WO2019072107A1 - 消费能力预测 - Google Patents

消费能力预测 Download PDF

Info

Publication number
WO2019072107A1
WO2019072107A1 PCT/CN2018/108340 CN2018108340W WO2019072107A1 WO 2019072107 A1 WO2019072107 A1 WO 2019072107A1 CN 2018108340 W CN2018108340 W CN 2018108340W WO 2019072107 A1 WO2019072107 A1 WO 2019072107A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature data
time series
neural network
target object
user
Prior art date
Application number
PCT/CN2018/108340
Other languages
English (en)
French (fr)
Inventor
徐俊
李尚强
翟艺涛
王子伟
Original Assignee
北京三快在线科技有限公司
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 北京三快在线科技有限公司 filed Critical 北京三快在线科技有限公司
Priority to US16/755,880 priority Critical patent/US20200285937A1/en
Publication of WO2019072107A1 publication Critical patent/WO2019072107A1/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
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q30/0201—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
    • G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
    • 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
    • G06Q30/0224—Discounts or incentives, e.g. coupons or rebates based on user history
    • 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
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements

Definitions

  • the present invention relates to the field of computer technologies, and in particular, to a consumer capability prediction method, apparatus, electronic device, and readable storage medium.
  • Coupons need to determine the user's spending power based on the user's consumption history.
  • the first one determine the spending power based on the price of the user's last purchase; the second: randomly select the price of the user to purchase the product at a time to determine its spending power.
  • the user's last and certain spending power is related to his specific consumption scenario, and may be purchased for a relatively high price for some reasons, and is determined to be a high spending power. user.
  • the user's purchase price of goods in a few years may increase year by year or decrease year by year, and the average value can only reflect a whole result. Therefore, the user's spending power is determined by the price of the user's last purchase of the product, the price of a random purchase of the product, or the average price of the historically purchased product, and the accuracy is low.
  • the present invention has been made in order to provide a consumption capability prediction method, apparatus, electronic device and readable storage medium that overcome the above problems or at least partially solve the above problems.
  • a method for predicting consumption power comprising:
  • a consumption capability prediction apparatus comprising:
  • a first data obtaining module configured to acquire one or more statistical feature data and one or more time series feature data for the target object from the historical data of the target user;
  • a consumption capability determining module configured to determine, according to the one or more statistical feature data and the one or more time series feature data, a consumption of the target user by the target by using a preset hybrid neural network prediction model ability.
  • an electronic device comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor executing the computer program
  • the consumption capability prediction method disclosed in the embodiment of the present invention is implemented.
  • a readable storage medium having stored thereon a computer program, the computer program being executed by a processor to implement the steps of the consumption capability prediction method disclosed in the embodiments of the present invention.
  • the consumption capability prediction method disclosed in the embodiment of the present invention acquires one or more statistical feature data and one or more time series feature data for a target object from historical data of the target user, and one or more feature data based on the statistics. And the one or more time series feature data, using a preset hybrid neural network prediction model to determine the consumption power of the target user for the target object.
  • the problem of determining the user's consumption ability by using the price of the user's last purchase of the product, the price of the random purchase of the commodity, or the average price of the historically purchased commodity is determined in the prior art, and the accuracy is low. Based on the time series feature data, the feature dimension extraction of historical data can be realized, which makes the consumption ability predicted by the hybrid neural network prediction model more accurate.
  • FIG. 1 is a flowchart of a consumption capacity prediction method according to Embodiment 1 of the present invention.
  • FIG. 2 is a flowchart of a consumption capacity prediction method according to Embodiment 2 of the present invention.
  • FIG. 3 is a schematic diagram showing a hybrid neural network prediction model of the present invention.
  • FIG. 4 is a specific flowchart of step 202 of Embodiment 2 of the present invention.
  • FIG. 5 is a flow chart showing the consumption capacity prediction of the present invention.
  • FIG. 6 is a structural block diagram of a consumption capability prediction apparatus according to Embodiment 3 of the present invention.
  • FIG. 7 is a structural block diagram of a consumption capability prediction apparatus according to Embodiment 4 of the present invention.
  • Fig. 8 is a block diagram showing the hardware configuration of a consumption capability predicting device of the present invention.
  • FIG. 1 a flowchart of a method for predicting consumption capability according to Embodiment 1 of the present invention is shown, which may specifically include the following steps:
  • Step 101 Acquire one or more statistical feature data and one or more time series feature data for the target object from historical data of the target user.
  • one or more statistical feature data and one or more time series feature data for the target object are first obtained from the historical data of the target user.
  • the statistical feature data includes one or any combination of the following data: a historical consumer price parameter of the target object in an arbitrary time period, a historical browsing price parameter of the target object in an arbitrary time period, and a historical consumption of the non-target object in an arbitrary time period.
  • the target object may be a hotel, a KTV, a movie ticket, a restaurant, etc., and the embodiment of the present invention may be selected according to actual needs.
  • the embodiment of the present invention does not limit the one-time, one-month, three-month, half-year, one-year, and the like.
  • the parameters can be selected as the average value, the maximum value, the minimum value, the variance, the median, etc.
  • the historical consumer price parameters include the average value, the maximum value, the minimum value, the variance, the median of the historical consumer price, and the historical browsing price parameters including the history browsing.
  • the average value, the maximum value, the minimum value, the variance, and the median of the price are not limited in the embodiment of the present invention.
  • the non-target object is a collection of objects other than the hotel, such as a collection of objects such as KTV, movie tickets, restaurants, etc.; when the target object is KTV, the non-target object is an object other than KTV. Collections, such as collections of hotels, movie tickets, restaurants, etc.
  • the time series sequence feature data includes one or more of the following sequence: a historical consumer price parameter sequence of the target object within the set time period, a historical browsing price parameter sequence of the target object within the set time period, and a set time period A historical consumer price parameter sequence of a non-target object, and a historical browsing price parameter sequence of a non-target object within a set time period.
  • Parameters can be selected as mean, maximum, minimum, variance, median, and so on. The embodiments of the present application do not limit them.
  • each set of data is the average historical consumer price of the hotel in the current month
  • the first set of data is the average historical consumer price of the hotel last month
  • the second set of data is the history of the hotel for the previous two months.
  • the average consumer price, and so on, the 24th group of data is the average historical consumer price of the hotel for the first 24 months.
  • the fifth group of data does not exist, it can be calculated from the average of the fourth group data and the sixth group data.
  • Step 102 Determine, according to the one or more statistical feature data and the one or more time series feature data, a consumption power of the target user for the target object by using a preset hybrid neural network prediction model.
  • the target user's consumption ability for the target object can be obtained.
  • consumption power can be understood as the predicted consumer price.
  • the coupon for the target object that matches the spending power may also be sent to the target user.
  • a coupon of a different amount may be set for different spending powers. After determining the target user's spending power for the target object, the coupon for the target object matching the spending power is searched for, and the coupon is sent to the target user.
  • the coupon amount when the target object is a hotel, when the spending power is between 100 yuan and 199 yuan, the coupon amount can be reduced to 5 yuan for 100 yuan, and when the spending power is between 200 yuan and 399 yuan, the coupon amount can be If the spending power is between 400 yuan and 799 yuan, the coupon amount can be 400 yuan or 50 yuan. In this way, when the target user's consumption capacity for the hotel predicted by the hybrid neural network prediction model is 240 yuan, a coupon of the amount of 200 yuan minus 10 yuan can be sent to the target user.
  • the spending power is 240 yuan
  • the feature extraction of the time series dimension of the historical data can be realized, so that the consumption power predicted by the hybrid neural network prediction model is predicted. More accurate, which in turn makes the matching coupons more accurate.
  • an advertisement for the target object that matches the consumption capability may also be delivered to the target user.
  • the target-targeted advertisement matching the spending power is delivered to the target user.
  • the accuracy and effectiveness of the delivery of the advertisement can be further improved.
  • the consumption capability prediction method disclosed in the embodiment of the present invention acquires statistical feature data and time series sequence feature data for the target object from the historical data of the target user, and uses the preset based on the statistical feature data and the time series feature data.
  • the hybrid neural network prediction model determines the consumption power of the target user for the target object.
  • the problem of determining the user's consumption ability by using the price of the user's last purchase of the product, the price of the random purchase of the commodity, or the average price of the historically purchased commodity is determined in the prior art, and the accuracy is low.
  • the feature dimension extraction of historical data can be realized, which makes the consumption ability predicted by the hybrid neural network prediction model more accurate.
  • FIG. 2 a flowchart of a method for predicting consumption capability according to Embodiment 2 of the present invention is shown, which may specifically include the following steps:
  • Step 201 Acquire one or more statistical feature data, one or more time series feature data, and an actual consumption price for the target object from historical data of the sample user.
  • the user who has consumed the target object can be used as a sample user.
  • one or more statistical feature data of the sample user one or more time series feature data, and an actual consumer price are obtained.
  • the actual consumer price is the actual cost of the target object for the date specified by the sample user.
  • Step 202 Train the hybrid neural network prediction model according to one or more statistical feature data of the sample user, one or more time series feature data, and the actual consumption price.
  • the hybrid neural network prediction model includes a cyclic neural network and a traditional neural network.
  • FIG. 3 a schematic diagram of a hybrid neural network prediction model of the present invention is shown.
  • X 1 , X 2 , ..., X n-1 , X n represent the feature data of the input sample user.
  • n is a positive integer greater than or equal to 2.
  • the part of the feature data is one or more statistical feature data of the sample user, and is represented by a specific numerical value.
  • X 1 may be the historical consumer price average for the target user in the most recent week
  • X 2 may be the historical browsing price average for the target user in the most recent week.
  • the feature data may also be the historical consumer price average of the non-target objects in the most recent week, and the historical browsing price average of the non-target objects in the most recent week.
  • statistical characteristic data such as maximum value, minimum value, variance, and median can be used.
  • the feature data may also be statistical feature data such as a user level, a user active state, and a user resident address.
  • Another portion of the feature data is one or more time series feature data of the sample user.
  • X n-1 [s 1 ,s 2 ,...,s 24 ], where s 1 to s 24 are the average historical consumer price values for the target object for each month for 24 months
  • X n [t 1 , t 2 , . . . , t 24 ], where t 1 to t 24 are the historical consumer price averages for non-target objects for each month for 24 months, respectively.
  • the feature data input by the hybrid neural network prediction model may be two, that is, one statistical feature data and one time series feature data.
  • the parameters of the hybrid neural network prediction model are less, and the deviation is larger when the consumption power is compared with the actual consumption price. Entering as much feature data as possible may result in data redundancy, computational complexity, and no improvement in prediction results.
  • 40 to 50 feature data can be selected, the computational amount of the model is relatively small, and the prediction result is relatively accurate.
  • One or more statistical feature data of one sample user, one or more pieces of time series feature data, and an actual consumption price may constitute a group of training data, and by training the plurality of sets of training data, a hybrid neural network of the target object is obtained. Forecast model.
  • the hybrid neural network prediction model may include a cyclic neural network and a traditional neural network.
  • the cyclic neural network is used to process the time series feature data of the sample user, and the distribution characteristics of the historical price of the sample user are learned according to the time series feature data, and the time series feature data is calculated to obtain the time series feature data and transmitted to the traditional neural network.
  • the traditional neural network can be a fully connected Deep Neural Network (DNN), which processes the time series feature data of the sample user and the statistical feature data of the sample user through the traditional neural network.
  • DNN Deep Neural Network
  • Step 202 can include sub-step 2021, sub-step 2022, and sub-step 2023.
  • Sub-step 2021 inputting each time series feature data of the sample user into the cyclic neural network to obtain corresponding time series feature data.
  • Each time series feature data of the sample user For each time series feature data of the sample user, it needs to be processed by the cyclic neural network. Each time series feature data of the sample user is input into the cyclic neural network to obtain time series feature data of the sample user.
  • a time series feature data of the sample user includes L sub-feature data arranged in time series
  • the first sub-feature data is input to the cyclic neural network for the time series feature data of the sample user to obtain the first sub-feature Outputting result of the data; inputting the output result of the mth sub-feature data and the m-1th sub-feature data into the cyclic neural network until all the L sub-feature data in the time series characteristic data of the sample user are all input Obtaining corresponding timing feature data; wherein m is a positive integer greater than 1, and less than or equal to L. If there is a plurality of time series feature data in the hybrid neural network prediction model, the other time series feature data is processed according to the same method to obtain corresponding time series feature data.
  • the neural network 10 cycles, M n-1 wherein a time sequence is a timing feature data corresponding to the data Xn-1, M n is the time sequence X n feature data corresponding to a timing feature data.
  • the first sub-feature data s 1 is input to the cyclic neural network, and the output result y 1 of the first sub-feature data is obtained.
  • y 1 f(U 1 s 1 ), where f denotes an activation function of the cyclic neural network, and U 1 is a weight value of the first sub-feature data s 1 .
  • the weight value of s 24 , W 24 is the weight value of the output result y 23 , and therefore, the time series feature data of the sample user is correlated with each of the sub-feature data of the time series feature data.
  • Sub-step 2022 inputting one or more statistical feature data of the sample user and the one or more time series feature data into the traditional neural network to obtain a predicted consumption capability of the sample user.
  • the statistical feature data and the time series feature data obtained through the time series feature data are input into the traditional neural network, and the consumption power predicted by the sample user can be obtained.
  • the statistical feature data of the sample user is X 1 and X 2
  • the time series feature data of the sample user is M n-1 and M n
  • the statistical feature data of the sample user is X 1 .
  • the sample user's time series feature data M n-1 , M n is input into the traditional neural network.
  • the conventional neural network can be divided into an input layer 21, a hidden layer 22, and an output layer 23, and the values of the hidden layers H 1 , H 2 , and H 3 are respectively obtained by the formulas (1)-(3):
  • H 1 g(a 1 X 1 +a 2 X 2 +a 3 M n-1 +a 4 M n ) (1)
  • H 2 g(b 1 X 1 +b 2 X 2 +b 3 M n-1 +b 4 M n ) (2)
  • H 3 g (c 1 X 1 + c 2 X 2 + c 3 M n-1 + c 4 M n ) (3).
  • Equation (1) in a 1 indicates for the feature data X 1 in the input layer to the hidden layer H weighting value 1
  • a 2 indicates for the feature data X 2 input layer to the hidden layer H weighting value 1
  • a 3 indicates for the feature data M n-1 input layer to the hidden layer weight value of 1 H
  • a 4 M n represents the data characteristic for the input layer to the hidden layer
  • H is the weight value 1.
  • Formula B (2) 1 represents for the feature data X 1 in the input layer to the hidden layer H weighting value 2
  • b 2 indicates for the feature data X 2 input layer to the hidden layer H weighting value 2
  • b 3 indicates for the feature data M n-1 input layer to the hidden layer H weighting value 2
  • b 4 represent for the feature data M n input layer to the hidden layer H weighting value 2
  • equation (3) c 1 represents for the feature data X 1 input layer the hidden layer H weight values 3
  • c 2 denotes a weight value for the feature data X 2 input layer to the hidden layer H weight of 3
  • c 3 indicates for the feature data M n-1 input layer to the hidden layer H weight values
  • c 4 represents a characteristic data M n for the input layer to the hidden layer 3 H weight values.
  • the value of the output layer Z is obtained by the formula (4).
  • d 1 represents the weight value of the hidden layer H 1 to the output layer Z
  • d 2 represents the weight value of the hidden layer H 2 to the output layer Z
  • d 3 represents the hidden layer H 3 to the output The weight value of layer Z.
  • Output layer Z represents the consumption power predicted by the sample user.
  • the hidden layer 22 in FIG. 3 is at least one layer, the specific layer number of the hidden layer, the activation function f of the cyclic neural network, and the activation function g of the traditional neural network all pass the statistical feature data and time series of the sample user. The feature data is determined along with the actual consumer price for the target object.
  • Sub-step 2023 correcting each weight value in the hybrid neural network prediction model according to a deviation between the predicted consumption ability of the sample user and the corresponding actual consumption price, until the deviation is less than a set threshold.
  • the weight values in the cyclic neural network and the conventional neural network can be set to arbitrary values when the consumption power of the sample user is first predicted. Then, the consumption power predicted by the sample user is subtracted from the actual consumption price of the sample user for the target object, and the difference between the two is obtained.
  • the weight values in the hybrid neural network prediction model are corrected according to the magnitude of the difference, that is, the weight values in the cyclic neural network and the traditional neural network are corrected. After continuous correction, the predicted consumption power can be made more accurate until the deviation between the predicted consumption power of the sample user and the actual consumption price is less than the set threshold. After the training is completed, the weights of the hybrid neural network prediction model are worthy of determination. Further, the predicted spending power of multiple users can be obtained through steps 2021 and 2022. Then, the deviation between the predicted consumption power of the plurality of users and the actual consumption price thereof is determined, and the respective weight values in the hybrid neural network prediction model are corrected. The resulting weight values are more accurate.
  • the hybrid neural network prediction model can only predict the consumption ability for a certain target object.
  • the statistical feature data and the time series feature data may be different.
  • different statistical feature data may be used; or the same timing feature data may be used, but the time period setting in the time series feature data is different.
  • the statistical characteristic data may include: the average historical consumer price of the hotel in the past week, the average historical consumer price of the target other than the hotel in the most recent week, and the like.
  • the statistical feature data may include: the median historical consumer price of the KTV in the last week, the median historical consumer price of the target object other than the KTV in the most recent week, and the like.
  • the time series feature data may include: an average historical price of the hotel in the last 24 months.
  • the target object is KTV
  • the time series feature data may include: KTV in the last 24 weeks. The average value of historical consumer prices.
  • the price parameter of the non-target object affects the price parameter of the target object to a certain extent
  • the price parameter of the non-target object can be used as a feature data.
  • Step 203 Acquire one or more statistical feature data and one or more time series feature data for the target object from the historical data of the target user according to the feature data extraction rule of the corresponding target object.
  • the feature data extraction rule of the target object is first determined, that is, the respective feature data of the use of the hybrid neural network prediction model is determined.
  • the price parameter of the target object can be used as one feature data
  • the price parameter of the non-target object can be used as another feature data.
  • the statistical feature data and the time series feature data for the two feature data of the target object are then acquired from the historical data of the target user.
  • the statistical feature data may include: a historical consumer price parameter of the target object, a historical browsing price parameter of the target object, a historical consumer price parameter of the non-target object, a historical browsing price parameter of the non-target object, and a user rating.
  • the historical consumer price parameter of the target object may include an average value, a maximum value, a minimum value, and the like of the historical consumer price of the target object.
  • the historical browsing price parameter of the target object may include an average value, a maximum value, a minimum value, and the like of the historical browsing price of the target object.
  • the historical consumer price parameter of the non-target object may include an average value, a maximum value, a minimum value, and the like of the historical consumer price of the non-target object.
  • the historical browsing price parameter of the non-target object is to include an average value, a maximum value, a minimum value, and the like of the historical browsing price of the non-target object.
  • Features such as user ratings may include user ratings, user active states, user resident addresses, and the like.
  • the time series feature data may include: an average sequence of historical consumption prices of the target object, a historical browsing price average sequence of the target object, a historical consumer price average sequence of the non-target object, a historical browsing price average sequence of the non-target object, and the like. .
  • Step 204 Determine, according to the one or more time series feature data of the target user, the corresponding one or more time series feature data by using the cyclic neural network of the hybrid neural network prediction model.
  • the time series feature data of the four target users are respectively input into the cyclic neural network, and the four corresponding time series feature data are determined by the cyclic neural network.
  • Step 205 Determine, according to the one or more statistical feature data of the target user and the corresponding one or more time series feature data, the traditional neural network that uses the hybrid neural network prediction model to determine the target user Describe the consumption power of the target audience.
  • a plurality of statistical feature data and four time series feature data of the target user are input into the traditional neural network, and the traditional user is used to determine the consumption ability of the target user for the target object.
  • the average price of the goods purchased according to the user's history is used to determine the spending power, with an error of 40 yuan.
  • a general machine learning model such as Linear Regression (LR) or Gradient Boosting Decision Tree (GBDT), has an error of around 33 yuan.
  • LR Linear Regression
  • GBDT Gradient Boosting Decision Tree
  • the final prediction error is about 30 yuan, and the predicted consumption power is more accurate.
  • the consumption capability prediction method disclosed in the embodiment of the present invention acquires statistical feature data, time series sequence feature data, and actual consumption price for the target object from the historical data of the sample user, according to the sample user
  • the hybrid feature data, the time series feature data, and the actual consumer price training for the target object obtain the hybrid neural network prediction model.
  • the statistical feature data and the time series feature data for the target object are obtained from the historical data of the target user according to the feature data extraction rule of the corresponding target object.
  • the cyclic neural network is used to determine the time series feature data of the target user.
  • the cyclic neural network can realize the feature extraction of the time series dimension of the historical data, which makes the consumption ability predicted by the hybrid neural network prediction model more accurate.
  • FIG. 6 a structural block diagram of a consumption capability prediction apparatus according to Embodiment 3 of the present invention is shown.
  • the first data obtaining module 501 is configured to acquire one or more statistical feature data and one or more time series feature data for the target object from the historical data of the target user.
  • the consumption capability determining module 502 is configured to determine, by using a preset hybrid neural network prediction model, the target user for the target object based on the one or more statistical feature data and the one or more time series feature data. Spending power.
  • the consumption capability prediction apparatus disclosed in the embodiment of the present invention acquires one or more statistical feature data and one or more time series feature data for the target object from the historical data of the target user, based on the one or more statistical feature data. And the one or more time series feature data, using a preset hybrid neural network prediction model to determine the consumption power of the target user for the target object.
  • the problem of determining the user's consumption ability by using the price of the user's last purchase of the product, the price of the random purchase of the commodity, or the average price of the historically purchased commodity is determined in the prior art, and the accuracy is low. Based on the time series feature data, the feature dimension extraction of historical data can be realized, which makes the consumption ability predicted by the hybrid neural network prediction model more accurate.
  • FIG. 7 a structural block diagram of a consumption capability prediction apparatus according to Embodiment 4 of the present invention is shown.
  • the consumption capability prediction apparatus further includes:
  • a second data obtaining module 503 configured to acquire, from the historical data of the sample user, one or more statistical feature data, one or more time series feature data, and an actual consumption price for the target object;
  • a model training module 504 configured to train the hybrid neural network prediction model according to one or more statistical feature data of the sample user, one or more time series feature data, and the actual consumption price; wherein the mixed neural Network prediction models include cyclic neural networks and traditional neural networks.
  • the model training module 504 includes:
  • the timing feature data generating sub-module 5041 is configured to input each time series feature data of the sample user into the cyclic neural network to obtain corresponding time series feature data;
  • the consumption capability generation sub-module 5042 is configured to input one or more statistical feature data of the sample user and the one or more time series feature data into the traditional neural network to obtain a predicted consumption capability of the sample user;
  • the weight value correction sub-module 5043 is configured to correct each weight value in the hybrid neural network prediction model according to a deviation between the predicted consumption capability of the sample user and the actual consumption price, until the deviation is less than a setting Threshold.
  • the time-series feature data generating sub-module 5041 includes:
  • the first output result generating unit 50411 is configured to input the first sub-feature data into the circulating neural network to obtain an output result of the first sub-feature data;
  • the timing feature data generating unit 50412 is configured to input the output result of the mth sub-feature data and the m-1th sub-feature data into the cyclic neural network until all the L sub-feature data are input, and obtain the corresponding Time series feature data.
  • m is a positive integer greater than one and less than or equal to L.
  • the consumption capability determining module 502 includes:
  • a timing feature data determining sub-module 5021 configured to determine one or more timing characteristics of the target user by using a cyclic neural network in the hybrid neural network prediction model based on one or more time series feature data of the target user data;
  • the consumption capability determination sub-module 5022 is configured to determine, by using the one or more statistical feature data of the target user and one or more time-series feature data of the target user, by using a traditional neural network in the hybrid neural network prediction model The target user's spending power for the target object.
  • the first data obtaining module 501 includes:
  • the first data acquisition sub-module 5011 is configured to acquire one or more statistical feature data and one or more time series sequences for the target object from the historical data of the target user according to the feature data extraction rule of the corresponding target object. Feature data.
  • the consumption capability prediction device further includes:
  • a issuing module 505 configured to send a coupon for the target object that matches the consumption capability to the target user; and/or to deliver an advertisement for the target object that matches the consumption capability To the target user.
  • the consumption capability prediction apparatus disclosed in the embodiment of the present invention acquires statistical feature data, time series sequence feature data, and actual consumption price for the target object from the historical data of the sample user, according to the sample user
  • the statistical feature data, the time series feature data, and the actual consumer price training for the target object obtain the hybrid neural network prediction model, and obtain the target data from the historical data of the target user according to the feature data extraction rule of the corresponding target object.
  • Generating statistical feature data and time series sequence feature data of the target object determining, based on the time series sequence feature data of the target user, time series feature data of the target user by using a cyclic neural network, based on the statistical feature data of the target user and the The time-series feature data of the target user determines the consumption ability of the target user for the target object by using a traditional neural network.
  • the cyclic neural network can realize the feature extraction of the time series dimension of the historical data, which makes the consumption ability predicted by the hybrid neural network prediction model more accurate.
  • the present invention also discloses an electronic device, see FIG. 8, comprising a memory 820, a processor 810, and a computer program 900 stored on the memory 820 and operable on the processor, wherein When the processor 810 executes the computer program 900, the consumption capability prediction method according to the first embodiment and the second embodiment of the present invention is implemented.
  • the device may also include a bus 830 and an external interface 840.
  • the processor 810 and the memory 820 are interconnected by a bus 830 and can also communicate with other devices or components via the external interface 840.
  • the invention also discloses a readable storage medium on which a computer program is stored, and when the computer program is executed by the processor, the steps of the consumption capability prediction method according to the first embodiment and the second embodiment of the present invention are implemented.
  • the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.
  • modules in the devices of the embodiments can be adaptively changed and placed in one or more devices different from the embodiment.
  • the modules or units or components of the embodiments may be combined into one module or unit or component, and further they may be divided into a plurality of sub-modules or sub-units or sub-components.
  • any combination of the features disclosed in the specification, including the accompanying claims, the abstract and the drawings, and any methods so disclosed, or All processes or units of the device are combined.
  • Each feature disclosed in this specification (including the accompanying claims, the abstract and the drawings) may be replaced by alternative features that provide the same, equivalent or similar purpose.
  • the various component embodiments of the present invention may be implemented in hardware, or in a software module running on one or more processors, or in a combination thereof.
  • a microprocessor or digital signal processor may be used in practice to implement some or all of the functionality of some or all of the components of the consumption capability prediction device in accordance with embodiments of the present invention.
  • the invention can also be implemented as a device or device program (e.g., a computer program and a computer program product) for performing some or all of the methods described herein.
  • a program implementing the invention may be stored on a computer readable medium or may be in the form of one or more signals. Such signals may be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Abstract

本发明实施例提供了一种消费能力预测方法、装置、电子设备及可读存储介质，涉及计算机技术领域。根据所述方法的一个示例，通过从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据，可基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。

Description

消费能力预测
相关申请的交叉引用
本专利申请要求于2017年10月11日提交的、申请号为2017109433882、发明名称为“一种消费能力预测方法、装置、电子设备及可读存储介质”的中国专利申请的优先权，该申请的全文以引用的方式并入本文中。
技术领域
本发明涉及计算机技术领域，特别是涉及一种消费能力预测方法、装置、电子设备及可读存储介质。
背景技术
使用优惠券的促销商业模式已变得普及，使用优惠券使得用户在购买商品时能够获得商品价格的折扣和/或在购买商品时获得附加服务，为了能够定向的给指定消费能力的用户群体发放优惠券，需要根据用户的消费历史来确定用户的消费能力。
目前，通常有以下三种方式确定用户的消费能力，第一种：根据用户最近一次购买商品的价格来确定其消费能力；第二种：随机选取用户某一次购买商品的价格来确定其消费能力；第三种：根据用户历史购买商品的价格平均值来确定其消费能力。
但对于第一种方式和第二种方式来说，用户最近一次、某一次的消费能力与其具体的消费场景相关，可能由于一些原因购买了价格比较高的商品，而被确定为高消费能力的用户。对于第三种方式，用户在几年内的商品购买价格可能是一个逐年上升或者逐年下降的，而平均值只能反映一个整体的结果。因此，利用用户最近一次购买商品的价格、随机某一次购买商品的价格，或者历史购买商品的价格平均值来确定用户的消费能力，其准确度较低。
发明内容
鉴于上述问题，提出了本发明以便提供一种克服上述问题或者至少部分地解决上述问题的消费能力预测方法、装置、电子设备及可读存储介质。
根据本发明的一个方面，提供了一种消费能力预测方法，包括：
从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据；
基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。
根据本发明的另一方面，提供了一种消费能力预测装置，包括：
第一数据获取模块，用于从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据；
消费能力确定模块，用于基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。
根据本发明的再一方面，提供了一种电子设备，包括存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现本发明实施例公开的所述的消费能力预测方法。
根据本发明的再一方面，提供了一种可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现本发明实施例公开的所述消费能力预测方法的步骤。
本发明实施例公开的消费能力预测方法，从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据，基于所述统计一个或多个特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。解决了现有技术中利用用户最近一次购买商品的价格、随机某一次购买商品的价格，或者历史购买商品的价格平均值来确定用户的消费能力，其准确度较低的问题，在统计特征数据的基础上结合时序序列特征数据，可实现对历史数据进行时序维度的特征提取，使得利用混合神经网络预测模型预测的消费能力更准确。
上述说明仅是本发明技术方案的概述，为了能够更清楚了解本发明的技术手段，而可依照说明书的内容予以实施，并且为了让本发明的上述和其它目的、特征和优点能够更明显易懂，以下特举本发明的具体实施方式。
附图说明
通过阅读下文优选实施方式的详细描述，各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的，而并不认为是对本发明的限制。而且在整个附图中，用相同的参考符号表示相同的部件。
图1示出了本发明实施例一的一种消费能力预测方法的流程图；
图2示出了本发明实施例二的一种消费能力预测方法的流程图；
图3示出了本发明的混合神经网络预测模型的示意图；
图4示出了本发明实施例二步骤202的具体流程图；
图5示出了本发明的消费能力预测的流程示意图；
图6示出了本发明实施例三的一种消费能力预测装置的结构框图；
图7示出了本发明实施例四的一种消费能力预测装置的结构框图；
图8示出了示出了本发明的一种消费能力预测装置的硬件结构图。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例，然而应当理解，可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反，提供这些实施例是为了能够更透彻地理解本公开，并且能够将本公开的范围完整的传达给本领域的技术人员。
实施例一
参照图1，示出了本发明实施例一的一种消费能力预测方法的流程图，具体可以包括如下步骤：
步骤101，从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据。
针对需要预测其消费能力的目标用户，首先要从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据。
统计特征数据包括以下数据中的一个或任意组合：任意时间段内的目标对象的 历史消费价格参数、任意时间段内的目标对象的历史浏览价格参数、任意时间段内的非目标对象的历史消费价格参数、任意时间段内的非目标对象的历史浏览价格参数、用户等级、用户活跃状态、用户常住地址。
目标对象可以是酒店、KTV、电影票、饭店等，本发明实施例可以根据实际需求进行选择。任意时间段可以选取一周、一个月、三个月、半年、一年等，本发明实施例不对其加以限制。参数可以选取平均值、最大值、最小值、方差、中位数等，历史消费价格参数包括历史消费价格的平均值、最大值、最小值、方差、中位数，历史浏览价格参数包括历史浏览价格的平均值、最大值、最小值、方差、中位数，本发明实施例不对其加以限制。
当目标对象为酒店时，非目标对象为除酒店以外的其他对象的集合，如KTV、电影票、饭店等对象的集合；当目标对象为KTV时，非目标对象为除KTV以外的其他对象的集合，如酒店、电影票、饭店等对象的集合。
时序序列特征数据包括以下序列中的一个或多个：设定时间段内的目标对象的历史消费价格参数序列、设定时间段内的目标对象的历史浏览价格参数序列、设定时间段内的非目标对象的历史消费价格参数序列、设定时间段内的非目标对象的历史浏览价格参数序列。参数可以选取平均值、最大值、最小值、方差、中位数等。本申请实施例不对其加以限制。
例如，当目标对象为酒店时，对于酒店的历史消费价格，可选取最近两年内的数据，且按月划分，设定时间段内的酒店的历史消费价格平均值序列为按照时间顺序排列的24组数据构成的序列，每一组数据为当前月份的酒店的历史消费价格平均值，第一组数据为上个月酒店的历史消费价格平均值，第二组数据为前两个月酒店的历史消费价格平均值，以此类推，第24组数据为前24个月酒店的历史消费价格平均值。当其中某一个月没有酒店的历史消费价格平均值时，通过相邻月份的酒店的历史消费价格平均值进行补全。例如，若第5组数据不存在，则可通过第4组数据和第6组数据的平均值计算得到。
步骤102，基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。
将目标用户的一个或多个统计特征数据和一个或多个时序序列特征数据，输入到预设的混合神经网络预测模型中，则可得到目标用户针对目标对象的消费能力。其中， 消费能力可理解为预测的消费价格。
在预测到目标用户针对目标对象的消费能力后，还可以将与所述消费能力匹配的针对所述目标对象的优惠券，发送至所述目标用户。
可以为不同消费能力设定不同金额的优惠券，在确定目标用户针对目标对象的消费能力后，查找与消费能力匹配的针对目标对象的优惠券，将优惠券发送给目标用户。
例如，当目标对象为酒店时，当消费能力在100元-199元之间，优惠券金额可为满100元减5元，当消费能力在200元-399元之间，优惠券金额可为满200元减10元，当消费能力在400元-799元之间，优惠券金额可为满400元减50元。这样，当利用混合神经网络预测模型预测得到的目标用户针对酒店的消费能力为240元时，可以将一张金额为满200元减10元的优惠券发送给目标用户。当然，当消费能力为240元时，也可以将一张满100元减5元的优惠券发送给目标用户，但满200元减10元的优惠券的优惠幅度更高，目标用户的实际消费的可能性更大。
本发明实施例中，在一个或多个统计特征数据的基础上结合一个或多个时序序列特征数据，可实现对历史数据进行时序维度的特征提取，使得利用混合神经网络预测模型预测的消费能力更准确，进而使得匹配到的优惠券更准确。
此外，在预测到目标用户针对目标对象的消费能力后，还可以将与所述消费能力匹配的针对所述目标对象的广告，投放给所述目标用户。
在确定目标用户针对目标对象的消费能力后，将与消费能力匹配的针对目标对象的广告，投放给目标用户。通过匹配目标用户的消费能力，可以进一步提高广告的投放的精准度和投放效果。
当然，在确定目标用户针对目标对象的消费能力后，除了优惠券的发送和广告数据的投放，还可以将其应用到其它的场景中，本发明实施例对此不作限制。
本发明实施例公开的消费能力预测方法，从目标用户的历史数据中获取针对目标对象的统计特征数据和时序序列特征数据，基于所述统计特征数据和所述时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。解决了现有技术中利用用户最近一次购买商品的价格、随机某一次购买商品的价格，或者历史购买商品的价格平均值来确定用户的消费能力，其准确度较低的问题，在统计特征数据的基础上结合时序序列特征数据，可实现对历史数据进行时序维度的特征提取，使得利用混合神经网络预测模型预测的消费能力更准确。
实施例二
参照图2，示出了本发明实施例二的一种消费能力预测方法的流程图，具体可以包括如下步骤：
步骤201，从样本用户的历史数据中，获取针对所述目标对象的一个或多个统计特征数据、一个或多个时序序列特征数据以及实际消费价格。
针对某一目标对象，可以将已消费该目标对象的用户作为样本用户。从样本用户的历史数据中，获取样本用户的一个或多个统计特征数据、一个或多个时序序列特征数据以及实际消费价格。实际消费价格为样本用户指定日期的为目标对象的实际花费。
步骤202，根据所述样本用户的一个或多个统计特征数据、一个或多个时序序列特征数据以及所述实际消费价格训练所述混合神经网络预测模型。其中，所述混合神经网络预测模型包括循环神经网络和传统神经网络。
参照图3，示出了本发明的混合神经网络预测模型的示意图。
图3中，X 1，X 2，…，X n-1，X n表示输入的样本用户的特征数据。n为大于等于2的正整数。其中，一部分特征数据为样本用户的一个或多个统计特征数据，用具体的数值来表示。例如，X 1可以是样本用户最近一周内针对目标对象的历史消费价格平均值，X 2可以是样本用户最近一周内针对目标对象的历史浏览价格平均值。所述特征数据还可以是最近一周内非目标对象的历史消费价格平均值，最近一周内非目标对象的历史浏览价格平均值。除了平均值以外，还可以使用最大值、最小值、方差、中位数等统计特征数据。此外，所述特征数据还可以是用户等级、用户活跃状态、用户常住地址等统计特征数据。另外一部分特征数据为样本用户的一个或多个时序序列特征数据。例如，X n-1＝[s 1，s 2，…，s 24]，其中，s 1至s 24分别为24个月以来每个月的针对目标对象的历史消费价格平均值，X n＝[t 1，t 2，…，t 24],其中，t 1至t 24分别为24个月以来每个月的针对非目标对象的历史消费价格平均值。
可以理解的是，所述混合神经网络预测模型输入的特征数据可以为2个，即1个统计特征数据和1个时序序列特征数据。但是，当输入的特征数据比较少时，使用混合神经网络预测模型的参数较少，最终得到消费能力和实际消费价格相比，偏差会比较大。而输入尽可能多的特征数据，可能会导致数据冗余，计算复杂，对于预测结果没有提升。在一个实施方式中，可以选择40到50个特征数据，模型的计算量相对较少，而预测结果相对准确。
一个样本用户的一个或多个统计特征数据一个或多个、时序序列特征数据以及实际消费价格可以构成一组训练数据，通过对多组的训练数据进行训练，得到所述目标对象的混合神经网络预测模型。
其中，所述混合神经网络预测模型可以包括循环神经网络和传统神经网络。利用循环神经网络处理样本用户的时序序列特征数据，根据时序序列特征数据学习样本用户的历史价格在时间上的分布特征，计算时序序列特征数据得到时序特征数据传给传统神经网络。传统神经网络可以为全连接的深度神经网络(Deep Neural Network,DNN)，通过传统神经网络处理样本用户的时序特征数据和样本用户的统计特征数据。
如图4所示，为本实施例的步骤202的具体流程图。步骤202可以包括子步骤2021、子步骤2022、子步骤2023。
子步骤2021，将所述样本用户的每个时序序列特征数据输入所述循环神经网络，获得对应的时序特征数据。
对于样本用户的每个时序序列特征数据，需要利用循环神经网络进行处理。将样本用户的每个时序序列特征数据输入循环神经网络中，获得该样本用户的时序特征数据。
当样本用户的一个时序序列特征数据包括按时序排列的L个子特征数据时，针对样本用户的该时序序列特征数据，将第一个子特征数据输入所述循环神经网络，获得第一个子特征数据的输出结果；将第m个子特征数据与第m-1个子特征数据的输出结果组合输入至所述循环神经网络，直至所述样本用户的时序序列特征数据中的L个子特征数据全部输入完成，获得对应的时序特征数据；其中，m为大于1、且小于或等于L的正整数。如果上述混合神经网络预测模型存在多个时序特征数据，则按照同样的方法，对其他时序序列特征数据进行处理，获得对应的时序特征数据。
如图3所示，10为循环神经网络，M n-1为时序序列特征数据Xn-1对应的时序特征数据，M n为时序序列特征数据X n对应的时序特征数据。
对于时序序列特征数据X n-1＝[s 1，s 2，…，s 24]，将第一个子特征数据s 1输入循环神经网络，获得第一个子特征数据的输出结果y 1，y 1＝f(U 1s 1)，其中，f表示循环神经网络的激活函数，U 1为第一个子特征数据s 1的权重值。然后将第二个子特征数据s 2与第一个子特征数据的输出结果y 1组合输入至循环神经网络，获得第二个子特征数据的输出结果y 2，y 2＝f(U 2s 2+W 2y 1)，U 2为第二个子特征数据s 2的权重值，W 2为第一个子特征数据的输 出结果y 1的权重值，以此类推，将第24个子特征数据s 24与第23个子特征数据的输出结果y 23组合输入至循环神经网络，获得对应的时序特征数据M n-1＝f(U 24s 24+W 24y 23)，U 24为第24个子特征数据s 24的权重值，W 24为输出结果y 23的权重值，因此，样本用户的时序特征数据与时序序列特征数据中的每一个子特征数据都相关。需要注意的是，本公开不限定循环神经网络的具体实现方式。例如，还可以使用改进的循环神经网络获得样本用户的时序特征数据。
子步骤2022，将所述样本用户的一个或多个统计特征数据与所述一个或多个时序特征数据输入所述传统神经网络，获得所述样本用户的预测消费能力。
对于一个样本用户，将统计特征数据与通过时序序列特征数据得到的时序特征数据输入传统神经网络中，则可得到所述样本用户预测的消费能力。
如图3所示，为简化说明传统神经网络，样本用户的统计特征数据为X 1和X 2，样本用户的时序特征数据为M n-1和M n，将样本用户的统计特征数据X 1、X 2，样本用户的时序特征数据M n-1、M n输入传统神经网络。一般，传统神经网络可划分为输入层21、隐藏层22和输出层23，则隐藏层H 1、H 2、H 3的值分别用公式(1)-(3)得到：
H 1＝g(a 1X 1+a 2X 2+a 3M n-1+a 4M n)    (1),
H 2＝g(b 1X 1+b 2X 2+b 3M n-1+b 4M n)    (2),
H 3＝g(c 1X 1+c 2X 2+c 3M n-1+c 4M n)    (3)。
其中，g表示传统神经网络的激活函数。公式(1)中的a 1表示对于特征数据X 1输入层到隐藏层H 1的权重值，a 2表示对于特征数据X 2输入层到隐藏层H 1的权重值，a 3表示对于特征数据M n-1输入层到隐藏层H 1的权重值，a 4表示对于特征数据M n输入层到隐藏层H 1的权重值。公式(2)中的b 1表示对于特征数据X 1输入层到隐藏层H 2的权重值，b 2表示对于特征数据X 2输入层到隐藏层H 2的权重值，b 3表示对于特征数据M n-1输入层到隐藏层H 2的权重值，b 4表示对于特征数据M n输入层到隐藏层H 2的权重值；公式(3)中的c 1表示对于特征数据X 1输入层到隐藏层H 3的权重值，c 2表示对于特征数据X 2输入层到隐藏层H 3的权重值，c 3表示对于特征数据M n-1输入层到隐藏层H 3的权重值，c 4表示对于特征数据M n输入层到隐藏层H 3的权重值。
输出层Z的值用公式(4)得到。
Z＝g(d 1H 1+d 2H 2+d 3H 3)    (4),
其中，g表示传统神经网络的激活函数，d 1表示隐藏层H 1到输出层Z的权重值，d 2表示隐藏层H 2到输出层Z的权重值，d 3表示隐藏层H 3到输出层Z的权重值。
输出层Z表示的是样本用户预测的消费能力。应该注意的是，图3中隐藏层22为至少一层，隐藏层的具体层数、循环神经网络的激活函数f、传统神经网络的激活函数g，均通过样本用户的统计特征数据、时序序列特征数据以及针对目标对象的实际消费价格确定。
子步骤2023，根据所述样本用户的预测消费能力与对应的实际消费价格之间的偏差，修正所述混合神经网络预测模型中的各个权重值，直至所述偏差小于设定阈值。
在初次预测样本用户的消费能力时，循环神经网络和传统神经网络中的各个权重值可以设置为任意值。然后，将样本用户预测得到的消费能力与样本用户针对目标对象的实际消费价格相减，得到两者之间的差值。根据差值的大小修正混合神经网络预测模型中的各个权重值，也就是修正循环神经网络和传统神经网络中的各个权重值。经过不断的修正，可以使得预测的消费能力更加准确，直至样本用户的预测消费能力与实际消费价格之间的偏差小于设定阈值。训练完成后，混合神经网络预测模型的各个权重值得以确定。更进一步的，可以通过步骤2021和2022，获得多个用户的预测消费能力。然后确定多个用户的预测消费能力和其实际的消费价格之间的偏差，对混合神经网络预测模型中的各个权重值进行修正。由此得出的权重值更加准确。
需要说明的是，得到混合神经网络预测模型只能预测针对某一种目标对象的消费能力。对于另一种目标对象，则统计特征数据和时序序列特征数据可以不同。例如，对于另一种目标对象，可以使用不同的统计特征数据；或者使用相同的时序特征数据，但是时序特征数据里面的时间段的设定不同。
例如，当目标对象为酒店时，统计特征数据可以包括：最近一周内酒店的历史消费价格平均值，最近一周内除酒店以外的其他目标对象的历史消费价格平均值等。当目标对象为KTV时，统计特征数据可以包括：最近一周内KTV的历史消费价格中值，最近一周内除KTV以外的其他目标对象的历史消费价格中值等。类似的，当目标对象为酒店时，时序序列特征数据可以包括：最近24个月内酒店的历史消费价格平均值，当目标对象为KTV时，时序序列特征数据可以包括：最近24周内KTV的历史消费价格平均值。
此外，由于非目标对象的价格参数在一定程度上会影响目标对象的价格参数， 所以非目标对象的价格参数可以作为一种特征数据。通过综合考虑目标对象的价格参数和非目标对象的价格参数，来预测针对目标对象的消费能力，可以得到更好的预测结果。
步骤203，根据对应目标对象的特征数据提取规则，从所述目标用户的历史数据中获取针对所述目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据。
针对需要预测其消费能力的目标用户，首先确定目标对象的特征数据提取规则，也即确定所述混合神经网络预测模型的使用的各个特征数据。在本例中可以将目标对象的价格参数作为一种特征数据，非目标对象的价格参数作为另一种特征数据。然后从目标用户的历史数据中获取针对目标对象的这两种特征数据的统计特征数据和时序序列特征数据。
参照图5，示出了本发明的消费能力预测的流程示意图。
统计特征数据可以包括：目标对象的历史消费价格参数、目标对象的历史浏览价格参数、非目标对象的历史消费价格参数、非目标对象的历史浏览价格参数、用户等级等特征。
其中，目标对象的历史消费价格参数可以包括目标对象的历史消费价格的平均值、最大值、最小值等。目标对象的历史浏览价格参数可以包括目标对象的历史浏览价格的平均值、最大值、最小值等。非目标对象的历史消费价格参数可以包括非目标对象的历史消费价格的平均值、最大值、最小值等。非目标对象的历史浏览价格参数以包括非目标对象的历史浏览价格的平均值、最大值、最小值等。用户等级等特征可以包括用户等级、用户活跃状态、用户常住地址等。
时序序列特征数据可以包括：目标对象的历史消费价格的平均值序列，目标对象的历史浏览价格平均值序列、非目标对象的历史消费价格平均值序列、非目标对象的历史浏览价格平均值序列等。
步骤204，基于所述目标用户的一个或多个时序序列特征数据，利用所述混合神经网络预测模型的所述循环神经网络确定对应的一个或多个时序特征数据。
如图5所示，将4个目标用户的时序序列特征数据分别输入到循环神经网络中，利用循环神经网络确定4个对应的时序特征数据。
步骤205，基于所述目标用户的一个或多个统计特征数据和所述对应的一个或多个时序特征数据，利用所述混合神经网络预测模型的所述传统神经网络确定所述目标用户针对所述目标对象的消费能力。
如图5所示，将目标用户的多个统计特征数据和4个时序特征数据输入传统神经网络中，利用传统神经网络确定目标用户针对目标对象的消费能力。
经过测试，在一个例子中，根据用户历史购买商品的价格平均值来确定其消费能力，其误差为40元。采用一般的机器学习模型，如线型回归模型(Linear Regression，LR)或梯度提升决策树(Gradient Boosting Decision Tree，GBDT)，其误差在33元左右。而采用本发明的混合神经网络预测模型，最终的预测误差在30元左右，其预测得到的消费能力更准确。
本发明实施例公开的消费能力预测方法，从样本用户的历史数据中，获取所述样本用户的统计特征数据、时序序列特征数据以及针对所述目标对象的实际消费价格，根据所述样本用户的统计特征数据、时序序列特征数据以及针对所述目标对象的实际消费价格训练获得所述混合神经网络预测模型。根据对应目标对象的特征数据提取规则，从所述目标用户的历史数据中获取针对所述目标对象的统计特征数据和时序序列特征数据。基于所述目标用户的时序序列特征数据，利用循环神经网络确定所述目标用户的时序特征数据。基于所述目标用户的统计特征数据和所述目标用户的时序特征数据，利用传统神经网络确定所述目标用户针对所述目标对象的消费能力。在统计特征数据的基础上结合时序序列特征数据，通过循环神经网络可实现对历史数据进行时序维度的特征提取，使得利用混合神经网络预测模型预测的消费能力更准确。
对于方法实施例，为了简单描述，故将其表述为一系列的动作组合，但是本领域技术人员应该知悉，本发明实施例并不受所描述的动作顺序的限制，因为依据本发明实施例，某些步骤可以采用其他顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作并不一定是本发明实施例所必须的。
实施例三
参照图6，示出了本发明实施例三的一种消费能力预测装置的结构框图。
本发明实施例的消费能力预测装置包括：
第一数据获取模块501，用于从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据。
消费能力确定模块502，用于基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目 标对象的消费能力。
本发明实施例公开的消费能力预测装置，从目标用户的历史数据中获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据，基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。解决了现有技术中利用用户最近一次购买商品的价格、随机某一次购买商品的价格，或者历史购买商品的价格平均值来确定用户的消费能力，其准确度较低的问题，在统计特征数据的基础上结合时序序列特征数据，可实现对历史数据进行时序维度的特征提取，使得利用混合神经网络预测模型预测的消费能力更准确。
实施例四
参照图7，示出了本发明实施例四的一种消费能力预测装置的结构框图。
基于实施例三，所述消费能力预测装置还包括：
第二数据获取模块503，用于从样本用户的历史数据中，获取针对所述目标对象的一个或多个统计特征数据、一个或多个时序序列特征数据以及实际消费价格；
模型训练模块504，用于根据所述样本用户的一个或多个统计特征数据、一个或多个时序序列特征数据以及所述实际消费价格训练所述混合神经网络预测模型；其中，所述混合神经网络预测模型包括循环神经网络和传统神经网络。
所述模型训练模块504，包括：
时序特征数据生成子模块5041，用于将所述样本用户的每个时序序列特征数据输入所述循环神经网络，获得对应的时序特征数据；
消费能力生成子模块5042，用于将所述样本用户的一个或多个统计特征数据与所述一个或多个时序特征数据输入所述传统神经网络，获得所述样本用户的预测消费能力；
权重值修正子模块5043，用于根据所述样本用户的预测消费能力与所述实际消费价格之间的偏差，修正所述混合神经网络预测模型中的各个权重值，直至所述偏差小于设定阈值。
可选的，当所述时序序列特征数据包括按时序排列的L个子特征数据时，所述时序特征数据生成子模块5041，包括：
第一输出结果生成单元50411，用于将第一个子特征数据输入所述循环神经网络，获得第一子特征数据的输出结果；
时序特征数据生成单元50412，用于将第m个子特征数据与第m-1个子特征数据的输出结果组合输入至所述循环神经网络，直至所述L个子特征数据全部输入完成，获得对应的所述时序特征数据。其中，m为大于1、且小于或等于L的正整数。
基于实施例三，所述消费能力确定模块502，包括：
时序特征数据确定子模块5021，用于基于所述目标用户的一个或多个时序序列特征数据，利用所述混合神经网络预测模型中的循环神经网络确定所述目标用户的一个或多个时序特征数据；
消费能力确定子模块5022，用于基于所述目标用户的一个或多个统计特征数据和所述目标用户的一个或多个时序特征数据，利用所述混合神经网络预测模型中的传统神经网络确定所述目标用户针对所述目标对象的消费能力。
基于实施例三，所述第一数据获取模块501，包括：
第一数据获取子模块5011，用于根据对应目标对象的特征数据提取规则，从所述目标用户的历史数据中获取针对所述目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据。
进一步的，所述消费能力预测装置还包括：
发放模块505，用于将与所述消费能力匹配的针对所述目标对象的优惠券，发送至所述目标用户；和/或将与所述消费能力匹配的针对所述目标对象的广告，投放给所述目标用户。
本发明实施例公开的消费能力预测装置，从样本用户的历史数据中，获取所述样本用户的统计特征数据、时序序列特征数据以及针对所述目标对象的实际消费价格，根据所述样本用户的统计特征数据、时序序列特征数据以及针对所述目标对象的实际消费价格训练获得所述混合神经网络预测模型，根据对应目标对象的特征数据提取规则，从所述目标用户的历史数据中获取针对所述目标对象的统计特征数据和时序序列特征数据，基于所述目标用户的时序序列特征数据，利用循环神经网络确定所述目标用户的时序特征数据，基于所述目标用户的统计特征数据和所述目标用户的时序特征数据，利用传统神经网络确定所述目标用户针对所述目标对象的消费能力。在统计特征数据的基础上结合时序序列特征数据，通过循环神经网络可实现对历史数据进行时序维度的特征 提取，使得利用混合神经网络预测模型预测的消费能力更准确。
相应的，本发明还公开了一种电子设备，参见图8，包括存储器820、处理器810以及存储在所述存储器820上并可在所述处理器上运行的计算机程序900，其特征在于，所述处理器810执行所述计算机程序900时实现如本发明实施例一和实施例二所述的消费能力预测方法。在其他可能的实现方式中，所述装置还可能包括总线830和外部接口840。处理器810和存储器820借由总线830相互连接，还可以通过外部接口840与其他设备或者部件进行通信。
本发明还公开了一种可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现如本发明实施例一和实施例二所述消费能力预测方法的步骤。
对于装置实施例而言，由于其与方法实施例基本相似，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
在此提供的算法和显示不与任何特定计算机、虚拟系统或者其它设备固有相关。各种通用系统也可以与基于在此的示教一起使用。根据上面的描述，构造这类系统所要求的结构是显而易见的。此外，本发明也不针对任何特定编程语言。应当明白，可以利用各种编程语言实现在此描述的本发明的内容，并且上面对特定语言所做的描述是为了披露本发明的最佳实施方式。
在此处所提供的说明书中，说明了大量具体细节。然而，能够理解，本发明的实施例可以在没有这些具体细节的情况下实践。在一些实例中，并未详细示出公知的方法、结构和技术，以便不模糊对本说明书的理解。
类似地，应当理解，为了精简本公开并帮助理解各个发明方面中的一个或多个，在上面对本发明的示例性实施例的描述中，本发明的各个特征有时被一起分组到单个实施例、图、或者对其的描述中。然而，并不应将该公开的方法解释成反映如下意图：即所要求保护的本发明要求比在每个权利要求中所明确记载的特征更多的特征。更确切地说，如下面的权利要求书所反映的那样，发明方面在于少于前面公开的单个实施例的所有特征。因此，遵循具体实施方式的权利要求书由此明确地并入该具体实施方式，其中每个权利要求本身都作为本发明的单独实施例。
本领域那些技术人员可以理解，可以对实施例中的设备中的模块进行自适应性地改变并且把它们设置在与该实施例不同的一个或多个设备中。可以把实施例中的模块或单元或组件组合成一个模块或单元或组件，以及此外可以把它们分成多个子模块或子 单元或子组件。除了这样的特征和/或过程或者单元中的至少一些是相互排斥之外，可以采用任何组合对本说明书(包括伴随的权利要求、摘要和附图)中公开的所有特征以及如此公开的任何方法或者设备的所有过程或单元进行组合。除非另外明确陈述，本说明书(包括伴随的权利要求、摘要和附图)中公开的每个特征可以由提供相同、等同或相似目的的替代特征来代替。
此外，本领域的技术人员能够理解，尽管在此所述的一些实施例包括其它实施例中所包括的某些特征而不是其它特征，但是不同实施例的特征的组合意味着处于本发明的范围之内并且形成不同的实施例。例如，在下面的权利要求书中，所要求保护的实施例的任意之一都可以以任意的组合方式来使用。
本发明的各个部件实施例可以以硬件实现，或者以在一个或者多个处理器上运行的软件模块实现，或者以它们的组合实现。本领域的技术人员应当理解，可以在实践中使用微处理器或者数字信号处理器(DSP)来实现根据本发明实施例的消费能力预测设备中的一些或者全部部件的一些或者全部功能。本发明还可以实现为用于执行这里所描述的方法的一部分或者全部的设备或者装置程序(例如，计算机程序和计算机程序产品)。这样的实现本发明的程序可以存储在计算机可读介质上，或者可以具有一个或者多个信号的形式。这样的信号可以从因特网网站上下载得到，或者在载体信号上提供，或者以任何其他形式提供。
应该注意的是上述实施例对本发明进行说明而不是对本发明进行限制，并且本领域技术人员在不脱离所附权利要求的范围的情况下可设计出替换实施例。在权利要求中，不应将位于括号之间的任何参考符号构造成对权利要求的限制。单词“包含”不排除存在未列在权利要求中的元件或步骤。位于元件之前的单词“一”或“一个”不排除存在多个这样的元件。本发明可以借助于包括有若干不同元件的硬件以及借助于适当编程的计算机来实现。在列举了若干装置的单元权利要求中，这些装置中的若干个可以是通过同一个硬件项来具体体现。单词第一、第二、以及第三等的使用不表示任何顺序。可将这些单词解释为名称。

Claims (10)

  1. 一种消费能力预测方法，包括：
    从目标用户的历史数据中，获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据；
    基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。
  2. 根据权利要求1所述的方法，其特征在于，还包括：
    从样本用户的历史数据中，获取针对所述目标对象的一个或多个统计特征数据、一个或多个时序序列特征数据以及实际消费价格；
    根据所述样本用户的一个或多个统计特征数据、一个或多个时序序列特征数据以及所述实际消费价格训练所述混合神经网络预测模型；其中，所述混合神经网络预测模型包括循环神经网络和传统神经网络。
  3. 根据权利要求2所述的方法，其特征在于，根据所述样本用户的一个或多个统计特征数据、一个或多个时序序列特征数据以及所述实际消费价格训练所述混合神经网络预测模型，包括：
    将所述样本用户的每个时序序列特征数据输入所述循环神经网络，获得对应的时序特征数据；
    将所述样本用户的一个或多个统计特征数据与所述一个或多个时序特征数据输入所述传统神经网络，获得所述样本用户的预测消费能力；
    根据所述样本用户的预测消费能力与所述实际消费价格之间的偏差，修正所述混合神经网络预测模型中的各个权重值，直至所述偏差小于设定阈值。
  4. 根据权利要求3所述的方法，其特征在于，当所述时序序列特征数据包括按时序排列的L个子特征数据时，获得对应的所述时序特征数据，包括：
    将第一个子特征数据输入所述循环神经网络，获得第一子特征数据的输出结果；
    将第m个子特征数据与第m-1个子特征数据的输出结果组合输入至所述循环神经网络，直至所述L个子特征数据全部输入完成，获得对应的所述时序特征数据；其中，m为大于1、且小于或等于L的正整数。
  5. 根据权利要求1所述的方法，其特征在于，基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用所述混合神经网络预测模型确定所述目标用户 针对所述目标对象的消费能力，包括：
    基于所述目标用户的一个或多个时序序列特征数据，利用所述混合神经网络预测模型中的循环神经网络确定所述目标用户的一个或多个时序特征数据；
    基于所述目标用户的一个或多个统计特征数据和所述目标用户的一个或多个时序特征数据，利用所述混合神经网络预测模型中的传统神经网络确定所述目标用户针对所述目标对象的消费能力。
  6. 根据权利要求1所述的方法，其特征在于，从所述目标用户的历史数据中，获取针对所述目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据，包括：
    根据对应所述目标对象的特征数据提取规则，从所述目标用户的历史数据中获取针对所述目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据。
  7. 根据权利要求1所述的方法，其特征在于，还包括：
    将与所述消费能力匹配的针对所述目标对象的优惠券，发送至所述目标用户；和/或
    将与所述消费能力匹配的针对所述目标对象的广告，投放给所述目标用户。
  8. 一种消费能力预测装置，包括：
    第一数据获取模块，用于从目标用户的历史数据中，获取针对目标对象的一个或多个统计特征数据和一个或多个时序序列特征数据；
    消费能力确定模块，用于基于所述一个或多个统计特征数据和所述一个或多个时序序列特征数据，利用预设的混合神经网络预测模型确定所述目标用户针对所述目标对象的消费能力。
  9. 一种电子设备，包括存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现权利要求1-7任意一项权利要求所述的消费能力预测方法。
  10. 一种可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现权利要求1-7任意一项所述消费能力预测方法的步骤。
PCT/CN2018/108340 2017-10-11 2018-09-28 消费能力预测 WO2019072107A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/755,880 US20200285937A1 (en) 2017-10-11 2018-09-28 Consumption capacity prediction

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710943388.2 2017-10-11
CN201710943388.2A CN107705155A (zh) 2017-10-11 2017-10-11 一种消费能力预测方法、装置、电子设备及可读存储介质

Publications (1)

Publication Number Publication Date
WO2019072107A1 true WO2019072107A1 (zh) 2019-04-18

Family

ID=61184992

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/108340 WO2019072107A1 (zh) 2017-10-11 2018-09-28 消费能力预测

Country Status (3)

Country Link
US (1) US20200285937A1 (zh)
CN (1) CN107705155A (zh)
WO (1) WO2019072107A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117708764A (zh) * 2024-02-06 2024-03-15 青岛天高智慧科技有限公司 基于校园一卡通的学生消费数据智能分析方法

Families Citing this family (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107705155A (zh) * 2017-10-11 2018-02-16 北京三快在线科技有限公司 一种消费能力预测方法、装置、电子设备及可读存储介质
CN108573399B (zh) * 2018-02-28 2022-03-18 中国银联股份有限公司 基于转移概率网络的商户推荐方法及其系统
CN108960563A (zh) * 2018-05-22 2018-12-07 深圳壹账通智能科技有限公司 一种商店的评级方法及其设备
CN110555714A (zh) * 2018-06-04 2019-12-10 百度在线网络技术（北京）有限公司 用于输出信息的方法和装置
CN108846692B (zh) * 2018-06-05 2021-03-30 浙江大学城市学院 一种基于多因素循环神经网络的消费者消费行为预测方法
CN109359758A (zh) * 2018-08-03 2019-02-19 阿里巴巴集团控股有限公司 外汇交易量预测方法和装置
KR20210059712A (ko) 2018-08-07 2021-05-25 블링크에이아이 테크놀로지스, 아이엔씨. 이미지 향상을 위한 인공지능 기법
CN109493134A (zh) * 2018-10-31 2019-03-19 北京字节跳动网络技术有限公司 用于推送信息的方法和装置
CN110188796A (zh) * 2019-04-25 2019-08-30 博彦科技股份有限公司 用户识别方法、装置、存储介质和处理器
CN110335062B (zh) * 2019-05-28 2022-07-12 成都美美臣科技有限公司 一个电子商务网站自动化优惠券推送方法
US20220270126A1 (en) * 2019-06-14 2022-08-25 Beijing DIDI Infinity Technology and Development Co., Ltd Reinforcement Learning Method For Incentive Policy Based On Historic Data Trajectory Construction
CN110689170A (zh) * 2019-09-04 2020-01-14 北京三快在线科技有限公司 对象参量的确定方法、装置、电子设备及存储介质
CN110874765B (zh) * 2019-10-22 2021-10-22 腾讯科技（深圳）有限公司 一种数据处理方法、装置、设备及存储介质
CN110880127B (zh) * 2019-10-24 2022-04-12 北京三快在线科技有限公司 消费水平的预测方法、装置、电子设备及存储介质
CN110992097B (zh) * 2019-12-03 2024-02-13 上海钧正网络科技有限公司 营收产品价格的处理方法、装置、计算机设备及存储介质
CN110990704A (zh) * 2019-12-06 2020-04-10 创新奇智(成都)科技有限公司 时间序列用户与内容互动行为的学习预测方法
CN111461866B (zh) * 2020-03-31 2023-09-26 中国银行股份有限公司 一种数据分析方法及装置
CN111708768A (zh) * 2020-05-20 2020-09-25 广州中智融通金融科技有限公司 自动柜员机现金调度方法、系统及存储介质
CN111915344A (zh) * 2020-06-20 2020-11-10 武汉海云健康科技股份有限公司 一种基于医疗大数据的新会员催熟方法和装置
CN112070523A (zh) * 2020-07-23 2020-12-11 盛威时代科技集团有限公司 基于云计算技术在智能交通管理产品中投送广告的方法
CN112256961B (zh) * 2020-10-19 2024-04-09 平安科技（深圳）有限公司 用户画像生成方法、装置、设备及介质
CN112381617B (zh) * 2020-11-30 2023-09-05 中国银行股份有限公司 网络购物方法及装置
CN113778979A (zh) * 2021-01-21 2021-12-10 北京沃东天骏信息技术有限公司 一种直播点击率的确定方法和装置
US11710148B2 (en) * 2021-01-31 2023-07-25 Walmart Apollo, Llc Deep learning-based revenue-per-click prediction model framework
CN113095890A (zh) * 2021-04-29 2021-07-09 平安普惠企业管理有限公司 电子优惠券发放方法、装置、计算机设备和存储介质
CN113935697B (zh) * 2021-12-15 2022-04-08 畅捷通信息技术股份有限公司 一种自动售货机的备货方法、装置及存储介质
CN114897517B (zh) * 2022-07-18 2022-09-27 山东美丽乡村云计算有限公司 基于区块链的文旅消费数据管理方法
CN117934064A (zh) * 2022-10-12 2024-04-26 华为云计算技术有限公司 一种数据处理方法、模型训练方法及相关设备
CN116414662B (zh) * 2023-04-24 2023-10-10 山东溯源安全科技有限公司 一种存储空间扩容提示方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101673385A (zh) * 2009-09-28 2010-03-17 百度在线网络技术（北京）有限公司 消费预估方法和装置
CN105335875A (zh) * 2015-10-30 2016-02-17 小米科技有限责任公司 购买力预测方法和装置
CN105868847A (zh) * 2016-03-24 2016-08-17 车智互联（北京）科技有限公司 一种购物行为的预测方法及装置
CN106296257A (zh) * 2015-06-11 2017-01-04 苏宁云商集团股份有限公司 一种基于用户行为分析的固定广告位投放方法及系统
CN107705155A (zh) * 2017-10-11 2018-02-16 北京三快在线科技有限公司 一种消费能力预测方法、装置、电子设备及可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101673385A (zh) * 2009-09-28 2010-03-17 百度在线网络技术（北京）有限公司 消费预估方法和装置
CN106296257A (zh) * 2015-06-11 2017-01-04 苏宁云商集团股份有限公司 一种基于用户行为分析的固定广告位投放方法及系统
CN105335875A (zh) * 2015-10-30 2016-02-17 小米科技有限责任公司 购买力预测方法和装置
CN105868847A (zh) * 2016-03-24 2016-08-17 车智互联（北京）科技有限公司 一种购物行为的预测方法及装置
CN107705155A (zh) * 2017-10-11 2018-02-16 北京三快在线科技有限公司 一种消费能力预测方法、装置、电子设备及可读存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117708764A (zh) * 2024-02-06 2024-03-15 青岛天高智慧科技有限公司 基于校园一卡通的学生消费数据智能分析方法
CN117708764B (zh) * 2024-02-06 2024-05-03 青岛天高智慧科技有限公司 基于校园一卡通的学生消费数据智能分析方法

Also Published As

Publication number Publication date
US20200285937A1 (en) 2020-09-10
CN107705155A (zh) 2018-02-16

Similar Documents

Publication Publication Date Title
WO2019072107A1 (zh) 消费能力预测
US20210049622A1 (en) Marketing method and apparatus based on deep reinforcement learning
JP6414363B2 (ja) 予測システム、方法およびプログラム
CN107133865B (zh) 一种信用分的获取、特征向量值的输出方法及其装置
US8732015B1 (en) Social media pricing engine
CN110796477A (zh) 广告展示方法、装置、电子设备及可读存储介质
US10262336B2 (en) Non-converting publisher attribution weighting and analytics server and method
CN107909392B (zh) 刺激行为敏感度预测方法、装置、设备及可读存储介质
US10572912B2 (en) System and method for integrating retail price optimization for revenue and profit with business rules
US10776816B2 (en) System and method for building a targeted audience for an online advertising campaign
CN108109011A (zh) 一种广告反作弊方法及计算设备
CN109284881A (zh) 订单分配方法、装置、计算机可读存储介质及电子设备
JP2016511906A (ja) 商品検索結果の順位付け
JP6311851B2 (ja) 共クラスタリングシステム、方法およびプログラム
WO2014173237A1 (zh) 一种推荐的方法及服务器
CN111008335B (zh) 一种信息处理方法、装置、设备及存储介质
CN112330358A (zh) 产品销量预测的方法和装置、存储介质和电子设备
CN108446297A (zh) 一种推荐方法及装置，电子设备
KR101639656B1 (ko) 광고 제공 방법 및 서버 장치
CN111680213B (zh) 信息推荐方法、数据处理方法及装置
CN112560105B (zh) 保护多方数据隐私的联合建模方法及装置
Chan et al. Forecasting online auctions via self‐exciting point processes
Chan et al. Modeling and forecasting online auction prices: a semiparametric regression analysis
CN113128597B (zh) 一种用户行为特征的提取和分类预测的方法及装置
CN103020081A (zh) 站外投放关键词方法和系统

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: 18866715

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18866715

Country of ref document: EP

Kind code of ref document: A1