name; Divide the numerical values of each correlation factor into multiple intervals, obtain and configure corresponding reference scores for each numerical interval of the correlation factor according to the corresponding relationship between the numerical size of each correlation factor and the high and low of commodity sales, and store them in association with the name of each correlation factor; Obtain each correlation factor and its corresponding reference score, configure the weight of each correlation factor and sum them up, and store them as the scoring model: The scoring model has the following scoring output methods: ； The above S c is the sum of the reference scores of each correlation factor output by the scoring model, n is the number of correlation factors, X i is the reference score of the i-th correlation factor, ω i is the weight coefficient of the correlation factor; The source data includes the information of the commodity to be pushed, the big data of the takeout platform, the real-time influencing factor data related to the current user and the commodity to be pushed, the historical order data of the user and the historical influencing factor data associated therewith.

2. The method of claim 1, wherein the method further comprises: The scoring output method of the scoring model also includes a numerical characteristic curve. The generating the numerical characteristic curve comprises: obtaining and calculating the characteristic scores of the correlation factors according to the reference scores and the weight coefficients of the correlation factors; and generating a numerical characteristic curve for reflecting the relative numerical relationship and the numerical change trend among the correlation factors based on the characteristic scores of the correlation factors in a set arrangement order. The method further comprises: Obtaining the corresponding relationship between the numerical characteristic curve shape and the commodity sales volume, and setting a reference curve according to the corresponding relationship; Obtaining the numerical characteristic curve corresponding to the current user based on the scoring model, and comparing the numerical characteristic curve with the reference curve, if the similarity exceeds a set value, pushing the commodity advertisement data to the user end; The Y-axis component of the numerical characteristic curve is represented as: M atrix =[ω1·X1,ω2·X2,···,ω n ·X n ] n is the number of correlation factors, X n is the reference score of the nth correlation factor, ω n is the weight coefficient of the nth correlation factor.

3. The method of claim 2, wherein the method further comprises: The method further comprises: Obtaining each correlation factor and its data associated with the commodity name; Extracting the data characteristics of each correlation factor based on deep neural network analysis, and obtaining the correlation factors with significant correlation changes in each correlation factor through clustering algorithm analysis and storing them as a correlation feature group; Taking the correlation feature group as a whole and calculating the sum of the numerical values, dividing the sum of the numerical values of the correlation feature group into multiple intervals, obtaining the corresponding relationship between the sum of the numerical values of the correlation feature group and the high and low of the commodity sales volume, and configuring corresponding reference scores for each numerical value interval of the correlation feature group and storing them in association with the correlation feature group name.

4. The method of claim 1, wherein the method further comprises: The method further comprises: Dividing the time of pushing the advertisement into multiple pushing periods, and statistically obtaining the click rate and the order rate corresponding to at least three pushing periods in which the advertisement has been pushed, and generating a first trend line and a second trend line respectively; Adjusting the pushing trigger threshold corresponding to the subsequent pushing period according to the first trend line; Adjusting the numerical value of the specific correlation factor according to the second trend line; The click rate and the pushing trigger threshold are inversely related; The specific correlation factor includes the unit price or the price discount of the commodity.

5. The method of claim 1, wherein the method further comprises: The method further comprises: Establishing and storing the association between the order sharing discount and the personnel concentration degree; Before pushing the commodity advertisement data to the user end, the method further comprises: Obtaining the current location of the target user, and detecting whether there are other target users within a set range near the location coordinates; If there are other target users and the number exceeds a set value, the order sharing discount information of the current commodity is pushed according to the association.

6. The method of claim 1, wherein the method further comprises: After determining the current user as the target user and pushing the commodity advertisement data to the user end, the method further comprises: marking the target user to whom the commodity advertisement has been pushed, and storing the pushing time of this commodity advertisement, the commodity name, and the information whether the user clicks the advertisement or purchases the commodity in association with the user information; Before obtaining the historical order data of the user, the method further comprises: detecting the user information and confirming whether there is an advertisement pushing record in the user information: If there is a commodity advertisement pushing record, detecting the time interval between the current time and the last pushing time: if the time interval does not exceed a set value, the pushing determination action of pushing the commodity advertisement to the current user is terminated; if the time interval exceeds a set value, the historical order data of the user is obtained and the subsequent pushing determination steps are executed.

7. An advertisement accurate pushing system based on take-out platform user data, characterized in that, The method comprises the following steps: a scoring model construction unit configured to construct a scoring model reflecting the correlation between commodity sales and specific influencing factor data according to input source data and output the scoring model; a popular scoring model construction unit configured to obtain takeout platform big data and construct a popular scoring model reflecting the corresponding relationship between specific influencing factor data scores and commodity sales to be pushed according to the scoring model construction method; a source data acquisition unit configured to acquire real-time influencing factor data related to the commodity information to be pushed, takeout platform big data, current users, and historical order data of the users and historical influencing factor data associated therewith; a push determination unit configured to query whether the current user historical order data contains order data identical to or similar to the commodity to be pushed; if not, a first reference score is generated based on the real-time influencing factor data and according to the popular scoring model; if yes, it is determined whether the number of orders exceeds a set value; if yes, a personalized scoring model is constructed by the scoring model construction method based on the current user historical order data and the historical influencing factor data associated therewith, and a second reference score is generated according to the personalized scoring model in combination with the real-time influencing factor data; if no, personalized influencing factors associated with commodity sales are analyzed and confirmed based on the current user historical order data and the historical influencing factor data associated therewith, the scores corresponding to the personalized influencing factors in the popular scoring model are revised according to the correlation degree of each personalized influencing factor with commodity sales, a comprehensive scoring model is constructed, and a third reference score is generated according to the comprehensive scoring model in combination with the real-time influencing factor data; an advertisement push unit configured to compare the first reference score, the second reference score, or the third reference score with a set push trigger threshold value, and if the push trigger threshold value is exceeded, the current user is determined as a target user and commodity advertisement data is pushed to the user end; wherein the influencing factors include user gender, user age, order time, location, season, weather, commodity unit price, price discount, commodity delivery time, commodity type, commodity taste, and commodity color style; the scoring model construction unit comprises: a data feature extraction subunit configured to convert commodity sales and influencing factor data related thereto into numerical values; a correlation factor acquisition subunit configured to acquire each influencing factor affecting commodity sales and the correlation coefficient between commodity sales and the corresponding influencing factor based on source data analysis, select the influencing factor with a correlation coefficient greater than a set value as a correlation factor, and store the correlation factor in association with the commodity name; a reference score subunit configured to divide the numerical values of each correlation factor into multiple intervals, acquire and determine the corresponding relationship between the numerical values of each correlation factor and the high and low of commodity sales, configure corresponding reference scores for each numerical value interval of the correlation factor, and store the reference scores in association with the names of each correlation factor. The scoring model generation subunit is configured to obtain each correlation factor and a corresponding reference score, configure a weight for each correlation factor and sum them up, and store the sum as the scoring model; The scoring output mode of the scoring model is: ； The above S c is the sum of the reference scores of each correlation factor output by the scoring model, n is the number of correlation factors, X i is the reference score of the i-th correlation factor, ω i is the weight coefficient of the correlation factor; The source data includes to-be-pushed commodity information, take-out platform big data, current users, real-time influence factor data related to to-be-pushed commodities, user historical order data, and historical influence factor data associated therewith. 8.The system according to claim 7, wherein, The advertisement precise pushing system further includes a parameter optimization unit, which includes: A pushing time period division subunit is configured to divide the time of advertisement pushing into multiple pushing time periods; An advertisement click rate statistical subunit is configured to count the click rates corresponding to at least three pushing time periods in which advertisement pushing has been completed, and generate a first trend line; A commodity order completion rate statistical subunit is configured to count the order completion rates corresponding to at least three pushing time periods in which advertisement pushing has been completed, and generate a second trend line; A triggering threshold adjustment subunit is configured to adjust the pushing triggering threshold corresponding to a subsequent pushing time period according to the first trend line; and a correlation factor value optimization subunit is configured to adjust the value of a specific correlation factor according to the second trend line, wherein the specific correlation factor includes a commodity unit price or a price discount.

Images