WO2019105226A1 - 一种用于营销活动效果预测的方法、装置及电子设备 - Google Patents

一种用于营销活动效果预测的方法、装置及电子设备 Download PDF

Info

Publication number
WO2019105226A1
WO2019105226A1 PCT/CN2018/115339 CN2018115339W WO2019105226A1 WO 2019105226 A1 WO2019105226 A1 WO 2019105226A1 CN 2018115339 W CN2018115339 W CN 2018115339W WO 2019105226 A1 WO2019105226 A1 WO 2019105226A1
Authority
WO
WIPO (PCT)
Prior art keywords
store
marketing
activity
similar
value
Prior art date
Application number
PCT/CN2018/115339
Other languages
English (en)
French (fr)
Inventor
樊翀
Original Assignee
口碑(上海)信息技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 口碑(上海)信息技术有限公司 filed Critical 口碑(上海)信息技术有限公司
Publication of WO2019105226A1 publication Critical patent/WO2019105226A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0207—Discounts or incentives, e.g. coupons or rebates

Definitions

  • the present application relates to the field of O2O catering industry marketing, and specifically relates to a method for predicting the effectiveness of marketing activities.
  • the present application also relates to an apparatus for predicting effectiveness of a marketing campaign, and an electronic device for implementing a method for predicting the effectiveness of a marketing campaign.
  • the estimation of the effectiveness of marketing activities is usually based on the similar marketing activity data of the history of the store, that is, the previous effect and the effect of the activity created now should be similar, and the discount rate is regarded as a dimension of the marketing activity. Calculate the effects of historically similar activities by simply counting the historical status of different discount rate activities.
  • the existing marketing activity forecasting scheme does not use the discount rate change as a key factor to quantitatively predict the effect of the activity, so there is a problem that the effect of the marketing campaign cannot be predicted more accurately in real time according to the change of the discount rate.
  • the discount rate has a great impact on the BI indicators (Business Indicators, or business indicators) of many activities.
  • a threshold has a threshold, and the threshold is generally higher than the price of the store. If the discount is large enough, Attractive, so many users will increase the price of the store in order to achieve the threshold of the use of coupons, and for example, for the repurchase rate, if the sent coupons are attractive enough to pull the user back, The repurchase rate naturally rises.
  • the present application provides a method for predicting the effectiveness of a marketing campaign to solve the problem that the existing marketing campaign effectiveness estimation scheme cannot accurately predict the effectiveness of the marketing campaign in real time according to the change of the discount rate.
  • the application additionally provides an apparatus for predicting the effectiveness of a marketing campaign.
  • the present application also provides an electronic device that implements a method for predicting the effectiveness of a marketing campaign.
  • the application provides a method for predicting the effectiveness of a marketing campaign, comprising:
  • a discount rate adjustment factor is constructed by linear fitting, and the initial value of the marketing activity effect is calibrated by using the discount rate adjustment factor, and the obtained calibration value is the target store marketing. Estimated activity performance.
  • the marketing activity performance data is a value increase of the business indicator before and after the marketing activity of the store, and at least includes any one of the following indicators: a percentage increase of the customer unit price and a percentage increase of the repurchase rate.
  • the campaign activity data of the selected target store includes the following processing:
  • Establishing a similar store model of the target store determining a similar store of the target store based on the similar store model, and selecting marketing campaign effect data of the target store and the similar store from the marketing campaign effect data of the different stores.
  • the establishing a similar store model of the target store includes:
  • the set store similarity ranking method includes:
  • the store similarity ranking is determined according to the feature values of the candidate similar stores, the proportion of each dimension in the similar store ranking, and the closeness of the feature values of the similar stores and the corresponding dimensions of the target store.
  • the preset condition includes:
  • the data smoothing process obtains an initial value of the marketing activity effect estimation of the target store, and includes the following smoothing process: taking the average value of the customer unit price increase percentage of the target store and the similar store as the target store The initial value of the customer unit price is estimated, and the average value of the resale rate increase percentage of the target store and its similar stores is taken as the initial value of the target store's repurchase rate.
  • the campaign activity effect data of the selected target store is an activity effect data of a historical marketing campaign that selects a release amount and a nuclear sales volume greater than a set value.
  • the discount activity adjustment factor is constructed by linear fitting according to the effect data of the marketing activities of different stores, including: a percentage increase of the customer unit price based on the historical marketing activity and a percentage increase of the repurchase rate, and the linearization by the least squares method
  • the formula for calculating the discount rate adjustment factor with the discount rate is obtained.
  • the linear fitting is performed by a least squares method, and a formula for calculating a discount rate adjustment factor with a discount rate is obtained, including:
  • the discount rate adjustment factor is calculated according to the discount rate: the discount rate adjustment factor is equal to the slope multiplied by the discount rate plus the intercept;
  • the intercept is equal to the average of the activity effect minus the slope multiplied by the average of the discount rate; the discount rate of each similar store is multiplied by the average difference of the activity effect, and the discount rate is divided by the discount rate of each similar store.
  • the sum of squares of the differences is obtained, that is, the slope is obtained; the deviation from the mean is the distance from which the actual value deviates from the average value.
  • the calibrating the initial value of the marketing activity effect using the discount rate adjustment factor includes using the following calibration formula:
  • the target store marketing activity effect estimated value is equal to the target store's marketing activity effect estimated initial value multiplied by the discount rate adjustment factor.
  • the method further includes: using the threshold adjustment factor for the target store marketing activity effect estimation value, wherein the threshold adjustment factor is the number of orders that meet the threshold according to the historical marketing activity and the number of orders that satisfy the customer unit price
  • the factor of the ratio calculation threshold to the effect of marketing activities is determined by:
  • the historical activity effect corresponding to the ratio is divided by the historical activity effect corresponding to the ratio of 1, and the obtained value is the threshold adjustment factor.
  • the estimated value of the target store marketing activity effect is further corrected by using a threshold adjustment factor, including: multiplying the threshold adjustment factor by the target store marketing activity effect estimation value to obtain a target store A further revised estimate of the effectiveness of marketing campaigns.
  • the marketing activity includes at least one of the following activity types: full reduction, each full reduction, and consumption delivery.
  • the calculating the performance data of the marketing activities of the different stores before and after the marketing campaign refers to calculating the customer unit price and the percentage of the customer unit price before and after the marketing activity according to the customer unit price and the repurchase rate in a certain period of time before and after the start of the marketing campaign. And the repurchase rate increased by a percentage point.
  • the application also provides a method for predicting the effect of a merchant activity, comprising:
  • the characteristic value of the store from any of the following dimensions: daily average number, pen unit price, store-owned business ID, location feature, and establishing an activity based on the KNN algorithm or based on the set store similarity ranking method according to the feature value Appropriate store model for the store;
  • the similar store of the activity applicable store is determined, and the effect data of the similar store marketing activity is obtained, and the data smoothing process is used to obtain the predicted value of the marketing activity effect of the activity applicable to the store;
  • the set store similarity ranking method includes:
  • the store similarity ranking is determined according to the feature value of each dimension of the candidate similar store, the proportion of each dimension in the similar store ranking, and the closeness of the feature value of the similar store and the corresponding store corresponding dimension of the store.
  • the preset condition includes:
  • the proportion of the dimensions in the similar store ranking is determined according to the following priority order:
  • the activity uses the store itself as the most similar store
  • the average number of daily pens is equal to the number of high daily averages divided by the number of low daily averages;
  • Pen unit price similarity the pen unit price similarity, equal to the high pen unit price divided by the low pen unit price
  • the distance from the applicable store is less than the set value
  • the application also provides an apparatus for predicting the effectiveness of a marketing campaign, comprising:
  • the historical data obtaining unit is configured to obtain historical marketing data of different stores as input samples, and calculate marketing activity effect data of different stores before and after the marketing activity;
  • the preliminary estimating unit is configured to select the marketing activity effect data of the target store, and obtain the initial estimated value of the marketing activity of the target store through data smoothing processing;
  • the estimated value calibration unit is configured to construct a discount rate adjustment factor by linear fitting based on the marketing activity effect data of different stores, and use the discount rate adjustment factor to calibrate the initial value of the marketing activity effect estimation, and obtain the calibration
  • the value is the estimated value of the target store marketing activity.
  • the application also provides an electronic device, including:
  • Memory Memory, and processor
  • the memory is for storing computer executable instructions for executing the computer executable instructions:
  • a discount rate adjustment factor is constructed by linear fitting, and the initial value of the marketing activity effect is calibrated by using the discount rate adjustment factor, and the obtained calibration value is the target store marketing. Estimated activity performance.
  • the method, the device and the electronic device for predicting the effect of the marketing activity obtain the historical marketing data of different stores as input samples, calculate the marketing activity effect data of different stores before and after the marketing activity; and select the marketing activity effect data of the target store.
  • the initial value of the marketing activity of the target store is obtained; based on the marketing activity effect data of different stores, the discount rate adjustment factor is constructed by linear fitting, and the discount rate adjustment factor is used to effect the marketing activity.
  • the estimated initial value is calibrated, and the obtained calibration value is the estimated value of the target store marketing activity effect, and the historical offline data of the target store is estimated according to the historical offline data of the store's marketing activity effect, and the offline value is used.
  • the data constructs a discount rate adjustment factor, and further adjusts the estimated value of the marketing activity, thereby solving the problem that the current marketing activity effect estimation scheme cannot accurately predict the effect of the marketing activity in real time according to the change of the discount rate. .
  • FIG. 1 is a process flowchart of a method for predicting effectiveness of a marketing activity provided by an embodiment of the present application
  • FIG. 2 is a schematic system diagram of an actual deployment example of a method for predicting effectiveness of a marketing activity provided by an embodiment of the present application;
  • FIG. 4 is a process flowchart of a method for predicting the effect of a merchant marketing activity provided by an embodiment of the present application
  • FIG. 5 is a schematic diagram of an apparatus for predicting the effectiveness of a marketing activity provided by an embodiment of the present application.
  • FIG. 6 is a schematic diagram of an electronic device for implementing the method for predicting the effectiveness of a marketing campaign provided by an embodiment of the present application.
  • the present application provides a method for predicting the effectiveness of marketing activities.
  • the present application also relates to an apparatus for predicting the effectiveness of marketing activities and an electronic device for implementing prediction of effects of marketing activities. Detailed description will be made one by one in the following embodiments.
  • An embodiment of the present application provides a method for predicting the effectiveness of a marketing campaign.
  • FIG. 1 is a process flow diagram of a method for predicting marketing activity effects provided by an embodiment of the present application
  • FIG. 2 is a system for practical application examples of a method for predicting marketing activity effects provided by an embodiment of the present application
  • FIG. 3 is an interaction diagram after actual deployment of an embodiment of the present application.
  • the method for forecasting the effectiveness of marketing activities shown in Figure 1 includes:
  • Step S101 obtaining historical marketing data of different stores as input samples, and calculating marketing activity effect data of different stores before and after the marketing activity;
  • Step S102 selecting marketing campaign effect data of the target store, and obtaining an initial estimated value of the marketing campaign effect of the target store through data smoothing processing;
  • Step S103 based on the marketing activity effect data of different stores, construct a discount rate adjustment factor by linear fitting, and use the discount rate adjustment factor to calibrate the initial value of the marketing activity effect estimation, and the obtained calibration value is the Target store marketing campaign performance estimates.
  • O2O catering industry business marketing planning and marketing programs continue to innovate, more accurate prediction of the effectiveness of marketing activities, so that businesses can clarify the value of marketing activities, plan to do marketing activities step by step, and the real-time effect of marketing activities Forecasting can help businesses more rationally configure marketing campaign resources, fees, and appropriate alternatives to set up activities that better match market expectations and marketing purposes.
  • the existing general forecasting method is based on the similar marketing campaign effect data of the store's history, that is, the previous effect and the activity effect created now should be similar. In fact, these alone do not provide accurate, real-time predictions of the effects of upcoming marketing campaigns. Among them, the discount rate has a great influence on the BI indicators (Business Indicators, or business indicators) of many activities.
  • a threshold with a threshold the threshold is generally higher than the price of the store, if the discount is large enough to attract People, then many users will spend more on the threshold of the use of coupons, thus increasing the unit price of the store, so the discount rate is one of the key factors affecting the effectiveness of forecasting marketing activities, and can not simply count the activities of different discount rates.
  • the embodiment of the present application takes the estimation of the effect of the marketing campaign activity of the O2O catering industry as an example, and describes the method for predicting the effect of the marketing activity in detail.
  • the model is explicitly constructed for the discount rate, and the effect of the activity is estimated more accurately in real time as the discount rate changes, instead of simply using the discount rate as one dimension of the activity to count only the different discount rate activities.
  • the historical situation in the calculation of the effect of historical similar activities, consider the effect of the historical marketing activities of similar stores around the store, and use the data of these similar stores to build a model to make predictions, and estimate the activity effect by the granularity of each store per activity.
  • the so-called similar store is a store that is selected for the target store and is close to the characteristics of the target store during the marketing campaign prediction process.
  • the so-called target store refers to the applicable activity store of the event selected by the merchant when launching the marketing campaign, that is, the online marketing campaign according to the store granularity.
  • the merchant of the word-of-mouth network can freely and flexibly choose which store to use for which promotion.
  • step S101 historical marketing data of different stores are acquired as input samples, and marketing activity effect data of different stores before and after the marketing activity are calculated.
  • the effect of the upcoming marketing campaign is predicted in advance, and the obtained reliable effect prediction result can be used as a pre-established comparison object, which is used to evaluate the marketing in the middle of the event or after the event is over.
  • the activity guides the operation of the store or the merchant by comparing the actual marketing situation with the comparison object.
  • the purpose of this step is to calculate the effect data of different store historical marketing activities based on the historical data of different stores.
  • the so-called marketing activity effect data is the value of the business indicators of the store before and after the marketing activities, including at least one of the following indicators: customer unit price increase percentage points and repurchase rate increase percentage points.
  • the calculation of the marketing activity effect data of different stores before and after the marketing campaign refers to calculating the customer unit price increase percentage and the repurchase rate of the store before and after the marketing activity according to the customer unit price and the repurchase rate within a certain period of time before and after the start of the marketing campaign. Increase the percentage points. For example, pull the raw data within one month before and after the marketing campaign to calculate.
  • Different stores can be stores of the same merchant or stores of different merchants.
  • the customer unit price refers to the average purchase price of each customer; the repurchase rate refers to the ratio of the returning customers of the secondary consumption to the total transaction UV under a certain time window, and the so-called UV is an independent visitor ( Unique Visitor), a computer client accessing the website is a visitor. The same client will only be counted once during 00:00-24:00. Obviously, the promotion can increase the amount of UV.
  • O2O catering industry business marketing activities have a wide range of names, including at least one of the following types of activities: full reduction, each full reduction, consumption delivery.
  • the full reduction is an activity type, which means that the corresponding discount can be enjoyed if the threshold is met, such as 100 yuan minus 10 yuan; the full reduction is another activity type different from the full reduction. It means that each time you meet the threshold, you can enjoy the corresponding discount multiple. For example, if you lose 100 yuan for every 100 yuan, you can get 30 yuan for consumers to spend 300 yuan.
  • the consumption is another type of activity, which means that after successful consumption, Automatically receive a coupon, the discount is generally greater than the corresponding full reduction of the store, and can be used when the user consumes twice.
  • the merchant can select which promotion activity is launched according to the store granularity, and can predict the effect of the target store of the launch event before the event, and comprehensively use the store data similar to the target store in the forecasting process, thereby balancing The impact of fluctuations in the operating conditions of the target store on the forecasting effect.
  • draw data from the past year to calculate the customer unit price and repurchase rate increase percentage per store per activity.
  • each customer For each store in the past year, each customer’s unit price and repurchase for one month before and after the statistical activity Rate, and then calculate the corresponding increase percentage point, which is the historical activity effect data of the customer unit price and the repurchase rate increase percentage of the store. If there is more than one activity of the same type in the history of a store, the customer unit price increase percentage and the repurchase rate increase percentage can be averaged for multiple activities.
  • the activity type can be subdivided, or the activities can be divided into categories according to different analysis purposes, for example, the activity is divided into two.
  • the major categories including the full reduction/per-full reduction activity and the consumer delivery activity, are targeted at the user's first transaction behavior and the return transaction behavior, that is, for the user's first transaction behavior, the full reduction/every reduction is classified into a large category. It can also be used as two separate activity types for more detailed analysis of data, such as analyzing the activity cost margin for each sub-category.
  • the underlying data is incomplete, for example, there is no activity in the history of similar stores in the vicinity, it is adjusted according to experience, abandoning the store without activity, or using the experience value of the global performance of the business.
  • step S102 the marketing activity effect data of the target store is selected, and the initial value of the marketing activity effect estimation of the target store is obtained through data smoothing processing.
  • the purpose of this step is to preliminarily estimate the initial value of the activity of the target store based on the historical activity data of the store.
  • the marketing activity effect data of the target store is extracted, and based on the historical data of the marketing activities, the marketing campaign effect launched by the target store is predicted in advance.
  • the historical data of the target store itself may be selected for prediction, or the historical data of the marketing activities of the similar stores may be selected at the same time as a data basis for predicting the effect of the marketing campaign of the target store.
  • the historical data of marketing activities of these similar stores it is possible to make more reliable predictions on the effects of the marketing activities launched by the target stores.
  • the so-called activity effect refers to the degree of improvement of the customer unit price and the repurchase rate
  • the estimation of the activity effect includes the percentage increase of the customer unit price of the predicted marketing activity and the percentage increase of the repurchase rate, and the marketing activity is predicted.
  • the campaign activity effect data of the selected target store includes the following processes: establishing a similar store model of the target store, determining similar stores of the target store based on the similar store model, and marketing campaign effect data from the different stores Select marketing activity effect data of the target store and its similar stores.
  • the establishing a similar store model of the target store includes: acquiring a store feature value from one or more dimensions, establishing a similar store of the target store based on the feature value based on a KNN algorithm or based on a set store similarity ranking method a model; wherein the one or more dimensions include at least one of the following dimensions: a daily average number, a pen unit price, a store-owned merchant ID, and a location feature.
  • the similar store model of the target store is established based on the set store similarity ranking method.
  • the set store similarity ranking method includes:
  • the store similarity ranking is determined according to the feature values of the candidate similar stores, the proportion of each dimension in the similar store ranking, and the closeness of the feature values of the similar stores and the corresponding dimensions of the target store.
  • the preset condition includes: belonging to the same city as the target store; and/or belonging to the same category as the target store. For example, a store that belongs to the same city as the target store and belongs to the same three-category category as a similar store candidate set, and then selects similar stores from the same dimension according to other dimensions.
  • the store feature value is obtained from multiple dimensions, including: the daily average number of pens, the unit price of the pen, the merchant ID of the store, and the location feature.
  • the so-called daily average is the average number of transactions per day, and a water bill is a transaction.
  • the unit price of the pen refers to the average transaction amount corresponding to each transaction record (a water list), and is generally measured by dividing the total amount of consumption by the number of consumption.
  • the store ID of the store that is, the store pid (or partner ID), for example, the KFC store pid is a yum group
  • the O2O catering business merchant will get the registered pid when the promotion website is registered, and the subordinate stores generally share the pid.
  • the same pid store is very similar.
  • the location feature is the distance from the geographic location of the target store and whether the location is in the same administrative area.
  • the close-to-door stores generally serve as a comparative dimension for selecting similar stores because of the similar customer group scope. For example, it belongs to the same administrative area as the target store.
  • the similar store model established for the target store includes sorting the similar stores, that is, sorting according to the similarity, so that when the similar store recalls the marketing campaign, the priority recall may be more preferred.
  • Similar stores because prioritizing more historical data of similar stores, can help to improve the prediction accuracy of the activity effect; wherein the recall marketing activity refers to obtaining activity data and other related information of the marketing activity.
  • the proportion of each dimension in the similar store ranking is determined according to the following priorities. proportion:
  • the target store itself is the most similar store
  • the average number of daily pens is equal to the number of high daily averages divided by the number of low daily averages;
  • Pen unit price similarity the pen unit price similarity, equal to the high pen unit price divided by the low pen unit price
  • the distance from the target store is less than the set value
  • the similar stores are defined as 8 sorting priorities of A to H, and the historical marketing activity data of the target store itself is used as the data of the highest priority in the forecasting process, and the priority is A;
  • the daily average number of similarities, the sorting priority is B, and the difference in the marketing effect between the two stores with large daily averages is also very large, so the feature value of the dimension is prior to the priority in the sorting, and the average number of days per day
  • the similarity is the relative value calculated by dividing the number of high daily averages by the number of low daily averages. For example, a store with a similar daily average number less than 1.5 is regarded as a similar store; the sort priority C is the pen price similarity, due to the effect of the activity.
  • the similarity of the unit price dimension is more important.
  • the value is taken from the high unit price divided by the low unit price.
  • the store with the similar unit price equal to 1.5 is a similar store;
  • the sort priority D is The same as the merchant ID of the target store;
  • the sorting priorities E to H are location features, including: the distance is less than 300 meters, the sorting priority E; the distance is less than 500 meters, the sorting priority F; the distance is less than 1000 , G is a sort priority; and the same area of the store, sort priority is H.
  • the KNN classification algorithm can also be used to output the stores corresponding to the target store for the multi-dimensional feature values of different stores
  • the so-called KNN algorithm ie K nearest neighbor (kNN, or k-NearestNeighbor) classification algorithm is one of the data mining classification techniques.
  • K nearest neighbor means that each sample can be represented by its nearest k neighbors.
  • the core idea is that if a sample is in a certain category, most of the k nearest neighbors in the feature space belong to a certain category. , the sample also belongs to this category and has the characteristics of the sample on this category.
  • the method determines the category to which the sample to be classified belongs according to only the category of the nearest one or several limited samples.
  • the campaign activity effect data of the selected target store and its similar store is the activity effect data of the historical marketing campaign that selects the release amount and the nuclear sales volume greater than the set value. That is, after selecting a similar store of the target store, the following screening process is performed on the marketing activity: selecting a marketing activity in which the issue amount and the nuclear sales volume are larger than the set value, for example, selecting activities of 10 similar stores. That is to say, in the process of selecting activities of similar stores, it is necessary to do a certain degree of screening of historical activities, because for historical activities, it is necessary to have sufficient statistical samples of the historical activities in order to have a relatively high degree of confidence. Sample to build the model.
  • the embodiment of the present application selects only those activities whose distribution amount is greater than 50 and the nuclear sales volume is greater than 10.
  • the write-off refers to the use of coupons issued by marketing activities in offline transactions.
  • the purpose of setting the threshold of coupon or card is to filter out the garbage activity.
  • Some marketing platforms do have a lot of invalid activities. Users of this kind of activity can't perceive it, so there is no need to use it as a reference for future prediction.
  • the amount of distribution If it is greater than 50, it can be considered that the various BI indicators of such activities, such as the write-off rate, the customer unit price increase, and the repurchase rate increase are statistically significant.
  • the threshold of the card core sales volume greater than 10 is a soft condition (so-called soft condition, not a mandatory condition), that is, the activity that satisfies this condition is preferentially selected, and the limit of the condition is relaxed when the number of activities satisfying the condition does not reach the set value, so that Try to pick enough activities.
  • the card core sales threshold is to try to select high-quality activities, because with the development of the platform, the quality of activities has become higher and higher, the garbage activities are less and less, and the reference of quality activities is naturally more important.
  • the activities of 10 similar stores are selected, and the activity effect data of the marketing activities of the stores are obtained, including the percentage increase of the customer unit price and the percentage of the repurchase rate of each of the stores before and after the marketing activity, and the effect data of the activity is adopted.
  • Data smoothing processing is performed to obtain an initial value of the marketing campaign effect of the target store.
  • the data smoothing process obtains an initial value of the marketing campaign effect of the target store, including the following smoothing process: taking the average value of the customer unit price increase percentage of the target store and the similar store as the customer unit price estimate of the target store
  • the initial value, the average value of the percentage increase percentage of the repurchase rate of the target store and its similar stores is taken as the initial value of the repurchase rate of the target store.
  • the increase in the unit price and the repurchase rate of the store takes the average of multiple activities.
  • the underlying data is incomplete, for example, there is no activity in the history of similar stores in the vicinity, then adjust according to experience, give up the store without activity, or use the experience value of the global performance of the business to carry out the bottom.
  • the average value is taken as the initial value of the effect of the marketing campaign of the type.
  • step S103 based on the marketing activity effect data of different stores, a discount rate adjustment factor is constructed by linear fitting, and the initial value of the marketing activity effect is calibrated using the discount rate adjustment factor, and the obtained calibration value is The target store marketing activity effect estimate.
  • the purpose of this step is to construct a discount rate adjustment factor by linear fitting according to different store historical activity data, and calibrate the initial value of the target store activity effect obtained in step S102.
  • the discount rate adjustment factor is constructed by linear fitting according to the effect data of the marketing activities of different stores, including: the percentage increase of the customer unit price based on the historical marketing activity and the percentage increase of the repurchase rate, and linear fitting by the least squares method.
  • the sample data used to construct the discount rate adjustment factor may be historical activity data of a similar store, or may be a statistical mean value of activity of all activities of the same type in the same city at a specific discount rate, using the sample data.
  • a linear fit is made to the discount rate to obtain the slope and intercept corresponding to the fitted line.
  • the customer unit price increase percentage and the repurchase rate increase percentage of the same type of activity in the same city are taken as samples, and only the activity of 50% or more is considered when selecting the sample, because the number of activities less than 5 fold is too small, the activity The effect is fluctuating significantly.
  • the linear fitting is performed by a least squares method, and a formula for calculating a discount rate adjustment factor with a discount rate is obtained, including: using a discount rate and a marketing activity effect, fitting a straight line intercept and a slope, and then discounting
  • the formula for calculating the rate adjustment factor as a function of discount rate is: the discount rate adjustment factor is equal to the slope multiplied by the discount rate plus the intercept;
  • the intercept is equal to the average of the activity effect minus the slope multiplied by the average of the discount rate; the discount rate of each similar store is multiplied by the average difference of the activity effect, and the discount rate is divided by the discount rate of each similar store.
  • the sum of squares of the differences is obtained, that is, the slope is obtained; the deviation from the mean is the distance from which the actual value deviates from the average value.
  • Discounted rate adjustment factor b * discount rate + a.
  • the initial value of the marketing campaign effect is calibrated using a discount rate adjustment factor, including the following calibration formula:
  • the target store marketing activity effect estimated value is equal to the target store's marketing activity effect estimated initial value multiplied by the discount rate adjustment factor, namely:
  • Forecast customer unit price increase percentage point customer unit price increase percentage estimate initial value * discount rate adjustment factor
  • Forecast repurchase rate increase percentage points repurchase rate increase percentage point estimate initial value * discount rate adjustment factor.
  • the method for predicting the effectiveness of the marketing activity is Including the estimated value of the target store marketing activity, using the threshold adjustment factor to further improve the prediction accuracy.
  • the threshold adjustment factor is a factor that calculates the influence of the threshold on the effect of the marketing activity according to the ratio of the number of orders that meet the threshold of the historical marketing activity and the number of orders that satisfy the customer unit price, and is determined by the following methods:
  • the historical activity effect corresponding to the ratio is divided by the historical activity effect corresponding to the ratio of 1, and the obtained value is the threshold adjustment factor.
  • the estimated value of the target store marketing activity effect is further corrected by using a threshold adjustment factor, including:
  • an initial value of the target store marketing activity is estimated, and after the discount rate adjustment factor and the threshold adjustment factor are corrected, the final activity effect prediction value is obtained, and the order of the two levels of adjustment is implemented. Does not affect the forecast results.
  • the user inputs relevant information of the marketing activity plan, for example, the promotion plan category, the maximum amount of the reduction amount, etc., and the processing of the embodiment of the present application can obtain the pre-promotion of the promotion activity.
  • the actual deployment of the system is shown in Figure 2.
  • the interaction between the user and the actual deployment system is shown in Figure 3.
  • the present application Based on an embodiment of the method for predicting the effectiveness of marketing activities provided by the present application, the present application also provides a method for predicting the effectiveness of a merchant activity.
  • FIG. 4 there is shown a process flow diagram of a method for merchant activity effect prediction provided in accordance with the present application.
  • the present embodiment is based on the foregoing embodiment of the method for predicting the effectiveness of the marketing activity, so the description is relatively simple.
  • the related part refer to the corresponding description of the above embodiment.
  • the embodiments of the method described below for merchant activity effects prediction are merely illustrative.
  • the present application provides a method for predicting the effectiveness of a merchant marketing campaign, including:
  • Step S401 obtaining the feature value of the store from any of the following dimensions: the daily average number of pens, the unit price of the pen, the merchant ID of the store, and the location feature, and based on the feature value, based on the KNN algorithm or the set ranking method based on the similarity of the store , establishing a similar store model for the event store;
  • Step S402 determining, according to the similar store model, a similar store in which the activity is applied to the store, obtaining similar store marketing activity effect data, and obtaining a predicted value of the marketing activity effect of the activity applicable to the store through data smoothing processing;
  • Step S403 the effect value of the marketing activity of the activity applicable store is aggregated, and the predicted value of the business activity effect is obtained.
  • the embodiment of the present application takes the O2O catering industry as an example to describe the method for predicting the effect of a merchant activity.
  • merchants are free to choose the applicable store for on-line promotion according to the granularity of the store.
  • the word-of-mouth network when estimating the effect of the event, first apply the effect of the store to the event, and then apply to all applicable stores of the online event.
  • the activity effect prediction values are aggregated to obtain the overall prediction result of the merchant.
  • the so-called activity-applied store refers to the applicable store that is selected for the event when the merchant launches the marketing campaign.
  • the historical marketing activity effect data of the similar stores around the store is taken into consideration together, instead of simply counting the applicable event store.
  • the historical situation can thus balance the fluctuations in store operations and make more accurate predictions of the effectiveness of marketing activities.
  • the feature value of the store is obtained from any of the following dimensions: the daily average number, the pen unit price, the store-owned merchant ID, and the location feature, and based on the feature value, based on the KNN algorithm or based on the set store similarity Sorting methods, establishing a similar store model for the event store.
  • the purpose of this step is to establish a similar store model for the event-applied store, so as to prioritize the historical data of the marketing activities of the most similar stores, and make high-precision predictions on the marketing activities of the store.
  • different store feature values are obtained from multiple dimensions, including: daily average number, pen unit price, store-owned business ID, location feature, and multi-dimensional feature values for different stores, based on the set store similarity
  • the ranking method establishes a similar store model of the target store.
  • the set store similarity ranking method includes:
  • the store similarity ranking is determined according to the feature values of the candidate similar stores, the proportion of each dimension in the similar store ranking, and the closeness of the feature values of the similar stores and the corresponding dimensions of the target store.
  • the preset condition includes: belonging to the same city as the target store; and/or belonging to the same category as the target store. For example, a store that belongs to the same city as the target store and belongs to the same three-category category as a similar store candidate set, and then selects similar stores from the same dimension according to other dimensions.
  • the similar store model established for the target store includes sorting the similar stores, that is, sorting according to the similarity, so that when the marketing campaign is recalled to a similar store, the priority recall may be more preferred. Similar stores, because prioritizing historical data that is more similar to the store, can help improve the prediction accuracy of the activity. And when sorting similar stores in the process of establishing a similar store model, consider the proportion of each dimension in the similar store ranking, wherein the proportion of each dimension in the similar store ranking is determined according to the following priorities. proportion:
  • the activity uses the store itself as the most similar store
  • the average number of daily pens is equal to the number of high daily averages divided by the number of low daily averages;
  • Pen unit price similarity the pen unit price similarity, equal to the high pen unit price divided by the low pen unit price
  • the distance from the applicable store is less than the set value
  • the similar stores are defined as 8 sorting priorities of A to H, and the historical marketing activity data of the target store itself is used as the data of the highest priority in the forecasting process, and the priority is A;
  • the daily average number of similarities, the sorting priority is B, and the difference in the marketing effect between the two stores with large daily averages is also very large, so the feature value of the dimension is prior to the priority in the sorting, and the average number of days per day
  • the similarity is the relative value calculated by dividing the number of high daily averages by the number of low daily averages. For example, a store with a similar daily average number less than 1.5 is regarded as a similar store; the sort priority C is the pen price similarity, due to the effect of the activity.
  • the similarity of the unit price dimension is more important.
  • the value is taken from the high unit price divided by the low unit price.
  • the store with the similar unit price equal to 1.5 is a similar store;
  • the sort priority D is The same as the merchant ID of the target store;
  • the sorting priorities E to H are location features, including: the distance is less than 300 meters, the sorting priority E; the distance is less than 500 meters, the sorting priority F; the distance is less than 1000 , G is a sort priority; and the same area of the store, sort priority is H.
  • the KNN classification algorithm can also be used to output various stores similar to the target store for the multi-dimensional feature values of different stores.
  • the similar store of the activity-applied store is determined based on the similar store model, and the similar store marketing activity effect data is acquired, and the marketing campaign effect estimation value of the activity applicable to the store is obtained through data smoothing processing.
  • the purpose of this step is to obtain the activity effect prediction data of the store to which the event is applicable.
  • the marketing activity performance data of the applicable store and its similar stores it is necessary to select the activity effect data of the historical marketing activities whose issuance amount and nuclear sales volume are larger than the set value. That is, the following marketing process is performed on the demand marketing activities: selecting marketing activities in which the issuing amount and the nuclear sales volume are larger than the set value, for example, selecting activities of 10 similar stores.
  • the historical activities selected in this way have sufficient statistical samples to have a high degree of confidence. Specifically, the embodiment of the present application selects only those activities whose release amount is greater than 50 and the nuclear sales volume is greater than 10, thereby filtering out the garbage activity.
  • the activities of 10 similar stores are selected, and the activity effect data of the marketing activities of the stores are obtained, including the percentage increase of the customer unit price and the percentage of the repurchase rate of each of the stores before and after the marketing activity, and the effect data of the activity is adopted.
  • the following smoothing process taking the average value of the customer unit price increase percentage of the target store and its similar stores as the customer unit price estimate of the target store
  • the initial value, the average value of the resale rate increase percentage of the target store and its similar stores is taken as the estimated value of the target store's repurchase rate.
  • step S403 the campaign activity effect value of the activity applicable store is aggregated, and the merchant activity effect prediction value is obtained.
  • the purpose of this step is to make an overall prediction of the merchant's promotional activities based on the predicted sales value of the store's different stores.
  • different applicable stores of the activity are predicted according to the activity type, and the predicted values of the applicable stores are added to obtain an overall prediction of the activity of the merchant.
  • the present application also provides an apparatus for predicting the effectiveness of marketing activities.
  • FIG. 5 a schematic diagram of an apparatus for predicting marketing activity effects according to the present application is shown. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple. For related parts, refer to the corresponding description of the method embodiment.
  • the device embodiments described below are merely illustrative.
  • the application provides an apparatus for predicting the effectiveness of a marketing campaign, comprising:
  • the historical data obtaining unit 501 is configured to obtain historical marketing data of different stores as an input sample, and calculate marketing activity effect data of different stores before and after the marketing activity;
  • the preliminary estimating unit 502 is configured to select marketing event effect data of the target store, and obtain an initial estimated value of the marketing activity of the target store through data smoothing processing;
  • the estimated value calibration unit 503 is configured to construct a discount rate adjustment factor by linear fitting based on the marketing activity effect data of different stores, and use the discount rate adjustment factor to calibrate the initial value of the marketing activity effect, and obtain the obtained
  • the calibration value is the estimated value of the target store marketing activity.
  • the marketing activity performance data is a value increase of the business indicator before and after the marketing activity of the store, and at least includes any one of the following indicators: a percentage increase of the customer unit price and a percentage increase of the repurchase rate.
  • the campaign activity effect data of the selected target store includes the following processes: establishing a similar store model of the target store, determining a similar store of the target store based on the similar store model, and marketing campaign effect from the different store
  • the campaign activity data of the target store and its similar stores are selected in the data.
  • the preliminary estimating unit 502 includes a modeling subunit for establishing a similar store model of the target store, including:
  • the set store similarity ranking method includes:
  • the store similarity ranking is determined according to the feature values of the candidate similar stores, the proportion of each dimension in the similar store ranking, and the closeness of the feature values of the similar stores and the corresponding dimensions of the target store.
  • the preset condition includes:
  • the data smoothing process obtains an initial value of the marketing activity effect estimation of the target store, and includes the following smoothing process: taking the average value of the customer unit price increase percentage of the target store and the similar store as the target store The initial value of the customer unit price is estimated, and the average value of the resale rate increase percentage of the target store and its similar stores is taken as the initial value of the target store's repurchase rate.
  • the preliminary estimating unit 502 includes an activity screening sub-unit, configured to select activity performance data of a historical marketing activity in which the issuing amount and the nuclear sales volume are greater than a set value.
  • the discount activity adjustment factor is constructed by linear fitting according to the effect data of the marketing activities of different stores, including: a percentage increase of the customer unit price based on the historical marketing activity and a percentage increase of the repurchase rate, and the linearization by the least squares method
  • the formula for calculating the discount rate adjustment factor with the discount rate is obtained.
  • the linear fitting is performed by a least squares method, and a formula for calculating a discount rate adjustment factor with a discount rate is obtained, including:
  • the discount rate adjustment factor is calculated according to the discount rate: the discount rate adjustment factor is equal to the slope multiplied by the discount rate plus the intercept;
  • the intercept is equal to the average of the activity effect minus the slope multiplied by the average of the discount rate; the discount rate of each similar store is multiplied by the average difference of the activity effect, and the discount rate is divided by the discount rate of each similar store.
  • the sum of squares of the differences is obtained, that is, the slope is obtained; the deviation from the mean is the distance from which the actual value deviates from the average value.
  • the calibrating the initial value of the marketing activity effect using the discount rate adjustment factor includes using the following calibration formula:
  • the target store marketing activity effect estimated value is equal to the target store's marketing activity effect estimated initial value multiplied by the discount rate adjustment factor.
  • the method for predicting the effectiveness of the marketing activity further includes: a threshold adjustment unit, configured to further correct the target store marketing activity effect estimate by using a threshold adjustment factor; wherein the threshold adjustment factor Is a factor that calculates the impact of the threshold on the effectiveness of the marketing campaign based on the ratio of the number of orders for which the historical marketing campaign meets the threshold and the number of orders that satisfy the customer's unit price, determined by:
  • the historical activity effect corresponding to the ratio is divided by the historical activity effect corresponding to the ratio of 1, and the obtained value is the threshold adjustment factor.
  • the estimated value of the target store marketing activity effect is further corrected by using a threshold adjustment factor, including: multiplying the threshold adjustment factor by the target store marketing activity effect estimation value to obtain a target store A further revised estimate of the effectiveness of marketing campaigns.
  • the marketing activity includes at least one of the following activity types: full reduction, each full reduction, and consumption delivery.
  • the calculating the performance data of the marketing activities of the different stores before and after the marketing campaign refers to calculating the customer unit price and the percentage of the customer unit price before and after the marketing activity according to the customer unit price and the repurchase rate in a certain period of time before and after the start of the marketing campaign. And the repurchase rate increased by a percentage point.
  • the present application also provides an electronic device for implementing the method for predicting the effectiveness of a marketing campaign.
  • FIG. 6 a schematic diagram of an electronic device provided by the embodiment is shown.
  • the embodiment of the electronic device provided by the present application is described in a relatively simple manner.
  • the embodiments described below are merely illustrative.
  • the application provides an electronic device, including:
  • the memory 601 is configured to store computer executable instructions, and the processor 602 is configured to execute the computer executable instructions:
  • a discount rate adjustment factor is constructed by linear fitting, and the initial value of the marketing activity effect is calibrated by using the discount rate adjustment factor, and the obtained calibration value is the target store marketing. Estimated activity performance.
  • the marketing activity performance data is a value increase of the business indicator before and after the marketing activity of the store, and at least includes any one of the following indicators: a percentage increase of the customer unit price and a percentage increase of the repurchase rate.
  • the campaign activity data of the selected target store includes the following processing:
  • Establishing a similar store model of the target store determining a similar store of the target store based on the similar store model, and selecting marketing campaign effect data of the target store and the similar store from the marketing campaign effect data of the different stores.
  • processor 602 is further configured to execute the following computer executable instructions:
  • the set store similarity ranking method includes:
  • the store similarity ranking is determined according to the feature values of the candidate similar stores, the proportion of each dimension in the similar store ranking, and the closeness of the feature values of the similar stores and the corresponding dimensions of the target store.
  • the preset condition includes:
  • the data smoothing process obtains an initial value of the marketing activity effect estimation of the target store, and includes the following smoothing process: taking the average value of the customer unit price increase percentage of the target store and the similar store as the target store The initial value of the customer unit price is estimated, and the average value of the resale rate increase percentage of the target store and its similar stores is taken as the initial value of the target store's repurchase rate.
  • the processor 602 is further configured to execute the following computer executable instructions: selecting activity effect data of the historical marketing activity in which the issue amount and the core sales volume are greater than the set value.
  • the discount activity adjustment factor is constructed by linear fitting according to the effect data of the marketing activities of different stores, including: a percentage increase of the customer unit price based on the historical marketing activity and a percentage increase of the repurchase rate, and the linearization by the least squares method
  • the formula for calculating the discount rate adjustment factor with the discount rate is obtained.
  • the linear fitting is performed by a least squares method, and a formula for calculating a discount rate adjustment factor with a discount rate is obtained, including:
  • the discount rate adjustment factor is calculated according to the discount rate: the discount rate adjustment factor is equal to the slope multiplied by the discount rate plus the intercept;
  • the intercept is equal to the average of the activity effect minus the slope multiplied by the average of the discount rate; the discount rate of each similar store is multiplied by the average difference of the activity effect, and the discount rate is divided by the discount rate of each similar store.
  • the sum of squares of the differences is obtained, that is, the slope is obtained; the deviation from the mean is the distance from which the actual value deviates from the average value.
  • the calibrating the initial value of the marketing activity effect using the discount rate adjustment factor includes using the following calibration formula:
  • the target store marketing activity effect estimated value is equal to the target store's marketing activity effect estimated initial value multiplied by the discount rate adjustment factor.
  • the processor 602 is further configured to execute the following computer executable instructions: the target store marketing activity effect estimation value is further corrected by using a threshold adjustment factor; wherein the threshold adjustment factor is based on history The ratio of the number of orders that meet the threshold of the marketing activity and the number of orders that satisfy the customer's unit price.
  • the factor that affects the effect of the marketing campaign on the effectiveness of the marketing campaign is determined by:
  • the historical activity effect corresponding to the ratio is divided by the historical activity effect corresponding to the ratio of 1, and the obtained value is the threshold adjustment factor.
  • the estimated value of the target store marketing activity effect is further corrected by using a threshold adjustment factor, including: multiplying the threshold adjustment factor by the target store marketing activity effect estimation value to obtain a target store A further revised estimate of the effectiveness of marketing campaigns.
  • the marketing activity includes at least one of the following activity types: full reduction, each full reduction, and consumption delivery.
  • the processor 602 is further configured to execute the following computer executable instructions: calculating a customer unit price increase percentage before and after the marketing activity according to the customer unit price and the repurchase rate in a certain period of time before and after the start of the marketing campaign of the store. And the repurchase rate increased by a percentage point.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • the memory may include non-persistent memory, random access memory (RAM), and/or non-volatile memory in a computer readable medium, such as read only memory (ROM) or flash memory.
  • RAM random access memory
  • ROM read only memory
  • Memory is an example of a computer readable medium.
  • Computer readable media including both permanent and non-persistent, removable and non-removable media may be implemented by any method or technology.
  • the information can be computer readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory. (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disk read only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, Magnetic tape cartridges, magnetic tape storage or other magnetic storage devices or any other non-transportable media can be used to store information that can be accessed by a computing device.
  • computer readable media does not include non-transitory computer readable media, such as modulated data signals and carrier waves.
  • embodiments of the present application can be provided as a method, system, or computer program product.
  • the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment in combination of software and hardware.
  • the application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.

Landscapes

  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种用于营销活动效果预测的方法，通过获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据（S101）；选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值（S102）；基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值（S103），从而解决了现有的营销活动效果预估方案无法根据折扣率的改变而实时地更准确地预估营销活动效果的问题。

Description

一种用于营销活动效果预测的方法、装置及电子设备
本申请要求于2017年11月29日提交中国专利局、申请号为201711222432.7、发明名称为“一种用于营销活动效果预测的方法、装置及电子设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及O2O餐饮行业营销领域，具体涉及一种用于营销活动效果预测的方法。本申请同时涉及一种用于营销活动效果预测装置，以及一种用于实现用于营销活动效果预测的方法的电子设备。
背景技术
在日益激烈的竞争环境中，O2O餐饮行业商家的营销手段层出不穷，各种名目的促销活动不断推出。商家在创建一个营销活动的时候，需要了解该活动的效果，即其能给门店给商家带来什么利益，这能让商家做到心中有数，明确营销活动的价值，有计划地按部就班做营销活动。
目前，对于营销活动效果的预估，通常做法是参考该门店的历史的类似营销活动效果数据，即以前的效果和现在创建的活动效果应该也类似，并且把折扣率当成营销活动的一个维度，通过简单地统计不同折扣率活动的历史情况来计算历史相似活动的效果。
现有的营销活动预测方案，并未将折扣率变化作为关键因素定量预估活动带来的效果，因而存在无法根据折扣率的改变而实时地更准确地预估营销活动效果的问题。事实上，折扣率对于很多活动效果的BI指标(Business Indicators，或商业指标)影响很大，例如一张有门槛的券，门槛一般会比门店客单价高，如果该券折扣力度够大，足以吸引人，那么很多用户为了达到券的使用门槛就会多消费，从而提高了门店客单价，再例如，对于复购率，如果发送的消费送券足够吸引人，能拉动用户的二次回头，复购率自然往上升。
发明内容
本申请提供一种用于营销活动效果预测的方法，以解决现有的营销活动效果预估方案无法根据折扣率的改变而实时地更准确地预估营销活动效果的问题。
本申请另外提供一种用于营销活动效果预测的装置。
本申请还提供一种实现用于营销活动效果预测的方法的电子设备。
本申请提供一种用于营销活动效果预测的方法，包括：
获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
可选的，所述营销活动效果数据，是门店在营销活动前后的商业指标提升值，至少包括下述任一指标：客单价提升百分点和复购率提升百分点。
可选的，所述选取目标门店的营销活动效果数据，包括下述处理：
建立目标门店的相似门店模型，基于所述相似门店模型确定目标门店的相似门店，从所述不同门店的营销活动效果数据中选取所述目标门店及其相似门店的营销活动效果数据。
可选的，所述建立目标门店的相似门店模型，包括：
从一个或多个维度获取门店特征值，基于KNN算法或者基于设定的门店相似度排序方法，根据所述特征值建立所述目标门店的相似门店模型；其中，所述一个或多个维度，至少包括下述任一维度：日均笔数、笔单价、门店所属商家ID、位置特征。
可选的，所述设定的门店相似度排序方法，包括：
通过预设条件筛选出所述目标门店的候选相似门店；
根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述目标门店对应维度的特征值接近程度，确定门店相似度排序。
可选的，所述预设条件，包括：
与目标门店属于同城；和/或
与目标门店属于同类目。
可选的，所述经过数据平滑处理得到所述目标门店的营销活动效果预估初值，包括下述平滑处理：取目标门店及其相似门店的客单价提升百分点的均值作为所述目标门店的客单价预估初值，取目标门店及其相似门店的复购率提升百分点的均值作为所述目标门店的复购率预估初值。
可选的，所述选取目标门店的营销活动效果数据，是选取发放量和核销量 大于设定值的历史营销活动的活动效果数据。
可选的，所述基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，包括：基于历史营销活动的客单价提升百分点和复购率提升百分点，通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式。
可选的，所述通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式，包括：
利用折扣率和营销活动效果，拟合出直线的截距和斜率，则折扣率调节因子随折扣率变化的计算公式为：折扣率调节因子等于斜率乘以折扣率再加上截距；
其中，截距等于活动效果平均值减去斜率乘以折扣率平均值；将各相似门店的折扣率离均差乘以活动效果离均差，求和后再除以各相似门店的折扣率离均差平方和，即得到斜率；所述离均差为实际值偏离平均值的距离。
可选的，所述使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，包括采用下述校准公式：
所述目标门店营销活动效果预估值等于所述目标门店的营销活动效果预估初值乘以折扣率调节因子。
可选的，还包括针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正；其中，所述门槛调节因子，是根据历史营销活动满足门槛的订单数和满足客单价的订单数的比值计算门槛对营销活动效果影响的因子，通过下述方式确定：
根据所述目标门店营销活动设定的门槛确定其对应的上述比值；
使用该比值对应的历史活动效果除以该比值为1时所对应的历史活动效果，得到的该数值，即为门槛调节因子。
可选的，所述针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正，包括：将所述门槛调节因子乘以所述目标门店营销活动效果预估值，得到目标门店的经过进一步修正的营销活动效果预估值。
可选的，所述营销活动，至少包括下述任一活动类型：满减、每满减、消费送。
可选的，所述计算营销活动前后不同门店的营销活动效果数据，是指根据门店营销活动开始前后一定时间段内的客单价和复购率，分别计算门店在营销 活动前后的客单价提升百分点和复购率提升百分点。
本申请还提供一种用于商家活动效果预测的方法，包括：
从下述任一维度获取门店的特征值：日均笔数、笔单价、门店所属商家ID、位置特征，根据所述特征值，基于KNN算法或者基于设定的门店相似度排序方法，建立活动适用门店的相似门店模型；
基于所述相似门店模型确定活动适用门店的相似门店，获取相似门店营销活动效果数据，经过数据平滑处理得到该活动适用门店的营销活动效果预测值；
对活动适用门店的营销活动效果值进行聚合，得到商家活动效果预测值。
可选的，所述设定的门店相似度排序方法，包括：
通过预设条件筛选出所述活动适用门店的候选相似门店；
根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述活动适用门店对应维度的特征值接近程度，确定门店相似度排序。
可选的，所述预设条件，包括：
与活动适用门店属于同城；和/或
与活动使用门店属于同类目。
可选的，所述各维度在相似门店排序中所占比重，是根据下述优先次序确定各维度比重：
所述活动使用门店本身为最相似门店；
日均笔数相似度；所述日均笔数相似度，等于高日均笔数除以低日均笔数；
笔单价相似度；所述笔单价相似度，等于高笔单价除以低笔单价；
与所述活动适用门店所属商家ID相同；
与活动适用门店的距离少于设定值；
与活动适用门店同属一个行政区。
本申请还提供一种用于营销活动效果预测的装置，包括：
历史数据获取单元，用于获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
初步预估单元，用于选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
预估值校准单元，用于基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值 进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
本申请还提供一种电子设备，包括：
存储器，以及处理器；
所述存储器用于存储计算机可执行指令，所述处理器用于执行所述计算机可执行指令：
获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
与现有技术相比，本申请具有以下优点：
本申请提供的用于营销活动效果预测的方法、装置及电子设备，通过获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值，根据门店的营销活动效果的历史离线数据得到目标门店的营销活动效果预估初值，并使用这些离线数据构建出折扣率调节因子，对该营销活动效果预估值进一步调节，从而解决了现有的营销活动效果预估方案无法根据折扣率的改变而实时地更准确地预估营销活动效果的问题。
附图说明
图1是本申请实施例提供的用于营销活动效果预测的方法的处理流程图；
图2是本申请实施例提供的用于营销活动效果预测的方法的实际部署举例的系统示意图；
图3是本申请实施例实际部署后交互效果图；
图4是本申请实施例提供的用于商家营销活动效果预测的方法的处理流程图；
图5是本申请实施例提供的用于营销活动效果预测的装置的示意图；
图6是本申请实施例提供的用于实现所述用于营销活动效果预测的方法的电子设备示意图。
具体实施方式
在下面的描述中阐述了很多具体细节以便于充分理解本申请。但是本申请能够以很多不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本申请内涵的情况下做类似推广，因此本申请不受下面公开的具体实施的限制。
本申请提供一种用于营销活动效果预测的方法。本申请同时涉及一种用于营销活动效果预测的装置以及一种用于实现用于营销活动效果预测的电子设备。在下面的实施例中逐一进行详细说明。
本申请其一实施例提供一种用于营销活动效果预测的方法。
以下结合图1至图3对本申请其一实施例提供的一种用于营销活动效果预测的方法的实施例进行说明。其中图1是本申请其一实施例提供的用于营销活动效果预测的方法的处理流程图；图2是本申请其一实施例提供的用于营销活动效果预测的方法的实际应用举例的系统示意图；图3是本申请其一实施例实际部署后交互效果图。
图1所示用于营销活动效果预测的方法，包括：
步骤S101，获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
步骤S102，选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
步骤S103，基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
目前，O2O餐饮行业商家营销策划和营销方案不断推陈出新，对营销活动效果进行更准确的预测，则让商家能明确营销活动的价值，有计划地按部就班做营销活动，而且对营销活动的效果进行实时预测，可以帮助商家更合理的配置营销活动资源、费用以及合适的备选方案，从而设置更符合市场预期和营销目的的活动。现有通用预测方法是参考该门店的历史的类似营销活动效果数据，即以前的效果和现在创建的活动效果应该也类似。事实上，仅仅这些并不能对 即将部署的营销活动的效果进行准确的实时的预测。其中，折扣率对于很多活动效果的BI指标(Business Indicators，或商业指标)影响很大，例如一张有门槛的券，门槛一般会比门店客单价高，如果该券折扣力度够大，足以吸引人，那么很多用户为了达到券的使用门槛就会多消费，从而提高了门店客单价，因而折扣率是影响预测营销活动效果的关键因素之一，而不能仅仅是简单地统计不同折扣率活动的历史情况来计算历史相似活动的效果；另外，现有的预估方法也没有考虑门店经营状况的波动情况。
本申请实施例以O2O餐饮行业的商家营销活动效果预估为例，对所述的用于营销活动效果预测的方法进行详细说明。本申请实施例中，显式地针对折扣率构建模型，随着折扣率的改变而实时地更准确地预估活动效果，而不是简单地把折扣率当成活动的一个维度仅统计不同折扣率活动的历史情况，在计算历史相似活动的效果时考虑了门店周边的相似门店的历史营销活动效果，并用这些相似门店的数据一起构建模型做预测，以每活动每门店的粒度预估活动效果，从而能够对营销活动效果进行更准确的实时的预测，商家也可以对于上线活动的适用门店的预估效果进行聚合获得整体的预测结果。所谓相似门店是在营销活动预测过程中，针对目标门店选择的与目标门店的特征接近的门店。其中，所谓目标门店，是指商家推出营销活动时所选择的上线该活动的适用活动门店，即按门店粒度上线营销活动，例如口碑网的商家可以自由灵活选择旗下哪个门店适用哪个促销活动。
所述步骤S101，获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据。
利用门店的历史营销活动数据，对即将上线的营销活动的效果进行事前预测，得到的可靠的活动效果预测结果可以作为预先建立的对比对象，该对比对象用于在活动中期或者活动结束后评估营销活动，通过将真实的营销情况与该对比对象进行对照分析，从而指导门店或商家的经营。
本步骤的目的是，基于不同门店的历史营销原始数据，计算出不同门店历史营销活动效果数据。
所谓营销活动效果数据，是门店在营销活动前后的商业指标提升值，至少包括下述任一指标：客单价提升百分点和复购率提升百分点。所述计算营销活动前后不同门店的营销活动效果数据，是指根据门店营销活动开始前后一定时间段内的客单价和复购率，分别计算门店在营销活动前后的客单价提升百分点 和复购率提升百分点，例如，拉取营销活动开始前后一个月内的原始数据进行计算，不同门店可以是同一商家的各门店，也可以是不同商家的门店。其中，所述客单价，是指每一位顾客平均购买商品金额；所述复购率是指某特定时间窗口下，二次消费的回头客占总交易UV的比例，所谓UV，是独立访客(Unique Visitor)，访问网站的一台电脑客户端为一个访客，00:00-24:00内相同的客户端只会被计算一次，显然促销活动能提升UV量。
O2O餐饮行业商家营销活动名目繁多，至少包括下述任一活动类型：满减、每满减、消费送。其中，所述满减，是一种活动类型，指满足门槛就可以享受对应的优惠，如满100减10元；所述每满减，是与所述满减不同的另一种活动类型，指每满足门槛多少倍就可以享受对应的优惠倍数，如每满100减10元，消费者消费300元即可优惠30元；所述消费送，是另一种活动类型，指成功消费后会自动领取一张优惠券，其优惠力度一般比门店对应的满减每满减活动要大，在用户二次消费的时候可以使用。
本申请实施例中，商家可以按照门店粒度选择推出哪种促销活动，并能对推出活动的目标门店在活动前预测其效果，在预测过程中综合使用与该目标门店相似的门店数据，从而平衡该目标门店经营状况波动对预测效果的影响。具体的，拉取过去一年的数据来统计每活动每门店的客单价和复购率提升百分点，对过去一年的每活动每门店，统计活动开始前后一个月的各自的客单价和复购率，然后计算相应的提升百分点，即为该类型活动该门店的客单价和复购率提升百分点的历史活动效果数据。如果一个门店历史上有多个同类型的活动，则客单价提升百分点和复购率提升百分点取多个活动的平均即可。
需要说明的是，实际处理中在对历史活动数据按照每活动每门店粒度进行分析时，活动类型可以细分，也可以为了不同的分析目的将活动按大类划分，例如，将活动分为两大类，包括满减/每满减类活动和消费送类活动，分别针对用户的首次交易行为和回头交易行为，即针对用户的首次交易行为，将满减/每满减划归一大类；也可以为了更细分析数据将其作为两个单独活动类型，如分析每细分类的活动成本利润率。另外，当底层数据不完备时，例如周边相似门店历史上未做过活动，则根据经验进行调整，放弃未做活动的店，或者使用业务全局表现的经验值进行兜底。
所述步骤S102，选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值。
本步骤的目的是，基于门店历史活动数据初步预估出所述目标门店的活动效果预估初值。
从步骤S101获得的不同门店的历史营销活动效果数据中，提取出目标门店的营销活动效果数据，根据这些营销活动历史数据，对目标门店推出的营销活动效果进行事前预测。需要说明的是，可以选取目标门店自身的历史数据进行预测，也可以同时选取其相似门店的营销活动历史数据，共同作为对该目标门店的营销活动效果进行预测的数据基础。同时考虑这些相似门店的营销活动历史数据，能够对目标门店推出的营销活动的效果做出更可靠的预测。
本申请实施例中，所谓活动效果是指客单价和复购率提升程度，对于活动效果的预估包括预测营销活动的客单价提升百分点和复购率提升百分点，在进行营销活动预测时，通过建立目标门店的相似门店模型来选取目标门店及其相似门店的营销活动效果数据。
具体的，所述选取目标门店的营销活动效果数据，包括下述处理：建立目标门店的相似门店模型，基于所述相似门店模型确定目标门店的相似门店，从所述不同门店的营销活动效果数据中选取所述目标门店及其相似门店的营销活动效果数据。
所述建立目标门店的相似门店模型，包括：从一个或多个维度获取门店特征值，基于KNN算法或者基于设定的门店相似度排序方法，根据所述特征值建立所述目标门店的相似门店模型；其中，所述一个或多个维度，至少包括下述任一维度：日均笔数、笔单价、门店所属商家ID、位置特征。
本申请实施例中，针对不同门店的多维度特征值，基于设定的门店相似度排序方法建立所述目标门店的相似门店模型，具体的，所述设定的门店相似度排序方法，包括：
通过预设条件筛选出所述目标门店的候选相似门店；
根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述目标门店对应维度的特征值接近程度，确定门店相似度排序。其中，所述预设条件，包括：与目标门店属于同城；和/或与目标门店属于同类目。例如，将与目标门店属于同城，并且属于同三级类目的门店作为相似门店候选集，然后再按照其他维度从其中筛选符合条件的相似门店。
本申请实施例中，从多个维度获取门店特征值，包括：日均笔数、笔单价、门店所属商家ID、位置特征。
所谓日均笔数，是平均每天的交易笔数，一张水单为一笔交易。
所述笔单价，指每一笔交易记录(一张水单)对应的平均交易金额，一般以消费总金额除以消费笔数来计量。
所述门店所属商家ID，即门店pid(或partner ID)，例如KFC的门店pid是百胜集团，O2O餐饮业商家在推广网站注册时会得到注册的pid，其下属各同门店一般共用该pid，同一个pid的门店相似度很高。
所述位置特征，是与目标门店所处地理位置的远近以及所处位置是否在同一行政区域，在O2O餐饮行业中，距离近的门店由于客户群体范围相似因而一般作为选取相似门店的一个比较维度，例如，和该目标门店所处位置同属一个行政区。
另外，本申请实施例中，对目标门店建立的相似门店模型，包括了对相似门店分档次，也即按照相似度对其进行排序，这样在选择相似门店召回营销活动的时候，可以优先召回更相似的门店，由于优先考虑更相似门店的历史数据，能有助于提高活动效果的预测精度；其中，所谓召回营销活动，是指获取所述营销活动的活动数据及其他相关信息。并且在建立相似门店模型过程中对相似门店排序时，考虑各维度在相似门店排序中所占比重，其中，所述各维度在相似门店排序中所占比重，是根据下述优先次序确定各维度比重：
所述目标门店本身为最相似门店；
日均笔数相似度；所述日均笔数相似度，等于高日均笔数除以低日均笔数；
笔单价相似度；所述笔单价相似度，等于高笔单价除以低笔单价；
与所述目标门店所属商家ID相同；
与目标门店的距离少于设定值；
与目标门店同属一个行政区。
具体的，本申请实施例，对相似门店定义为A至H共8个排序优先级，将目标门店自身的历史营销活动数据作为预测过程中最优先考率的数据，优先级为A；其次，日均笔数相似度，排序优先级为B，日均笔数差别很大的两家门店其营销效果差别也很大，因此该维度的特征值在排序中优先级靠前，日均笔数相似度是用高日均笔数除以低日均笔数计算的相对值，例如将日均笔数相似度小于1.5的门店作为相似门店；排序优先级C为笔单价相似度，由于活动效果涉及到销售额，因此笔单价维度的相似程度比较重要，另外，该值取自于高笔单价除以低笔单价，例如笔单价相似度小等于1.5的门店为相似门店；排序优先级 D为，与所述目标门店所属商家ID相同；排序优先级E至H为位置特征，包括：距离少于300米，其排序优先级E；距离少于500米，排序优先级F；距离少于1000米，排序优先级G为；和该门店同一个区，排序优先级为H。
需要说明的是，选择目标门店的相似门店时，也可以采用KNN分类算法，针对不同门店的多维度特征值，输出和目标门店相似的各门店，所谓KNN算法，即K最近邻(kNN，或k-NearestNeighbor)分类算法，是数据挖掘分类技术方法之一。所谓K最近邻，是指每个样本都可以用它最接近的k个邻居来代表，其核心思想是如果一个样本在特征空间中的k个最相邻的样本中的大多数属于某一个类别，则该样本也属于这个类别，并具有这个类别上样本的特性。该方法在确定分类决策上只依据最邻近的一个或者几个有限的样本的类别来决定待分样本所属的类别。
另外，所述选取目标门店及其相似门店的营销活动效果数据，是选取发放量和核销量大于设定值的历史营销活动的活动效果数据。即在选择出目标门店的相似门店后，对营销活动进行下述筛选处理：选取发放量和核销量大于设定值的营销活动，例如选取10个相似门店的活动。即在选择相似门店的活动的过程中，还需要对历史的活动做一定程度的筛选，因为对历史活动而言，需要该历史活动有充分的统计样本，才能有比较高的置信度，才能作为样本来构建模型。具体的，本申请实施例是只选取那些发放量大于50核销量大于10的活动。所述核销，是指营销活动所发的券被线下交易中使用。券或卡发放量阈值的设置目的，一方面是过滤掉垃圾活动，有些营销平台确实存在很多无效活动，这种活动用户根本无法感知，所以也没必要用作未来预测的参考，二是发放量大于50，则可以认为这种活动的各项BI指标比如核销率，客单价提升，复购率提升才有统计意义。卡核销量大于10的阈值是一个软条件(所谓软条件，不是强制条件)，即优先选择满足这个条件的活动，当满足条件的活动数量达不到设定值就放松这个条件的限制，以便能尽量选取足够多的活动。卡核销量阈值是为了尽量选择质量较高的活动，因为随着平台的发展，活动质量已经越来越高，垃圾活动越来越少，质量好的活动参考意义自然更大。
本申请实施例中选择10家相似门店的活动，获取这些门店营销活动的活动效果数据，包括这些门店各自在营销活动前后的客单价提升百分点和复购率提升百分点，通过对所述活动效果数据进行数据平滑处理得到所述目标门店的营销活动效果预估初值。所述经过数据平滑处理得到所述目标门店的营销活动效 果预估初值，包括下述平滑处理：取目标门店及其相似门店的客单价提升百分点的均值作为所述目标门店的客单价预估初值，取目标门店及其相似门店的复购率提升百分点的均值作为所述目标门店的复购率预估初值。
另外，如果一个门店历史上有多个同类型的活动，则客单价，复购率的提升百分点取多个活动的平均。当底层数据不完备时，例如周边相似门店历史上未做过活动，则根据经验进行调整，放弃未做活动的店，或者使用业务全局表现的经验值进行兜底。对于上述处理中获取到的每活动每门店的活动效果数据取其平均值作为该类型营销活动的效果预估初值。
所述步骤S103，基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
由于折扣率变化对客单价和复购率等BI指标影响很大，因此对营销活动效果进行更精确的实时的预测则需要将折扣率的影响考虑在内。
本步骤的目的，是根据不同门店历史活动数据，通过线性拟合构建折扣率调节因子，对于步骤S102中得到的目标门店活动效果预估初值进行校准。
所述基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，包括：基于历史营销活动的客单价提升百分点和复购率提升百分点，通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式。需要说明的是，构建折扣率调节因子所使用的样本数据，可以是相似门店的历史活动数据，也可以是统计同城同类型的所有活动在特定折扣率下的活动效果数值均值，用这些样本数据对折扣率进行线性拟合，得到拟合出的直线对应的斜率和截距。具体到本申请实施例，以同城同类型活动的客单价提升百分点和复购率提升百分点作为样本，并且选择样本时只考虑大于等于5折的活动，因为小于5折的活动数量太少，活动效果波动明显。
具体的，所述通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式，包括：利用折扣率和营销活动效果，拟合出直线的截距和斜率，则折扣率调节因子随折扣率变化的计算公式为：折扣率调节因子等于斜率乘以折扣率再加上截距；
其中，截距等于活动效果平均值减去斜率乘以折扣率平均值；将各相似门店的折扣率离均差乘以活动效果离均差，求和后再除以各相似门店的折扣率离均差平方和，即得到斜率；所述离均差为实际值偏离平均值的距离。
本申请实施例通过最小二乘拟合出的斜率b和截距a的公式如下：
Figure PCTCN2018115339-appb-000001
Figure PCTCN2018115339-appb-000002
式中，
Figure PCTCN2018115339-appb-000003
则折扣率调节因子计算方法：折扣率调节因子＝b*折扣率+a。使用折扣率调节因子对所述营销活动效果预估初值进行校准，包括采用下述校准公式：
所述目标门店营销活动效果预估值等于所述目标门店的营销活动效果预估初值乘以折扣率调节因子，即：
预测客单价提升百分点＝客单价提升百分点预估初值*折扣率调节因子；
预测复购率提升百分点＝复购率提升百分点预估初值*折扣率调节因子。
另外，由于活动门槛反映了参与活动的人群特征和人群规模，对于营销商业指标的改善和活动的落地有直接的影响，因此，本申请实施例中提供的用于营销活动效果预测的方法，还包括针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正，提高预测准确性。其中，所述门槛调节因子，是根据历史营销活动满足门槛的订单数和满足客单价的订单数的比值计算门槛对营销活动效果影响的因子，通过下述方式确定：
根据所述目标门店营销活动设定的门槛确定其对应的上述比值；
使用该比值对应的历史活动效果除以该比值为1时所对应的历史活动效果，得到的该数值，即为门槛调节因子。
其中，所述针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正，包括：
将所述门槛调节因子乘以所述目标门店营销活动效果预估值，得到目标门店的经过进一步修正的营销活动效果预估值。
具体到本申请实施例，对所述目标门店营销活动效果预估初值，经过折扣率调节因子和门槛调节因子两级修正，得到最终的活动效果预测值，实现时这两级调节的先后顺序不影响预测结果。该营销活动效果预测方案实际部署后，使用者输入营销活动方案的相关信息，例如：促销方案类目、最大立减金额等，通过本申请实施例的处理，即能得到本次促销活动的预估效果，实际部署的系统示意图如图2所示，使用者与实际部署系统的交互效果如图3所示。
以本申请提供的用于营销活动效果预测的方法的实施例为基础，本申请还 提供了一种用于商家活动效果预测的方法。
参照图4，其示出了根据本申请提供的用于商家活动效果预测的方法处理流程图。由于本实施例以上述用于营销活动效果预测的方法的实施例为基础，所以描述得比较简单，相关的部分请参见上述实施例的对应说明即可。下述描述的用于商家活动效果预测的方法的实施例仅仅是示意性的。
本申请提供一种用于商家营销活动效果预测的方法，包括：
步骤S401，从下述任一维度获取门店的特征值：日均笔数、笔单价、门店所属商家ID、位置特征，根据所述特征值，基于KNN算法或者基于设定的门店相似度排序方法，建立活动适用门店的相似门店模型；
步骤S402，基于所述相似门店模型确定活动适用门店的相似门店，获取相似门店营销活动效果数据，经过数据平滑处理得到该活动适用门店的营销活动效果预估值；
步骤S403，对活动适用门店的营销活动效果值进行聚合，得到商家活动效果预测值。
本申请实施例以O2O餐饮行业为例，对所述用于商家活动效果预测的方法进行说明。
实际应用中，商家促销时按照门店粒度自由选择上线促销活动的适用门店，例如口碑网，在对活动效果进行预估时，首先针对活动适用门店的活动效果，再对上线活动的所有适用门店的活动效果预测值进行聚合获得商家整体的预测结果。所谓活动适用门店，是指商家推出营销活动时所选择的上线该活动的适用门店。
本申请实施例中，在针对适用上线活动的门店的促销效果进行预测时，将门店周边的相似门店的历史营销活动效果数据考虑在内一起做预测，而不是仅简单地统计该适用活动门店的历史情况，因而能够平衡门店经营波动，对营销活动效果进行更准确的预测。
所述步骤S401，从下述任一维度获取门店的特征值：日均笔数、笔单价、门店所属商家ID、位置特征，根据所述特征值，基于KNN算法或者基于设定的门店相似度排序方法，建立活动适用门店的相似门店模型。
本步骤的目的是建立活动适用门店的相似门店模型，从而优先考虑最相似门店的营销活动历史数据，对该门店上线的营销活动效果做出高精度预测。
本申请实施例中，从多个维度获取不同门店特征值，包括：日均笔数、笔 单价、门店所属商家ID、位置特征，并针对不同门店的多维度特征值，基于设定的门店相似度排序方法建立所述目标门店的相似门店模型，具体的，所述设定的门店相似度排序方法，包括：
通过预设条件筛选出所述目标门店的候选相似门店；
根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述目标门店对应维度的特征值接近程度，确定门店相似度排序。其中，所述预设条件，包括：与目标门店属于同城；和/或与目标门店属于同类目。例如，将与目标门店属于同城，并且属于同三级类目的门店作为相似门店候选集，然后再按照其他维度从其中筛选符合条件的相似门店。
另外，本申请实施例中，对目标门店建立的相似门店模型，包括了对相似门店分档次，也即按照相似度对其进行排序，这样在对相似门店召回营销活动的时候，可以优先召回更相似的门店，由于优先考虑更相似门店的历史数据，能有助于提高活动效果的预测精度。并且在建立相似门店模型过程中对相似门店排序时，考虑各维度在相似门店排序中所占比重，其中，所述各维度在相似门店排序中所占比重，是根据下述优先次序确定各维度比重：
所述活动使用门店本身为最相似门店；
日均笔数相似度；所述日均笔数相似度，等于高日均笔数除以低日均笔数；
笔单价相似度；所述笔单价相似度，等于高笔单价除以低笔单价；
与所述活动适用门店所属商家ID相同；
与活动适用门店的距离少于设定值；
与活动适用门店同属一个行政区。
具体的，本申请实施例，对相似门店定义为A至H共8个排序优先级，将目标门店自身的历史营销活动数据作为预测过程中最优先考率的数据，优先级为A；其次，日均笔数相似度，排序优先级为B，日均笔数差别很大的两家门店其营销效果差别也很大，因此该维度的特征值在排序中优先级靠前，日均笔数相似度是用高日均笔数除以低日均笔数计算的相对值，例如将日均笔数相似度小于1.5的门店作为相似门店；排序优先级C为笔单价相似度，由于活动效果涉及到销售额，因此笔单价维度的相似程度比较重要，另外，该值取自于高笔单价除以低笔单价，例如笔单价相似度小等于1.5的门店为相似门店；排序优先级D为，与所述目标门店所属商家ID相同；排序优先级E至H为位置特征，包括：距离少于300米，其排序优先级E；距离少于500米，排序优先级F；距离少于 1000米，排序优先级G为；和该门店同一个区，排序优先级为H。
另外，选择活动适用门店的相似门店时，也可以采用KNN分类算法，针对不同门店的多维度特征值，输出和目标门店相似的各门店。
所述步骤S402，基于所述相似门店模型确定活动适用门店的相似门店，获取相似门店营销活动效果数据，经过数据平滑处理得到该活动适用门店的营销活动效果预估值。
本步骤的目的是获取所述活动适用门店的活动效果预测数据。
在选取活动适用门店及其相似门店的营销活动效果数据时，需要选取发放量和核销量大于设定值的历史营销活动的活动效果数据。即对需求的营销活动进行下述筛选处理：选取发放量和核销量大于设定值的营销活动，例如选取10个相似门店的活动。这样选取的历史活动有充分的统计样本，才能有较高的置信度。具体的，本申请实施例是只选取那些发放量大于50核销量大于10的活动，从而过滤掉垃圾活动。
本申请实施例中选择10家相似门店的活动，获取这些门店营销活动的活动效果数据，包括这些门店各自在营销活动前后的客单价提升百分点和复购率提升百分点，通过对所述活动效果数据进行数据平滑处理得到所述目标门店的营销活动效果预估值，具体的，包括下述平滑处理：取目标门店及其相似门店的客单价提升百分点的均值作为所述目标门店的客单价预估初值，取目标门店及其相似门店的复购率提升百分点的均值作为所述目标门店的复购率预估值。
所述步骤S403，对活动适用门店的营销活动效果值进行聚合，得到商家活动效果预测值。
本步骤的目的是基于商家不同门店的促销活动预测值对商家的促销活动作出整体的预测。
具体到本申请实施例，按照活动类型对上线该活动的不同的适用门店作出预测，将各适用门店的预测值进行加和得到商家该类活动的整体预测。
与本申请提供的一种用于营销活动效果预测的方法的实施例相对应，本申请还提供了一种用于营销活动效果预测的装置。
参照图5，其示出了根据本申请提供的一种用于营销活动效果预测的装置示意图。由于装置实施例基本相似于方法实施例，所以描述得比较简单，相关的部分请参见方法实施例的对应说明即可。下述描述的装置实施例仅仅是示意性的。
本申请提供一种用于营销活动效果预测的装置，包括：
历史数据获取单元501，用于获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
初步预估单元502，用于选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
预估值校准单元503，用于基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
可选的，所述营销活动效果数据，是门店在营销活动前后的商业指标提升值，至少包括下述任一指标：客单价提升百分点和复购率提升百分点。
可选的，所述选取目标门店的营销活动效果数据，包括下述处理：建立目标门店的相似门店模型，基于所述相似门店模型确定目标门店的相似门店，从所述不同门店的营销活动效果数据中选取所述目标门店及其相似门店的营销活动效果数据。
可选的，初步预估单元502，包括建模子单元，用于建立目标门店的相似门店模型，包括：
从一个或多个维度获取门店特征值，基于KNN算法或者基于设定的门店相似度排序方法，根据所述特征值建立所述目标门店的相似门店模型；其中，所述一个或多个维度，至少包括下述任一维度：日均笔数、笔单价、门店所属商家ID、位置特征。
可选的，所述设定的门店相似度排序方法，包括：
通过预设条件筛选出所述目标门店的候选相似门店；
根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述目标门店对应维度的特征值接近程度，确定门店相似度排序。
可选的，所述预设条件，包括：
与目标门店属于同城；和/或
与目标门店属于同类目。
可选的，所述经过数据平滑处理得到所述目标门店的营销活动效果预估初值，包括下述平滑处理：取目标门店及其相似门店的客单价提升百分点的均值作为所述目标门店的客单价预估初值，取目标门店及其相似门店的复购率提升百分点的均值作为所述目标门店的复购率预估初值。
可选的，所述初步预估单元502，包括活动筛选子单元，用于选取发放量和核销量大于设定值的历史营销活动的活动效果数据。
可选的，所述基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，包括：基于历史营销活动的客单价提升百分点和复购率提升百分点，通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式。
可选的，所述通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式，包括：
利用折扣率和营销活动效果，拟合出直线的截距和斜率，则折扣率调节因子随折扣率变化的计算公式为：折扣率调节因子等于斜率乘以折扣率再加上截距；
其中，截距等于活动效果平均值减去斜率乘以折扣率平均值；将各相似门店的折扣率离均差乘以活动效果离均差，求和后再除以各相似门店的折扣率离均差平方和，即得到斜率；所述离均差为实际值偏离平均值的距离。
可选的，所述使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，包括采用下述校准公式：
所述目标门店营销活动效果预估值等于所述目标门店的营销活动效果预估初值乘以折扣率调节因子。
可选的，所述的用于营销活动效果预测的方法，还包括门槛调节单元，用于针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正；其中，所述门槛调节因子，是根据历史营销活动满足门槛的订单数和满足客单价的订单数的比值计算门槛对营销活动效果影响的因子，通过下述方式确定：
根据所述目标门店营销活动设定的门槛确定其对应的上述比值；
使用该比值对应的历史活动效果除以该比值为1时所对应的历史活动效果，得到的该数值，即为门槛调节因子。
可选的，所述针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正，包括：将所述门槛调节因子乘以所述目标门店营销活动效果预估值，得到目标门店的经过进一步修正的营销活动效果预估值。
可选的，所述营销活动，至少包括下述任一活动类型：满减、每满减、消费送。
可选的，所述计算营销活动前后不同门店的营销活动效果数据，是指根据 门店营销活动开始前后一定时间段内的客单价和复购率，分别计算门店在营销活动前后的客单价提升百分点和复购率提升百分点。
本申请还提供了一种用于实现所述用于营销活动效果预测的方法的电子设备，参照图6，其示出了本实施例提供的一种电子设备的示意图。
本申请提供的所述电子设备实施例描述得比较简单，相关的部分请参见上述提供的所述用于营销活动效果预测的方法的实施例的对应说明即可。下述描述的实施例仅仅是示意性的。
本申请提供一种电子设备，包括：
存储器601，以及处理器602；
所述存储器601用于存储计算机可执行指令，所述处理器602用于执行所述计算机可执行指令：
获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
可选的，所述营销活动效果数据，是门店在营销活动前后的商业指标提升值，至少包括下述任一指标：客单价提升百分点和复购率提升百分点。
可选的，所述选取目标门店的营销活动效果数据，包括下述处理：
建立目标门店的相似门店模型，基于所述相似门店模型确定目标门店的相似门店，从所述不同门店的营销活动效果数据中选取所述目标门店及其相似门店的营销活动效果数据。
可选的，所述处理器602还用于执行下述计算机可执行指令：
从一个或多个维度获取门店特征值，基于KNN算法或者基于设定的门店相似度排序方法，根据所述特征值建立所述目标门店的相似门店模型；其中，所述一个或多个维度，至少包括下述任一维度：日均笔数、笔单价、门店所属商家ID、位置特征。
可选的，所述设定的门店相似度排序方法，包括：
通过预设条件筛选出所述目标门店的候选相似门店；
根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述目标门店对应维度的特征值接近程度，确定门店相似度排序。
可选的，所述预设条件，包括：
与目标门店属于同城；和/或
与目标门店属于同类目。
可选的，所述经过数据平滑处理得到所述目标门店的营销活动效果预估初值，包括下述平滑处理：取目标门店及其相似门店的客单价提升百分点的均值作为所述目标门店的客单价预估初值，取目标门店及其相似门店的复购率提升百分点的均值作为所述目标门店的复购率预估初值。
可选的，所述处理器602还用于执行下述计算机可执行指令：选取发放量和核销量大于设定值的历史营销活动的活动效果数据。
可选的，所述基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，包括：基于历史营销活动的客单价提升百分点和复购率提升百分点，通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式。
可选的，所述通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式，包括：
利用折扣率和营销活动效果，拟合出直线的截距和斜率，则折扣率调节因子随折扣率变化的计算公式为：折扣率调节因子等于斜率乘以折扣率再加上截距；
其中，截距等于活动效果平均值减去斜率乘以折扣率平均值；将各相似门店的折扣率离均差乘以活动效果离均差，求和后再除以各相似门店的折扣率离均差平方和，即得到斜率；所述离均差为实际值偏离平均值的距离。
可选的，所述使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，包括采用下述校准公式：
所述目标门店营销活动效果预估值等于所述目标门店的营销活动效果预估初值乘以折扣率调节因子。
可选的，所述处理器602还用于执行下述计算机可执行指令：针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正；其中，所述门槛调节因子，是根据历史营销活动满足门槛的订单数和满足客单价的订单数的比值计算门槛对营销活动效果影响的因子，通过下述方式确定：
根据所述目标门店营销活动设定的门槛确定其对应的上述比值；
使用该比值对应的历史活动效果除以该比值为1时所对应的历史活动效果，得到的该数值，即为门槛调节因子。
可选的，所述针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正，包括：将所述门槛调节因子乘以所述目标门店营销活动效果预估值，得到目标门店的经过进一步修正的营销活动效果预估值。
可选的，所述营销活动，至少包括下述任一活动类型：满减、每满减、消费送。
可选的，所述处理器602还用于执行下述计算机可执行指令：根据门店营销活动开始前后一定时间段内的客单价和复购率，分别计算门店在营销活动前后的客单价提升百分点和复购率提升百分点。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
1、计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括非暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
2、本领域技术人员应明白，本申请的实施例可提供为方法、系统或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请虽然以较佳实施例公开如上，但其并不是用来限定本申请，任何本领域技术人员在不脱离本申请的精神和范围内，都可以做出可能的变动和修改，因此本申请的保护范围应当以本申请权利要求所界定的范围为准。

Claims (21)

  1. 一种用于营销活动效果预测的方法，其特征在于，包括：
    获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
    选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
    基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
  2. 根据权利要求1所述的用于营销活动效果预测的方法，其特征在于，所述营销活动效果数据，是门店在营销活动前后的商业指标提升值，至少包括下述任一指标：客单价提升百分点和复购率提升百分点。
  3. 根据权利要求1或2所述的用于营销活动效果预测的方法，其特征在于，所述选取目标门店的营销活动效果数据，包括下述处理：
    建立目标门店的相似门店模型，基于所述相似门店模型确定目标门店的相似门店，从所述不同门店的营销活动效果数据中选取所述目标门店及其相似门店的营销活动效果数据。
  4. 根据权利要求3所述的用于营销活动效果预测的方法，其特征在于，所述建立目标门店的相似门店模型，包括：
    从一个或多个维度获取门店特征值，基于KNN算法或者基于设定的门店相似度排序方法，根据所述特征值建立所述目标门店的相似门店模型；其中，所述一个或多个维度，至少包括下述任一维度：日均笔数、笔单价、门店所属商家ID、位置特征。
  5. 根据权利要求4所述的用于营销活动效果预测的方法，其特征在于，所述设定的门店相似度排序方法，包括：
    通过预设条件筛选出所述目标门店的候选相似门店；
    根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述目标门店对应维度的特征值接近程度，确定门店相似度排序。
  6. 根据权利要求5所述的用于营销活动效果预测的方法，其特征在于，所述预设条件，包括：
    与目标门店属于同城；和/或
    与目标门店属于同类目。
  7. 根据权利要求3所述的用于营销活动效果预测的方法，其特征在于，所述经过数据平滑处理得到所述目标门店的营销活动效果预估初值，包括下述平滑处理：取目标门店及其相似门店的客单价提升百分点的均值作为所述目标门店的客单价预估初值，取目标门店及其相似门店的复购率提升百分点的均值作为所述目标门店的复购率预估初值。
  8. 根据权利要求1所述的用于营销活动效果预测的方法，其特征在于，所述选取目标门店的营销活动效果数据，是选取发放量和核销量大于设定值的历史营销活动的活动效果数据。
  9. 根据权利要求2所述的用于营销活动效果预测的方法，其特征在于，所述基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，包括：基于历史营销活动的客单价提升百分点和复购率提升百分点，通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式。
  10. 根据权利要求9所述的用于营销活动效果预测的方法，其特征在于，所述通过最小二乘法进行线性拟合，得出折扣率调节因子随折扣率变化的计算公式，包括：
    利用折扣率和营销活动效果，拟合出直线的截距和斜率，则折扣率调节因子随折扣率变化的计算公式为：折扣率调节因子等于斜率乘以折扣率再加上截距；
    其中，截距等于活动效果平均值减去斜率乘以折扣率平均值；将各相似门店的折扣率离均差乘以活动效果离均差，求和后再除以各相似门店的折扣率离均差平方和，即得到斜率；所述离均差为实际值偏离平均值的距离。
  11. 根据权利要求1所述的用于营销活动效果预测的方法，其特征在于，所述使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，包括采用下述校准公式：
    所述目标门店营销活动效果预估值等于所述目标门店的营销活动效果预估初值乘以折扣率调节因子。
  12. 根据权利要求1所述的用于营销活动效果预测的方法，其特征在于，还包括针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正；其中，所述门槛调节因子，是根据历史营销活动满足门槛的订单数和满足客单价的订单数的比值计算门槛对营销活动效果影响的因子，通过下述方式确定：
    根据所述目标门店营销活动设定的门槛确定其对应的上述比值；
    使用该比值对应的历史活动效果除以该比值为1时所对应的历史活动效果，得到的该数值，即为门槛调节因子。
  13. 根据权利要求11所述的用于营销活动效果预测的方法，其特征在于，所述针对所述目标门店营销活动效果预估值，使用门槛调节因子进一步修正，包括：将所述门槛调节因子乘以所述目标门店营销活动效果预估值，得到目标门店的经过进一步修正的营销活动效果预估值。
  14. 根据权利要求1所述的用于营销活动效果预测的方法，其特征在于，所述营销活动，至少包括下述任一活动类型：满减、每满减、消费送。
  15. 根据权利要求2所述的用于营销活动效果预测的方法，其特征在于，所述计算营销活动前后不同门店的营销活动效果数据，是指根据门店营销活动开始前后一定时间段内的客单价和复购率，分别计算门店在营销活动前后的客单价提升百分点和复购率提升百分点。
  16. 一种用于商家活动效果预测的方法，其特征在于，包括：
    从下述任一维度获取门店的特征值：日均笔数、笔单价、门店所属商家ID、位置特征，根据所述特征值，基于KNN算法或者基于设定的门店相似度排序方法，建立活动适用门店的相似门店模型；
    基于所述相似门店模型确定活动适用门店的相似门店，获取相似门店营销活动效果数据，经过数据平滑处理得到该活动适用门店的营销活动效果预测值；
    对活动适用门店的营销活动效果值进行聚合，得到商家活动效果预测值。
  17. 根据权利要求16所述的用于商家活动效果预测的方法，其特征在于，所述设定的门店相似度排序方法，包括：
    通过预设条件筛选出所述活动适用门店的候选相似门店；
    根据候选相似门店各维度特征值、各维度在相似门店排序中所占比重，以及相似门店与所述活动适用门店对应维度的特征值接近程度，确定门店相似度排序。
  18. 根据权利要求17所述的用于商家活动效果预测的方法，其特征在于，所述预设条件，包括：
    与活动适用门店属于同城；和/或
    与活动使用门店属于同类目。
  19. 根据权利要求17所述的用于商家活动效果预测的方法，其特征在于， 所述各维度在相似门店排序中所占比重，是根据下述优先次序确定各维度比重：
    所述活动使用门店本身为最相似门店；
    日均笔数相似度；所述日均笔数相似度，等于高日均笔数除以低日均笔数；
    笔单价相似度；所述笔单价相似度，等于高笔单价除以低笔单价；
    与所述活动适用门店所属商家ID相同；
    与活动适用门店的距离少于设定值；
    与活动适用门店同属一个行政区。
  20. 一种用于营销活动效果预测的装置，其特征在于，包括：
    历史数据获取单元，用于获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
    初步预估单元，用于选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
    预估值校准单元，用于基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
  21. 一种电子设备，其特征在于，包括：
    存储器，以及处理器；
    所述存储器用于存储计算机可执行指令，所述处理器用于执行所述计算机可执行指令：
    获取不同门店的历史营销数据作为输入样本，计算营销活动前后不同门店的营销活动效果数据；
    选取目标门店的营销活动效果数据，经过数据平滑处理得到该目标门店的营销活动效果预估初值；
    基于不同门店的营销活动效果数据，通过线性拟合构建折扣率调节因子，使用所述折扣率调节因子对所述营销活动效果预估初值进行校准，获得的校准值即为所述目标门店营销活动效果预估值。
PCT/CN2018/115339 2017-11-29 2018-11-14 一种用于营销活动效果预测的方法、装置及电子设备 WO2019105226A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711222432.7 2017-11-29
CN201711222432.7A CN107886369A (zh) 2017-11-29 2017-11-29 一种用于营销活动效果预测的方法、装置及电子设备

Publications (1)

Publication Number Publication Date
WO2019105226A1 true WO2019105226A1 (zh) 2019-06-06

Family

ID=61775760

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/115339 WO2019105226A1 (zh) 2017-11-29 2018-11-14 一种用于营销活动效果预测的方法、装置及电子设备

Country Status (2)

Country Link
CN (1) CN107886369A (zh)
WO (1) WO2019105226A1 (zh)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107886369A (zh) * 2017-11-29 2018-04-06 口碑(上海)信息技术有限公司 一种用于营销活动效果预测的方法、装置及电子设备
CN108764958A (zh) * 2018-04-11 2018-11-06 口碑(上海)信息技术有限公司 基于用户特征标签的推荐方法以及装置
CN108880945B (zh) * 2018-08-02 2020-07-28 浙江口碑网络技术有限公司 一种云监控系统及方法
CN109299979A (zh) * 2018-09-17 2019-02-01 口碑(上海)信息技术有限公司 数据预估方法、装置、计算机设备及计算机可读存储介质
CN110033156B (zh) * 2018-12-14 2023-05-09 创新先进技术有限公司 一种业务活动效果的确定方法及装置
CN110458609A (zh) * 2019-07-23 2019-11-15 安徽迪科数金科技有限公司 一种新店经营管理预测分析系统
CN110796474A (zh) * 2019-09-18 2020-02-14 北京你财富计算机科技有限公司 一种活动运营自动调整方法、装置及电子设备
CN111415193B (zh) * 2020-01-12 2023-09-29 杭州览众数据科技有限公司 基于关联商品的门店相似度的品类销售提升度计算方法
CN111724037B (zh) * 2020-05-21 2024-03-05 口碑(上海)信息技术有限公司 运营资源分配方法、装置、计算机设备及可读存储介质
CN114862472A (zh) * 2022-05-19 2022-08-05 上海钧正网络科技有限公司 一种文案生成与投放方法、装置、电子设备以及介质
CN114971747A (zh) * 2022-07-14 2022-08-30 广州卓铸网络科技有限公司 一种基于大数据商品精准营销的数据分析方法及系统
CN117853146A (zh) * 2024-01-04 2024-04-09 江苏海獭电商科技有限公司 一种基于大数据的电商运营数据管理系统及方法
CN117593044B (zh) * 2024-01-18 2024-05-31 青岛网信信息科技有限公司 一种双角度的营销活动效果预测方法、介质及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102073960A (zh) * 2010-09-15 2011-05-25 江苏仕德伟网络科技股份有限公司 一种网站营销过程中运行效果的评估方法
CN102136119A (zh) * 2010-11-25 2011-07-27 中国移动（深圳）有限公司 营销服务效果评估方法和系统
CN107038190A (zh) * 2016-10-28 2017-08-11 厦门大学 一种应用于淘宝网的智能促销方案建模方法
CN107886369A (zh) * 2017-11-29 2018-04-06 口碑(上海)信息技术有限公司 一种用于营销活动效果预测的方法、装置及电子设备

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8265989B2 (en) * 2009-05-05 2012-09-11 The Nielsen Company, LLC Methods and apparatus to determine effects of promotional activity on sales

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102073960A (zh) * 2010-09-15 2011-05-25 江苏仕德伟网络科技股份有限公司 一种网站营销过程中运行效果的评估方法
CN102136119A (zh) * 2010-11-25 2011-07-27 中国移动（深圳）有限公司 营销服务效果评估方法和系统
CN107038190A (zh) * 2016-10-28 2017-08-11 厦门大学 一种应用于淘宝网的智能促销方案建模方法
CN107886369A (zh) * 2017-11-29 2018-04-06 口碑(上海)信息技术有限公司 一种用于营销活动效果预测的方法、装置及电子设备

Also Published As

Publication number Publication date
CN107886369A (zh) 2018-04-06

Similar Documents

Publication Publication Date Title
WO2019105226A1 (zh) 一种用于营销活动效果预测的方法、装置及电子设备
US8620720B2 (en) Embedding calendar knowledge in event-driven inventory forecasting
US20110238550A1 (en) Systems and methods for predicting financial behaviors
US20130151423A1 (en) Valuation of data
JP6727356B2 (ja) 特許評価システム
US20150332292A1 (en) System and method for monitoring market information for deregulated utilities based on transaction data
CN108171530B (zh) 一种用于客单价和复购率的提升方法及装置
WO2020221022A1 (zh) 业务对象推荐方法
KR102097045B1 (ko) 사용자의 특성을 반영하여 상품을 추천하는 방법 및 장치
CN107886241A (zh) 资源分析方法、装置、介质和电子设备
US11068798B2 (en) Systems and methods for short identifier behavioral analytics
Hummel et al. Loss functions for predicted click‐through rates in auctions for online advertising
CN111524001A (zh) 用户授信额度的预测方法、装置和相关设备
WO2015096742A1 (zh) 一种信息处理方法、装置和系统
JP5084968B1 (ja) 市場リスク予測装置、市場リスク予測方法及び市場リスク予測プログラム
US20170068962A1 (en) Forecasting customer channel choice using cross-channel loyalty
McCarthy et al. V (CLV): Examining variance in models of customer lifetime value
JP2016012223A (ja) 需要予測装置およびプログラム
WO2014020299A1 (en) Location evaluation
US20170178164A1 (en) Systems and Methods for Use in Processing Transaction Data
CN111464337A (zh) 资源配置方法、装置、电子设备
US20130073439A1 (en) Methods and systems for patent valuation using financial ratios
US20130282433A1 (en) Methods and apparatus to manage marketing forecasting activity
Ceritoğlu Self-insurance and consumption risk-sharing between birth-year cohorts in Turkey
Lhoest-Snoeck et al. Customer value modelling in the energy market and a practical application for marketing decision making

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18882350

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 31/08/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18882350

Country of ref document: EP

Kind code of ref document: A1