CN111292106A - Method and device for determining business demand influence factors - Google Patents

Method and device for determining business demand influence factors Download PDF

Info

Publication number
CN111292106A
CN111292106A CN201811489839.0A CN201811489839A CN111292106A CN 111292106 A CN111292106 A CN 111292106A CN 201811489839 A CN201811489839 A CN 201811489839A CN 111292106 A CN111292106 A CN 111292106A
Authority
CN
China
Prior art keywords
user
order
sample
feature vector
characteristic
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN201811489839.0A
Other languages
Chinese (zh)
Inventor
薄琳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Didi Infinity Technology and Development Co Ltd
Original Assignee
Beijing Didi Infinity Technology and Development Co Ltd
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 Beijing Didi Infinity Technology and Development Co Ltd filed Critical Beijing Didi Infinity Technology and Development Co Ltd
Priority to CN201811489839.0A priority Critical patent/CN111292106A/en
Publication of CN111292106A publication Critical patent/CN111292106A/en
Pending legal-status Critical Current

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Artificial Intelligence (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application provides a method and a device for determining business demand influence factors, wherein the method comprises the following steps: the method comprises the steps of obtaining order information of a plurality of orders of a user to be predicted within a preset historical time period; according to the order information of the orders, constructing an order feature vector sequence of the user to be predicted, wherein the order feature vector sequence comprises order feature vectors corresponding to the orders; and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model. The method and the device for predicting the business demand influence factor of the user to be predicted in the future preset time period can be determined with higher accuracy.

Description

Method and device for determining business demand influence factors
Technical Field
The application relates to the technical field of machine learning, in particular to a method and a device for determining business demand influence factors.
Background
The silent user refers to a user who has generated historical behavior data for a certain service in a certain time period, but has not generated historical behavior data for the service in another time period after the certain time period, for example, a user who has passed a car reservation on a car reservation platform but has not reserved the car on the car reservation platform in the last month is a silent user of the car reservation platform.
There is a need to obtain the reason for the silence of the user so as to better perform resource allocation and better provide service for the user. When the cause of the user's silence is obtained, it is usually obtained based on statistics, such as the number of user logins, the number of invoices made by the user in a preset time period, the number of bubbles made by the user in the preset time period, and the like. The statistics can generally characterize the accumulated characteristics of the historical behaviors of the users in a certain period of time, but the silence reason of the users determined based on the statistics has the problem of low prediction accuracy.
Disclosure of Invention
In view of this, an object of the present application is to provide a method and an apparatus for determining a business demand influencing factor, which can determine the business demand influencing factor of a user to be predicted in a future preset time period with a higher accuracy.
In a first aspect, a method for determining business requirement influencing factors is provided, where the method includes:
the method comprises the steps of obtaining order information of a plurality of orders of a user to be predicted within a preset historical time period;
according to the order information of the orders, constructing an order feature vector sequence of the user to be predicted, wherein the order feature vector sequence comprises order feature vectors corresponding to the orders;
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model.
In an optional implementation manner, constructing an order feature vector sequence of the user to be predicted according to the order information of the plurality of orders includes:
for each order, determining characteristic values of the order under a plurality of order characteristics according to order information of the order;
according to the characteristic values of the order under the characteristics of the orders, constructing an order characteristic vector corresponding to the order;
and according to the order characteristic vectors corresponding to the orders respectively, constructing the order characteristic vector sequence according to the sequence of order generation time.
In an optional implementation manner, the determining, based on the order feature vector sequence and a pre-trained prediction model, influence factor information of a business demand of the user to be predicted in a future preset time period includes:
inputting the order characteristic vector sequence into the prediction model to obtain a prediction result corresponding to the order characteristic vector sequence;
determining a second contribution value corresponding to each order feature based on a prediction result corresponding to the order feature vector sequence, elements in each feature vector sequence and reference values corresponding to the elements in the feature vector sequence;
and determining influence factor information of the user to be predicted in the order characteristic dimension based on the second contribution value.
In an optional implementation manner, the determining, based on the second contribution value, influence factor information of the user to be predicted in an order feature dimension includes:
determining at least one target order characteristic from the order characteristics according to the magnitude of the second contribution value;
and determining the influence factor information of the user to be predicted in order feature dimensions based on the influence factor classification corresponding to the target order features.
In an alternative embodiment, the classification of the influencing factors includes: the system is sensitive to one or more of price, few vehicles, waiting time, driver service, potential safety hazard and vehicle type and vehicle condition.
In an alternative embodiment, the second contribution value integrated gradsi(x) Satisfies the following formula:
Figure BDA0001895419220000031
wherein x isiRepresenting the ith element in the order feature vector; x is the number ofi' indicating an orderA reference value corresponding to the ith element in the feature vector; f represents a prediction function corresponding to the prediction model; x' represents a feature vector composed of reference values of the features of each order; x represents an order feature vector in the sequence of order feature vectors.
In an optional implementation manner, after obtaining order information of a plurality of orders of a user to be predicted within a preset historical time period, the method further includes:
according to the user information of the user to be predicted, constructing a user characteristic vector of the user to be predicted;
the determining of the influence factor information of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence and the pre-trained prediction model comprises the following steps:
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model.
In an optional implementation manner, the determining, based on the order feature vector sequence, the user feature vector input, and the pre-trained prediction model, influence factor information of a service demand of the user to be predicted in a future preset time period includes:
inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, obtaining attention distribution weights corresponding to the order feature vectors from a target network layer of the prediction model, and taking the attention distribution weights as first contribution values of orders corresponding to the order feature vectors to silence the user to be predicted;
and determining influence factor information of the user to be predicted in the order dimension based on the first contribution value.
In an optional implementation manner, the constructing, according to the user information, a user feature vector of the user to be predicted includes:
determining characteristic values of the user to be predicted under various user characteristics according to the user information;
and constructing a user characteristic vector of the user to be predicted according to the characteristic values of the user to be predicted under the various user characteristics.
In an alternative embodiment, the user features include a plurality of the following features:
age, gender, occupation, complaint factors of the service requester.
In an alternative embodiment, the order characteristics include a plurality of the following characteristics:
the order starting point distance, the queuing waiting time, the order price, the order duration, the driving receiving distance, the pre-evaluation value and the service provider information.
In an optional embodiment, the service provider information comprises at least one of the following information:
age, gender, rating level, service score, service equipment level, service equipment cleanliness, order completion rate, complained factors.
In an alternative embodiment, the predictive model is trained in the following way:
obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and an actual business demand result of each sample user in a prediction historical time period;
for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders;
determining a service demand prediction result of each sample user in a prediction history time period based on the sample order feature vector sequence of each sample user and a basic prediction model;
and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the prediction model.
In an optional implementation, after obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period, the method further includes:
for each sample user, generating a user feature vector of the sample user according to the user information of the sample user;
the determining a service demand prediction result of each sample user based on the sample order feature vector sequence and the basic prediction model of each sample user comprises:
and inputting the sample order characteristic vector sequence and the user characteristic vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user.
In an optional implementation manner, the inputting the sample order feature vector sequence and the user feature vector of each sample user into a basic prediction model to obtain a business demand prediction result of each sample user includes:
for each sample user, inputting a sample order feature vector in a sample order feature vector sequence of the sample user into a first neural network, and acquiring a middle feature vector corresponding to each order feature vector;
inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and acquiring attention distribution weights corresponding to each intermediate feature vector;
generating a fused feature vector based on each of the intermediate feature vectors of the sample user and the attention distribution weight;
and inputting the fusion characteristic vector of the sample user into a third neural network, extracting a target characteristic vector for the fusion characteristic vector, and inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
In an optional implementation manner, the inputting, for each sample user, a sample order feature vector in a sample order feature vector sequence of the sample user into the first neural network, and obtaining an intermediate feature vector corresponding to each order feature vector includes:
selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors of the sample user;
inputting the obtained characteristic vector of the previous sample order into the first neural network, wherein a target characteristic extraction layer of the first neural network is an intermediate characteristic vector output by the characteristic vector of the previous sample order;
inputting the current sample order feature vector and the intermediate feature vector of the previous sample order feature vector into the first neural network, and acquiring the intermediate feature vector corresponding to the current sample order feature vector;
and returning a sample order feature vector sequence aiming at the sample user, and selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors until the intermediate feature vectors of all the sample order feature vectors are extracted.
In an alternative embodiment, the inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and obtaining the attention allocation weight corresponding to each intermediate feature vector, includes:
for each intermediate feature vector, splicing the intermediate feature vector with the user feature vector to generate a spliced vector corresponding to the intermediate feature vector;
inputting the splicing vector into the second neural network to obtain the matching degree corresponding to the splicing vector;
and performing activation operation on the matching degree based on a preset activation function to obtain the attention distribution weight corresponding to the intermediate feature vector.
In an optional embodiment, the generating a fused feature vector based on each of the intermediate feature vectors and the attention assignment weight of the sample user includes:
and carrying out weighted summation on the intermediate feature vectors according to the attention distribution weights corresponding to the intermediate feature vectors to generate the fusion feature vectors.
In an optional implementation manner, the training the basic prediction model according to the service demand prediction result of each sample user and the corresponding actual service demand result includes:
taking any sample user in sample users who have not completed training in the current round as a current sample user, and determining the cross entropy loss of the current sample user in the current round according to the business demand prediction result of the current sample user and the corresponding actual business demand result;
adjusting parameters of the basic prediction model according to the cross entropy loss of the current sample user in the current round;
and taking the current sample user as a sample user completing the training in the current round, and returning to the step of determining the cross entropy loss of the current sample user in the current round until all sample users complete the training in the current round.
In an alternative embodiment, after completing the current round of training of the base prediction model, the method further includes:
detecting whether the number of the current wheel reaches a preset number; if so, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
verifying the basic prediction model obtained in the current round by using a test set; if the cross entropy loss is not greater than the number of the test data of the preset cross entropy loss threshold value in the test set, the percentage of the total number of the test data in the test set is occupied, and the percentage is greater than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the sample user corresponding to the previous round in sequence; and if the cross entropy loss of the sample user in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample user in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as the prediction model.
In an alternative embodiment, the sample order feature vector is generated as follows:
for each sample order, determining characteristic values of the sample order under a plurality of order characteristics according to order information of the sample order;
detecting whether the dimension of the characteristic value corresponding to each order characteristic is larger than a preset dimension or not;
if the dimension of the characteristic value corresponding to any order characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any order characteristic to obtain a compressed characteristic value of any order characteristic;
and constructing the sample order feature vector based on the compressed feature value corresponding to any order feature and the feature values corresponding to other order features.
In an alternative embodiment, the user order feature vector is generated by:
for each sample user, determining a characteristic value of the sample user under a plurality of user characteristics according to the user information of the sample user;
detecting whether the dimensionality of the characteristic value corresponding to each user characteristic is larger than a preset dimensionality;
if the dimension of the characteristic value corresponding to any user characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any user characteristic to obtain a compressed characteristic value of any user characteristic;
and constructing the user feature vector based on the compressed feature value corresponding to any user feature and the feature values corresponding to other user features.
In an alternative embodiment, the feature values are dimension compressed in the following manner:
acquiring a characteristic embedding matrix;
determining a product of the eigenvalue and the feature embedding matrix as the compressed eigenvalue.
In a second aspect, an embodiment of the present application provides a device for determining business requirement influencing factors, where the device includes:
the obtaining module 11 is configured to obtain order information of a plurality of orders of a user to be predicted within a preset historical time period;
a constructing module 12, configured to construct an order feature vector sequence of the user to be predicted according to the order information of the multiple orders, where the order feature vector sequence includes an order feature vector corresponding to each order;
and the determining module 13 is configured to determine, based on the order feature vector sequence and a pre-trained prediction model, influence factor information of a service demand of the user to be predicted in a future preset time period.
In an optional embodiment, the constructing module 12 is configured to construct the order feature vector sequence of the user to be predicted according to the order information of the multiple orders by using the following steps, including:
for each order, determining characteristic values of the order under a plurality of order characteristics according to order information of the order;
according to the characteristic values of the order under the characteristics of the orders, constructing an order characteristic vector corresponding to the order;
and according to the order characteristic vectors corresponding to the orders respectively, constructing the order characteristic vector sequence according to the sequence of order generation time.
In an optional embodiment, the determining module 13 is configured to determine, based on the order feature vector sequence and a pre-trained prediction model, influence factor information of a business demand of the user to be predicted in a future preset time period by using the following steps:
inputting the order characteristic vector sequence into the prediction model to obtain a prediction result corresponding to the order characteristic vector sequence;
determining a second contribution value corresponding to each order feature based on a prediction result corresponding to the order feature vector sequence, elements in each feature vector sequence and reference values corresponding to the elements in the feature vector sequence;
and determining influence factor information of the user to be predicted in the order characteristic dimension based on the second contribution value.
In an optional embodiment, the determining module 13 is configured to determine, based on the second contribution value, influence factor information of the user to be predicted in an order feature dimension by using the following steps, including:
determining at least one target order characteristic from the order characteristics according to the magnitude of the second contribution value;
and determining the influence factor information of the user to be predicted in order feature dimensions based on the influence factor classification corresponding to the target order features.
In an alternative embodiment, the classification of the influencing factors includes: the system is sensitive to one or more of price, few vehicles, waiting time, driver service, potential safety hazard and vehicle type and vehicle condition.
In an alternative embodiment, the second contribution value integrated gradsi(x) Satisfies the following formula:
Figure BDA0001895419220000101
wherein x isiRepresenting the ith element in the order feature vector; x'iRepresenting a reference value corresponding to the ith element in the order feature vector; f represents a prediction function corresponding to the prediction model; x' represents a feature vector composed of reference values of the features of each order; x represents an order feature vector in the sequence of order feature vectors.
In an alternative embodiment, the building module 12 is further configured to:
according to the user information of the user to be predicted, constructing a user characteristic vector of the user to be predicted;
the determining module 13 is configured to determine, based on the order feature vector sequence and a pre-trained prediction model, influence factor information of a service demand of the user to be predicted in a future preset time period by using the following steps:
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model.
In an optional embodiment, the determining module 13 is configured to determine, based on the order feature vector sequence, the user feature vector input, and the pre-trained prediction model, influence factor information of a business demand of the user to be predicted in a future preset time period by using the following steps:
inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, obtaining attention distribution weights corresponding to the order feature vectors from a target network layer of the prediction model, and taking the attention distribution weights as first contribution values of orders corresponding to the order feature vectors to silence the user to be predicted;
and determining influence factor information of the user to be predicted in the order dimension based on the first contribution value.
In an optional embodiment, the constructing module 12 is configured to construct the user feature vector of the user to be predicted according to the user information by adopting the following steps:
determining characteristic values of the user to be predicted under various user characteristics according to the user information;
and constructing a user characteristic vector of the user to be predicted according to the characteristic values of the user to be predicted under the various user characteristics.
In an alternative embodiment, the user features include a plurality of the following features:
age, gender, occupation, complaint factors of the service requester.
In an alternative embodiment, the order characteristics include a plurality of the following characteristics:
the order starting point distance, the queuing waiting time, the order price, the order duration, the driving receiving distance, the pre-evaluation value and the service provider information.
In an optional embodiment, the service provider information comprises at least one of the following information:
age, gender, rating level, service score, service equipment level, service equipment cleanliness, order completion rate, complained factors.
In an alternative embodiment, the method further comprises: a training module 14 for training the predictive model in the following way:
obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and an actual business demand result of each sample user in a prediction historical time period;
for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders;
determining a service demand prediction result of each sample user in a prediction history time period based on the sample order feature vector sequence of each sample user and a basic prediction model;
and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the prediction model.
In an optional embodiment, the training module 14 is further configured to, after obtaining sample order information of a plurality of sample orders of a plurality of sample users within a target historical time period, for each sample user, generate a user feature vector of the sample user according to the user information of the sample user;
the training module 14 is configured to determine a service demand prediction result of each sample user based on the sample order feature vector sequence and the basic prediction model of each sample user by using the following steps:
and inputting the sample order characteristic vector sequence and the user characteristic vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user.
In an optional embodiment, the training module 14 is configured to input the sample order feature vector sequence and the user feature vector of each sample user into a basic prediction model to obtain a business demand prediction result of each sample user by using the following steps:
for each sample user, inputting a sample order feature vector in a sample order feature vector sequence of the sample user into a first neural network, and acquiring a middle feature vector corresponding to each order feature vector;
inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and acquiring attention distribution weights corresponding to each intermediate feature vector;
generating a fused feature vector based on each of the intermediate feature vectors of the sample user and the attention distribution weight;
and inputting the fusion characteristic vector of the sample user into a third neural network, extracting a target characteristic vector for the fusion characteristic vector, and inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
In an optional embodiment, the training module 14 is configured to, for each sample user, input a sample order feature vector in a sample order feature vector sequence of the sample user into the first neural network, and obtain an intermediate feature vector corresponding to each order feature vector, where:
selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors of the sample user;
inputting the obtained characteristic vector of the previous sample order into the first neural network, wherein a target characteristic extraction layer of the first neural network is an intermediate characteristic vector output by the characteristic vector of the previous sample order;
inputting the current sample order feature vector and the intermediate feature vector of the previous sample order feature vector into the first neural network, and acquiring the intermediate feature vector corresponding to the current sample order feature vector;
and returning a sample order feature vector sequence aiming at the sample user, and selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors until the intermediate feature vectors of all the sample order feature vectors are extracted.
In an alternative embodiment, the training module 14 is configured to input each intermediate feature vector of the sample user and the user feature vector to a second neural network, and obtain an attention allocation weight corresponding to each intermediate feature vector by:
for each intermediate feature vector, splicing the intermediate feature vector with the user feature vector to generate a spliced vector corresponding to the intermediate feature vector;
inputting the splicing vector into the second neural network to obtain the matching degree corresponding to the splicing vector;
and performing activation operation on the matching degree based on a preset activation function to obtain the attention distribution weight corresponding to the intermediate feature vector.
In an alternative embodiment, the training module 14 is configured to generate a fused feature vector based on each of the intermediate feature vectors and the attention assignment weight of the sample user by:
and carrying out weighted summation on the intermediate feature vectors according to the attention distribution weights corresponding to the intermediate feature vectors to generate the fusion feature vectors.
In an optional implementation manner, the training module 14 is configured to train the basic prediction model according to the service demand prediction result of each sample user and the corresponding actual service demand result by:
taking any sample user in sample users who have not completed training in the current round as a current sample user, and determining the cross entropy loss of the current sample user in the current round according to the business demand prediction result of the current sample user and the corresponding actual business demand result;
adjusting parameters of the basic prediction model according to the cross entropy loss of the current sample user in the current round;
and taking the current sample user as a sample user completing the training in the current round, and returning to the step of determining the cross entropy loss of the current sample user in the current round until all sample users complete the training in the current round.
In an optional embodiment, the training module 14 is further configured to detect whether the current round reaches a preset number of rounds after completing the current round of training of the basic prediction model; if so, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
verifying the basic prediction model obtained in the current round by using a test set; if the cross entropy loss is not greater than the number of the test data of the preset cross entropy loss threshold value in the test set, the percentage of the total number of the test data in the test set is occupied, and the percentage is greater than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the sample user corresponding to the previous round in sequence; and if the cross entropy loss of the sample user in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample user in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as the prediction model.
In an alternative embodiment, the training module 14 is configured to generate the sample order feature vector by:
for each sample order, determining characteristic values of the sample order under a plurality of order characteristics according to order information of the sample order;
detecting whether the dimension of the characteristic value corresponding to each order characteristic is larger than a preset dimension or not;
if the dimension of the characteristic value corresponding to any order characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any order characteristic to obtain a compressed characteristic value of any order characteristic;
and constructing the sample order feature vector based on the compressed feature value corresponding to any order feature and the feature values corresponding to other order features.
In an alternative embodiment, the training module 14 is configured to generate the user order feature vector by:
for each sample user, determining a characteristic value of the sample user under a plurality of user characteristics according to the user information of the sample user;
detecting whether the dimensionality of the characteristic value corresponding to each user characteristic is larger than a preset dimensionality;
if the dimension of the characteristic value corresponding to any user characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any user characteristic to obtain a compressed characteristic value of any user characteristic;
and constructing the user feature vector based on the compressed feature value corresponding to any user feature and the feature values corresponding to other user features.
In an alternative embodiment, the training module 14 is configured to perform dimension compression on the feature values by using the following steps:
acquiring a characteristic embedding matrix;
determining a product of the eigenvalue and the feature embedding matrix as the compressed eigenvalue.
In a third aspect, an embodiment of the present application provides an electronic device, including: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, when the electronic device runs, the processor and the storage medium communicate with each other through the bus, and the processor executes the machine-readable instructions to perform the steps of the business requirement influencing factor method according to any one of the first aspect.
In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the business requirement influencing factor method according to any one of the first aspects are performed.
According to the method and the device, the order information of a plurality of orders of the user to be predicted in the preset historical time period is obtained, the order characteristic vector sequence of the user to be predicted is built, and the business demand influence factors of the user to be predicted in the future preset time period are determined with higher accuracy according to the order characteristic vector sequence and a pre-trained business demand prediction model, so that resource allocation is better performed based on the influence factors, and the service quality is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.
Fig. 1 is a block diagram of a system 100 in a scenario in which a business requirement influencing factor determining method according to some embodiments of the present application is applied;
FIG. 2 illustrates a schematic diagram of exemplary hardware and software components of an electronic device 200 of a server 110, a service requester terminal 130, a service provider terminal 140, which may implement the concepts of the present application, according to some embodiments of the present application;
fig. 3 is a flowchart illustrating a method for determining business requirement influencing factors according to an embodiment of the present application;
fig. 4 is a flowchart illustrating a specific method for constructing an order feature vector sequence of a user to be predicted according to order information of a plurality of orders in determining business demand influencing factors provided in embodiments of the present application;
fig. 5 is a flowchart illustrating a specific method for constructing a user feature vector of a user to be predicted in determining a business demand influencing factor according to embodiments of the present application;
fig. 6 is a flowchart illustrating a specific method for determining influence factor information of a service demand of a user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model in determining the influence factor of the service demand provided in embodiments of the present application;
fig. 7 is a flowchart illustrating a specific method for determining influence factor information of the user to be predicted in the order feature dimension based on the second contribution value in determining the business demand influence factor provided in the embodiments of the present application;
FIG. 8 is a flowchart illustrating a specific method for obtaining a pre-trained business demand prediction model in determining business demand influencing factors provided by embodiments of the present application;
fig. 9 is a flowchart illustrating a specific method for determining influence factor information of a service demand of a user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector and a pre-trained prediction model in determining the service demand influence factor provided in embodiments of the present application;
FIG. 10 is a flowchart illustrating another specific method for obtaining a pre-trained business need prediction model in determining business need influencing factors provided by embodiments of the present application;
fig. 11 is a schematic diagram illustrating a specific method for obtaining a business demand prediction result of each sample user by inputting a sample order feature vector sequence and a user feature vector of each sample user into a basic prediction model in determining business demand influence factors provided in embodiments of the present application;
fig. 12 is a schematic structural diagram illustrating a basic prediction network in determining business demand influencing factors provided in embodiments of the present application;
fig. 13 is a schematic structural diagram illustrating a service demand influence factor determining apparatus according to an embodiment of the present application.
Detailed Description
In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it should be understood that the drawings in the present application are for illustrative and descriptive purposes only and are not used to limit the scope of protection of the present application. Additionally, it should be understood that the schematic drawings are not necessarily drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of the present application. It should be understood that the operations of the flow diagrams may be performed out of order, and steps without logical context may be performed in reverse order or simultaneously. One skilled in the art, under the guidance of this application, may add one or more other operations to, or remove one or more operations from, the flowchart.
In addition, the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.
In order to enable those skilled in the art to use the present disclosure, a related introduction is made in conjunction with a specific application scenario "net appointment travel scenario". It will be apparent to those skilled in the art that the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the application. Although the present application is described primarily in the context of predicting traffic demands of network car booking passengers, it should be understood that this is merely one exemplary embodiment. The application can be applied to any other traffic type. For example, the present application may be applied to different transportation system environments, including terrestrial, marine, or airborne, among others, or any combination thereof. The vehicle of the transportation system may include a taxi, a private car, a windmill, a bus, a train, a bullet train, a high speed rail, a subway, a ship, an airplane, a spacecraft, a hot air balloon, or an unmanned vehicle, etc., or any combination thereof. The present application may also include any service system for providing a service to a user based on the internet, for example, a system for sending and/or receiving a courier, a service system for a business to a seller. Applications of the system or method of the present application may include web pages, plug-ins for browsers, client terminals, customization systems, internal analysis systems, or artificial intelligence robots, among others, or any combination thereof.
It should be noted that in the embodiments of the present application, the term "comprising" is used to indicate the presence of the features stated hereinafter, but does not exclude the addition of further features.
One aspect of the present application relates to a method for determining business demand impact factors. According to the method, the order characteristic vector sequence of the user to be predicted can be constructed through the obtained order information of a plurality of orders of the user to be predicted in the preset historical time period, and the influence factor information of the service demand of the user to be predicted in the future preset time period can be determined with higher accuracy according to the order characteristic vector sequence and a pre-trained prediction model, so that resource configuration can be better carried out based on the influence factors, and the service quality is improved.
It is noted that, before the application is filed, the determination of the influence factors of the user service demand is generally performed based on the statistics of the user to be predicted, which results in a problem of low prediction accuracy. However, the service demand influence factor determination system provided by the application can predict the service demand of the user to be predicted with higher accuracy.
Fig. 1 is a block diagram of a system 100 in a scenario in which a business requirement influencing factor determining method according to some embodiments of the present application is applied. For example, the system 100 may be an online transportation service platform for transportation services such as taxi cab, designated drive service, express, carpool, bus service, driver rental, or regular service, or any combination thereof. The system 100 may include one or more of a server 110, a network 120, a service requester terminal 130, a service provider terminal 140, and a database 150, and the server 110 may include a processor 112 therein that performs operations of instructions. The service requirement determining method provided by the embodiment of the present application may be applied to the server 110 in the system 100, and specifically, the processor 112 may execute the relevant operation instruction.
Fig. 2 illustrates a schematic diagram of exemplary hardware and software components of an electronic device 200 that may implement the server 110 of the present concepts according to some embodiments of the present application. For example, the processor 112 may be used on the electronic device 200 and to perform the functions herein.
The electronic device 200 may be a general purpose computer or a special purpose computer, both of which may be used to implement the business requirement determination method of the present application. Although only a single computer is shown, for convenience, the functions described herein may be implemented in a distributed fashion across multiple similar platforms to balance processing loads.
For example, the electronic device 200 may include a network port 210 connected to a network, one or more processors 220 for executing program instructions, a communication bus 230, and a different form of storage medium 240, such as a disk, ROM, or RAM, or any combination thereof. Illustratively, the computer platform may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application may be implemented in accordance with these program instructions. The electronic device 200 also includes an Input/Output (I/O) interface 250 between the computer and other Input/Output devices (e.g., keyboard, display screen).
For ease of illustration, only one processor is depicted in the electronic device 200. However, it should be noted that the electronic device 200 in the present application may also comprise a plurality of processors, and thus the steps performed by one processor described in the present application may also be performed by a plurality of processors in combination or individually. For example, if the processor of the electronic device 200 executes steps a and B, it should be understood that steps a and B may also be executed by two different processors together or separately in one processor. For example, the processor performs step a and the second processor performs step B, or the processor and the second processor perform steps a and B together.
Fig. 3 is a schematic diagram illustrating a method for determining business demand influencing factors provided by the embodiment of the present application, including S301 to S303.
S301: the method comprises the steps of obtaining order information of a plurality of orders of a user to be predicted within a preset historical time period.
In a specific implementation, the users to be predicted can be all users of the online car booking platform; in addition, because the number of users of the network car booking is very large, generally in the number of tens of millions or even billions, the service requirement of each user needs to be determined, and the calculation amount is very large, the users can be screened from all users of the network car booking platform based on certain conditions for more pertinence.
For example, the order sending frequency of each user on the online booking platform can be calculated periodically, and if the order sending frequency calculated in the period is reduced compared with the order sending frequency calculated in the last period and the quantity of the reduction reaches a preset percentage of the order sending frequency in the last period, the current user is determined as the user to be predicted. Here, the order issuing frequency refers to an action that a user wants to initiate a car booking order through a client used by the user.
Illustratively, the time interval between its last invoice and the current time may also be calculated for each user; and when the time interval is larger than a preset time interval threshold, determining the user as the user to be predicted.
In addition, other modes can be adopted to screen users with possibly reduced service requirements from all users of the network car booking platform as the users to be predicted.
The preset historical time period may be specifically set according to actual requirements, for example, the preset historical time period may be set to 3 months, 4 months, 5 months, half a year, and the like, and may also be set to 1 week, 2 weeks, 3 weeks, and the like. The preset history time period is, for example, a time period in which the current time is the latest time. The orders in the preset historical time period refer to orders in the preset historical time period.
In one implementation mode, the order information of the user feature vector sequence to be predicted is constructed, and the order information is the order information except the general user information in the order information; in another embodiment, the user features may also be in the order feature vector as general features of the order feature vector.
S302: and according to the order information of the orders, constructing an order feature vector sequence of the user to be predicted, wherein the order feature vector sequence comprises order feature vectors corresponding to the orders respectively.
Here, each order corresponds to order information, and an order feature vector corresponding to the order can be constructed according to the order information of each order. And after the order characteristic vectors of the orders are arranged according to the order time sequence, the order characteristic vector sequence is formed.
For example, for a user a to be predicted, if there are n orders within a preset historical time period, the order feature vector sequence generated for the user a to be predicted is as follows: x1, X2, X3, … … Xn.
Assuming that there are m elements in each order feature vector, the order feature vectors X1-Xn of the to-be-tested user are respectively expressed as:
X1＝(x11,x12,x13,x14,…,x1m)；
X2＝(x21,x22,x23,x24,…,x2m)；
……
Xn＝(xn1,xn2,xn3,xn4,…,xnm)。
specifically, referring to fig. 4, an embodiment of the present application further provides a specific method for constructing an order feature vector sequence of a user to be predicted according to order information of multiple orders, including:
s401: and for each order, determining a characteristic value of the order under a plurality of order characteristics according to the order information of the order.
Illustratively, taking the order information for constructing the user feature vector sequence to be predicted as the order information other than the general user information in the order information as an example, the order features include but are not limited to one or more of the following (1) to (7):
(1) the distance of the starting point of the order; here, before issuing an order through the online car appointment platform, the user inputs the starting place and the destination of the order from the client used by the user; the order starting point distance is the distance between the starting point and the destination.
Here, after the client transmits the start and destination input by the user to the server, the server considers that the client has a bubbling behavior.
(2) Queuing waiting time; here, the queuing waiting time may be estimated queuing waiting time determined by the server for the user to be predicted according to road condition information, transportation capacity information, car appointment information of other users, and the like when the user performs the bubbling behavior, or may be actual queuing waiting time between the user issuing the order and the real appointment of the user.
(3) Price of the order; here, the order price is a price at which the net appointment carries the user from the departure place to the destination.
(4) The order duration; here, the order duration may be one or more of a length of time taken for the net appointment vehicle to carry the user from the departure place to the destination, a total duration taken for the user to carry the user from the departure place to the destination from the departure place to the net appointment vehicle, and a total duration taken for the user to carry the user from the departure place to the destination from the net appointment vehicle to the net appointment vehicle.
(5) A driving receiving distance; here, the pick-up distance is a distance between the current position of the networked car appointment and the departure point of the user when the user appoints the networked car appointment.
(6) Pre-estimating; the pre-estimation is an estimated order price determined for a user according to a departure place and a destination sent by the client after the server detects that the bubble behavior occurs at the client.
(7) Service provider information; here, the service provider information may include one or more of: age, gender, rating level, service score, service equipment level, service equipment model, service equipment cleanliness, order completion rate, complained factors.
The evaluation level is an evaluation level of the service provider obtained after the user evaluates the service provider. And the service score is obtained after the user scores the service for the service provider. The service device class is obtained by classifying the service devices of the users according to a certain classification rule, for example, classifying the network appointment car according to the brand, classifying the network appointment car according to the purchase price section, classifying the network appointment car according to the vehicle type, classifying the network appointment car according to the use time, and the like. The cleanliness of the service equipment is generally obtained by scoring the cleanliness of the service equipment by the service provider. The order completion rate refers to a ratio between an order for which the service provider completes service and a received order; the complaint factor is obtained after the user complains about the service provider.
In the above order characteristics, both numerical characteristics and category characteristics are included. For numerical characteristics, the corresponding numerical values are directly used for representation. And a one-hot (one-hot) coding mode is used for corresponding class characteristics, namely each class characteristic corresponds to a vector consisting of 0 and 1, the number of classes corresponds to the dimension of the vector, namely one dimension of the vector corresponding to one class, when the preset operation behavior characteristics are of a certain class, the vector position corresponding to the class is 1, and all other parts are set to be 0.
S402: according to the characteristic values of the order under the characteristics of the orders, constructing an order characteristic vector corresponding to the order;
s403: and according to the order characteristic vectors corresponding to the orders respectively, constructing the order characteristic vector sequence according to the sequence of order generation time.
In another embodiment of the present application, after obtaining order information of a plurality of orders of a user to be predicted within a preset historical time period, the method further includes:
according to the user information of the user to be predicted, constructing a user characteristic vector of the user to be predicted;
at this time, the determining, based on the order feature vector sequence and the pre-trained prediction model, influence factor information of the service demand of the user to be predicted in a future preset time period includes:
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model.
Specifically, referring to fig. 5, an embodiment of the present application further provides a specific method for constructing a user feature vector of a user to be predicted, including:
s501: determining characteristic values of the user to be predicted under various user characteristics according to the user information;
s502: and constructing a user characteristic vector of the user to be predicted according to the characteristic values of the user to be predicted under the various user characteristics.
In a specific implementation, the user characteristics include: one or more of age, gender, occupation, and complaint factors of the service requester.
The complaint factor may be statistical information of a factor of a service requester complaint, for example, a plurality of complaint factors are preset, and the complaint factor may be represented as a vector formed by the number of times that the user complains for each complaint factor.
Illustratively, the preset complaint factors include 10 complaint factors a 1-a 10, and the number of complaints made by a certain user for the 10 complaint factors is as follows: 1. 0,5,0,1,15,0, then the complaint factors can be expressed as: (1,0,0,5,0,1,15,0,0,0).
In addition, in another embodiment of the present application, because part of the order features or the user features are high-dimensional sparse features, for example, when the order features include service device models, there are hundreds of models of vehicles serving as service devices according to different vehicle models, and because the order features are category features, the dimension of the order features obtained by the single hot method can also reach hundreds or even thousands, and only one element in the corresponding feature values is 1, and all other elements are 0. For such high-dimensional sparse features, dimension compression is required to be performed on the high-dimensional sparse features, so as to facilitate subsequent model calculation.
Specifically, for the order features, the order feature vector may be generated in the following manner:
for each order, determining characteristic values of the order under a plurality of order characteristics according to order information of the order;
detecting whether the dimension of a characteristic value corresponding to each order characteristic is larger than a first preset dimension or not;
if the dimension of the characteristic value corresponding to any order characteristic is larger than the first preset dimension, performing dimension compression on the characteristic value of any order characteristic to obtain a compressed characteristic value of any order characteristic;
and constructing an order feature vector corresponding to the order based on the compressed feature value corresponding to any order feature and the feature values corresponding to other order features.
For the user features, the user feature vector can be generated in the following manner:
determining characteristic values of a user to be predicted under various user characteristics according to user information of the user to be predicted;
detecting whether the dimension of the characteristic value corresponding to each user characteristic is larger than a second preset dimension or not;
if the dimension of the characteristic value corresponding to any user characteristic is larger than the second preset dimension, performing dimension compression on the characteristic value of any user characteristic to obtain a compressed characteristic value of any user characteristic;
and constructing a user feature vector corresponding to the user to be predicted based on the compressed feature value corresponding to any user feature and the feature values corresponding to other user features.
Here, the feature values may be dimension compressed in the following manner:
and acquiring a characteristic embedding matrix, and determining the product of the characteristic value and the characteristic embedding matrix as the compressed characteristic value.
Here, it should be noted that, for the same order feature, the feature embedding matrices of the order feature corresponding to different orders are the same; and aiming at the same user characteristic, the characteristic embedding matrixes of the user characteristic corresponding to different users to be predicted are the same.
The feature embedding matrix is typically initialized to random values and then trained with model parameters during the predictive model training process.
After the step S302 is performed, after the order feature vector sequence of the user to be predicted is constructed, the service demand influence factor determining scheme provided in the embodiment of the present application further includes:
s303: and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model.
One is as follows: in some embodiments, if the influence factor information of the service demand of the user to be predicted in the future preset time period is determined only through the constructed order feature vector sequence and the pre-trained prediction model, the constructed feature vector sequence is input into the pre-trained prediction model, and the influence factor information of the service demand of the user to be predicted in the future preset time period is determined.
Specifically, referring to fig. 6, the following method may be adopted to determine the information of the influence factor of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence and the pre-trained prediction model:
s601: inputting the order characteristic vector sequence into the prediction model to obtain a prediction result corresponding to the order characteristic vector sequence;
s602: determining a second contribution value corresponding to each order feature based on a prediction result corresponding to the order feature vector sequence, elements in each feature vector sequence and reference values corresponding to the elements in the feature vector sequence;
s603: and determining influence factor information of the user to be predicted in the order characteristic dimension based on the second contribution value.
Illustratively, any order characteristics x that are ultimately determinediCorresponding second contribution IntegratedGradsi(x) Satisfies the following formula:
Figure BDA0001895419220000271
wherein x isiRepresenting the ith element in the order feature vector; x'iRepresenting a reference value corresponding to the ith element in the order feature vector; f represents a prediction function corresponding to the prediction model; x' represents a feature vector composed of reference values of the features of each order; x represents an order feature vector in the sequence of order feature vectors.
After the second contribution value corresponding to each order feature is determined, influence factor information of the user to be predicted in the order feature dimension can be determined based on the second contribution value corresponding to each feature.
Referring to fig. 7, an embodiment of the present application further provides a specific method for determining influence factor information of the user to be predicted in the order feature dimension based on the second contribution value, where the specific method includes:
s701: determining at least one target order characteristic from the order characteristics according to the magnitude of the second contribution value;
s702: and determining the influence factor information of the user to be predicted in order feature dimensions based on the influence factor classification corresponding to the target order features.
In the embodiment of the present application, the classification of the influence factors of the order feature dimension includes: the price is sensitive, the vehicle is less sensitive, the waiting time is sensitive, the driver service is sensitive, the potential safety hazard is sensitive, and the vehicle type and vehicle condition are sensitive.
Here, according to the order from the largest to the smallest of the second contribution values, a preset number of order features may be selected from the order features as target order features, and influence factor information that influences the service demand of the user to be predicted is determined according to the influence factor classification to which each target order feature belongs.
For example, one of the target order characteristics is that the attribution of the impact factor classification includes: and if the price is sensitive, determining the influence factor information of the service demand of the user to be predicted comprises the following steps: the price is sensitive.
For another example, there are two target order characteristics, and the attributive influencing factors are classified as: and if the driver service is sensitive and the vehicle type and vehicle condition are sensitive, determining the influence factor information of the user service demand to be predicted comprises the following steps: driver service sensitivity and vehicle type condition sensitivity.
It should be noted that the prediction model provided in the embodiment of the present application predicts the service requirement of the user instead of directly predicting the influence factor information of the service requirement of the user. After the prediction model is determined, the determination of the influence factor information of the user service demand is realized according to the method corresponding to the above fig. 6.
In this case, referring to fig. 8, a pre-trained predictive model may be obtained in the following manner:
s801: the method comprises the steps of obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and actual business demand results of the sample users in a forecast historical time period.
Here, the sample order information of the plurality of sample orders of the sample user in the target historical time period is similar to the order information of the plurality of orders of the user to be predicted in the preset historical time period, and is not repeated herein.
S802: for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders.
Here, the manner for generating the sample order feature vector sequence of each sample user is similar to the manner for generating the order feature vector sequence of the user to be predicted, and is not described herein again.
S803: and determining a service demand prediction result of each sample user in a prediction history time period based on the sample order feature vector sequence of each sample user and a basic prediction model.
Here, the basic prediction model includes, for example: recurrent Neural Networks (RNN).
S804: and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the business demand prediction model.
Here, the process of training the basic model according to the business demand prediction result of the sample user and the corresponding actual business demand result is a process of inputting the feature vector sequence of each sample user into the basic prediction model, so that the basic model performs feature learning on the feature vector sequence, outputs the business demand prediction result, and adjusts the parameters of the basic prediction model according to the business demand prediction result and the corresponding actual business demand result, so that the business demand prediction result output by the basic prediction model and the actual business demand result tend to be consistent constantly.
Specifically, the basic prediction model may be trained according to the service demand prediction result of each sample user and the corresponding actual service demand result in the following manner:
and selecting one sample order feature vector from the sample order feature vector sequence as the current sample order feature vector according to the sequence of order generation time corresponding to the sample order feature vector in each sample order feature vector sequence.
And acquiring an intermediate characteristic vector output by a target characteristic extraction layer of the basic prediction model for the characteristic vector of the previous sample order after the characteristic vector of the previous sample order is input into the basic prediction model.
And inputting the current sample order characteristic vector and the intermediate characteristic vector into a basic prediction model to obtain a business demand prediction result corresponding to the sample order characteristic vector.
And adjusting parameters of the basic prediction model based on the obtained service demand prediction result and the corresponding actual service demand result, and returning to the step of selecting a sample order feature vector from the sample order feature vector sequence and inputting the sample order feature vector into the basic prediction model based on the adjusted basic prediction model until a training cut-off condition is met.
Here, the training stopping condition may be set according to actual needs, and may be, for example: and performing preset rounds of training on the basic prediction model by using the plurality of sample order characteristic vector sequences, and taking the basic prediction model obtained in the last round of training as a business demand prediction model. In each round of training, the model is trained once by using each sample order feature vector in the plurality of sample order feature vector sequences in sequence.
Alternatively, the training cut-off condition may also be: verifying the basic prediction model obtained in the current round by using a test set; and if the cross entropy loss is not more than the number of the test data with the preset cross entropy loss in the test set and occupies the percentage of the total number of the test data in the test set, and the percentage is more than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as a business demand prediction model.
Alternatively, the training cut-off condition may also be: comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the corresponding sample user in the previous round in sequence; and if the cross entropy loss of the sample users in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample users in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as a business demand prediction model.
Illustratively, the base prediction model may be trained in the following manner: and constructing an initial basic prediction model, and performing initial assignment on each parameter of the initial basic prediction model.
The sample users are determined to include s sample users of A1-As.
For the ith sample user Ai, i e [1, s ] therein:
for the ith sample user Ai, if there are n sample orders of the sample user Ai in the target historical time period and there are m corresponding order features, the sample order feature vector sequence generated for the sample user Ai is: xi1、Xi2、Xi3、……、Xin。
Performing a current round of training on the basic prediction model by using the feature vector sequence of the sample user Ai:
when the sample order feature vector sequence of the sample user A is used for carrying out the training of the basic prediction model in the current round: firstly, a sample order feature vector X is obtainedi1 into the basic prediction model:
at this time, if the sample order feature vector X is being usedi1 before input into the base prediction model, if there is no other sample order feature vector to train it, then only the initial assignment based on the base prediction model, and the sample order feature vector Xi1, obtaining the sample order feature vector Xi1, predicting the corresponding business demand. Predicting result based on business demand and sample order characteristic vector Xi1 actual business demand results, parameters to the basic prediction modelThe number is adjusted.
If the sample order feature vector X is in usei1 before being input into the basic prediction model, other sample order feature vectors are used for training the basic prediction model, then an intermediate feature vector Z0 extracted for the previous sample order feature vector by a target feature extraction layer of the basic prediction model after the previous sample order feature vector is input into the basic prediction model is obtained, and the intermediate feature vector Z0 is used as a sample order feature vector X used this timei1 parameters for training the basic prediction model, inputting the parameters into the basic prediction model, and based on the intermediate feature vector Z0 and the sample order feature vector Xi1, obtaining the sample order feature vector Xi1, predicting the corresponding business demand. Predicting result based on business demand and sample order characteristic vector Xi1, adjusting parameters of the basic prediction model according to the actual service demand result.
And obtaining the characteristic vector X of the order form of the sample from the target extraction layer of the basic prediction modeli1 extracted intermediate feature vector Z1.
Here, the prediction result is based on the business demand and the sample order feature vector Xi1, when the parameters of the basic prediction model are adjusted, the sample order feature vector X can be adjustedi1, comparing a corresponding service demand prediction result with an actual service demand result; and adjusting the parameters of the basic prediction model under the condition that the difference value between the two is smaller than a preset difference value threshold value.
Based on the business demand prediction result and the sample order feature vector Xi1 when the parameters of the basic prediction model are adjusted according to the actual business demand result corresponding to the basic prediction model, the characteristic vector X of the sample order can be usedi1, calculating cross entropy loss according to a service demand prediction result and an actual service demand result corresponding to the service demand prediction result; and adjusting parameters of the basic prediction model according to the cross entropy loss.
In some embodiments, if the cross entropy loss is large, the adjustment step size of the parameter of the basic prediction model is larger than that of the parameter of the basic prediction model when the cross entropy loss is small.
Then sample order feature vector Xi2 and the intermediate feature vector Z1, to the sample order feature vector Xi1 training the obtained basic prediction model to obtain a sample order characteristic vector Xi2, business demand forecasting result.
Based on sample order feature vector Xi2, continuously adjusting parameters of the basic prediction model according to the service demand prediction result and the actual service demand result.
And obtaining the characteristic vector X of the order form of the sample from the target extraction layer of the basic prediction modeli2 extracted intermediate feature vector Z2.
Respectively characterizing the sample to Xi3、……、Xin, training the basic prediction model, and finishing the feature vector sequence based on the current sample order as follows: xi1、Xi2、Xi3、……、XiAnd n pairs of basic prediction models are trained.
And training the basic prediction model based on the sample order characteristic vector sequence corresponding to each sample user in sequence, and finishing the training of the basic prediction model in the current round.
In some embodiments, the underlying predictive model is a recurrent neural network, including an input layer, a target feature extraction layer, and an output layer. Inputting the current sample order feature vector and the intermediate feature vector into a basic prediction model to obtain a service demand prediction result corresponding to the sample order feature vector, wherein the following method can be adopted: inputting the current sample order feature vector into an input layer, and extracting an intermediate feature vector for the sample order feature vector by using the input layer; and inputting the intermediate characteristic vector for sample order characteristic vector extraction by the intermediate characteristic vector input and target characteristic extraction layer of the basic prediction model, performing weighted summation on the intermediate characteristic vector and the intermediate characteristic vector extracted for the sample order characteristic vector by the target characteristic extraction layer of the basic prediction model to obtain an intermediate characteristic vector extracted for the current sample order characteristic vector by the target characteristic extraction layer of the basic prediction model, and inputting the intermediate characteristic vector extracted for the current sample order characteristic vector by the target characteristic extraction layer of the basic prediction model to an output layer to obtain a service demand prediction result.
The second step is as follows:
if the influence factor information of the service demand of the user to be predicted in the future preset time period is obtained by constructing the order feature vector sequence and the user feature vector and inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, in this case, the method for determining the influence factor information of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence and the pre-trained prediction model provided in the embodiments corresponding to fig. 6 and 7 may be adopted.
In another embodiment, the following method for determining the influence factor information of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence, the user feature vector and the pre-trained prediction model, which are provided in the embodiment corresponding to fig. 9, may also be used, and includes:
s901: inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, obtaining attention distribution weights corresponding to the order feature vectors from a target network layer of the prediction model, and taking the attention distribution weights as first contribution values of orders corresponding to the order feature vectors to silence the user to be predicted;
s902: and determining influence factor information of the user to be predicted in the order dimension based on the first contribution value.
Here, it should be noted that the influence factor information of the order dimension is a target order having an influence on the business demand of the user to be predicted.
Specifically, at least one target order feature vector may be determined from each order feature vector according to the magnitude of the attention distribution weight corresponding to the order feature vector, and an order corresponding to the determined target order feature vector is determined as a target order having a large influence on the business demand of the user to be predicted.
In this case, referring to fig. 10, a pre-trained predictive model may be obtained in the following manner:
s1001: obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and an actual business demand result of each sample user in a prediction historical time period; and acquiring user information of each sample user.
Here, the sample order information of the plurality of sample orders is similar to the order information of the user to be predicted, and the user information of the sample user is similar to the user information of the user to be predicted, which is not described herein again.
S1002: for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders; and generating the user feature vector of the sample user according to the user information of the sample user.
Here, the generation manner of the sample order feature vector sequence and the user feature vector of the sample user is similar to the acquisition manner of the order feature vector sequence and the user feature vector of the user to be predicted, and is not described herein again.
S1003: and inputting the sample order characteristic vector sequence and the user characteristic vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user.
S1004: and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the prediction model.
In specific implementation, fig. 11 shows a specific manner diagram of inputting a sample order feature vector sequence and a user feature vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user according to the embodiment of the present application, and fig. 12 shows a structural diagram of a basic prediction network according to the embodiment of the present application. Referring to fig. 11 and 12, inputting the sample order feature vector sequence and the user feature vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user, including:
s1101: and aiming at each sample user, inputting the sample order feature vector in the sample order feature vector sequence of the sample user into a first neural network, and acquiring a middle feature vector corresponding to each sample order feature vector.
Here, the sample order feature vector sequence includes a plurality of sample order feature vectors, and in order to establish a relationship between the sample order feature vectors, each sample order feature vector may be obtained in the following manner and input into the first neural network, so as to obtain an intermediate feature vector corresponding to each sample order feature vector:
selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors of the sample user;
inputting the obtained characteristic vector of the previous sample order into the first neural network, wherein a target characteristic extraction layer of the first neural network is an intermediate characteristic vector output by the characteristic vector of the previous sample order;
inputting the current sample order feature vector and the intermediate feature vector of the previous sample order feature vector into the first neural network, and acquiring the intermediate feature vector corresponding to the current sample order feature vector;
and returning a sample order feature vector sequence aiming at the sample user, and selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors until the intermediate feature vectors of all the sample order feature vectors are extracted.
S1102: and inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and acquiring attention distribution weights corresponding to each intermediate feature vector.
Here, the attention assignment weight corresponding to each intermediate feature vector may be acquired in the following manner:
and for each intermediate characteristic vector, splicing the intermediate characteristic vector with the user characteristic vector to generate a spliced vector corresponding to the intermediate characteristic vector. Inputting the splicing vector into the second neural network to obtain the matching degree corresponding to the splicing vector; and performing activation operation on the matching degree based on a preset activation function to obtain the attention distribution weight corresponding to the intermediate feature vector.
Here, it should be noted that the network layer that performs activation operation on the matching degree based on a preset activation function is a target extraction layer in the embodiment of the present application.
Specifically, the second neural network is a fully-connected network with at least one layer, the relationship between each element in the splicing vector can be established through the second neural network, the determined matching degree corresponding to the splicing vector is the matching degree between the intermediate feature vector forming the splicing vector and the user feature vector, and the matching degree can be regarded as the matching degree between the sample order and the user because the intermediate feature vector is obtained based on the sample order feature vector. And then, performing activation operation on the matching degrees corresponding to the splicing vectors by using the same activation function, and distributing weights to the attention corresponding to the intermediate feature vectors.
S1103: generating a fused feature vector based on each of the intermediate feature vectors and the attention-assigned weights for the sample user.
Here, the fused feature vector may be generated by performing weighted summation on each intermediate feature vector. And the weights obtained by performing weighted summation on the intermediate feature vectors are the attention distribution weights corresponding to the intermediate feature vectors respectively.
S1104: and inputting the fusion characteristic vector of the sample user into a third neural network, extracting a target characteristic vector for the fusion characteristic vector, and inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
For example, referring to fig. 12, the sample order feature vector sequence of the sample user includes 10 sample order feature vectors from X1 to X10, which are sequentially input to the first neural network, and the user feature vector of the sample user is T. Wherein:
inputting X1 into the first neural network to obtain an intermediate feature vector h 1;
inputting h1 and X2 into the first neural network to obtain an intermediate feature vector h 2;
……
inputting h9 and X10 into the first neural network to obtain an intermediate feature vector h 10;
then inputting h 1-h 10 and T into a second neural network to obtain attention distribution weights w 1-w 10 corresponding to each intermediate feature vector;
and carrying out weighted summation on H1-H10 according to the attention distribution weights w 1-w 10 to form a fusion feature vector H.
And inputting the fusion characteristic vector H into a third neural network to obtain a service demand prediction result P of the sample user.
Illustratively, the third neural network is a neural network including a plurality of fully connected layers, and through the third neural network, the connection between each element in the fused feature vector can be established, and the target feature vector capable of representing the service requirement of the user can be extracted. And then inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
In the embodiment of the application, the service demand of the user is characterized by the silence probability of the user, so that the service demand prediction result can be expressed as the silence probability prediction result of the user.
After obtaining the service demand prediction results of the sample users according to the basic prediction model, the basic prediction model can be trained according to the service demand prediction results of the sample users and the corresponding actual service demand results in the following way:
taking any sample user in sample users who have not completed training in the current round as a current sample user, and determining the cross entropy loss of the current sample user in the current round according to the business demand prediction result of the current sample user and the corresponding actual business demand result;
adjusting parameters of the basic prediction model according to the cross entropy loss of the current sample user in the current round;
and taking the current sample user as a sample user completing the training in the current round, and returning to the step of determining the cross entropy loss of the current sample user in the current round until all sample users complete the training in the current round.
Here, the process of training the basic prediction model according to the service demand prediction result of each sample user and the corresponding actual service demand result is similar to the process of S804, and is not repeated here.
Similar to the above S804, after the current round of training of the basic prediction model is completed, the embodiment of the present application may further detect whether to stop training by any one of the following three ways:
(1) detecting whether the number of the current wheel reaches a preset number; if so, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model.
When the model is specifically realized, a preset number of training rounds is preset during model training, if the fact that the number of training rounds reaches the preset number of training rounds is detected, the training of the basic prediction model is stopped, and the basic prediction model obtained in the last round of training is used as the prediction model.
(2) Verifying the basic prediction model obtained in the current round by using a test set; if the cross entropy loss is not greater than the number of the test data of the preset cross entropy loss threshold value in the test set, the percentage of the total number of the test data in the test set is occupied, and the percentage is greater than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
in the process of model training, the value of cross entropy loss needs to be gradually reduced, so that when the basic prediction model obtained in the current round is verified by using a test set, if the number of pieces of test data, in the test set, of which the cross entropy loss is not greater than a preset cross entropy loss threshold reaches a certain preset proportion, such as 90% and 95%, the training of the basic prediction model is stopped, and the basic prediction model obtained in the last round of training is used as the prediction model.
(3) Comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the sample user corresponding to the previous round in sequence; and if the cross entropy loss of the sample user in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample user in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as the prediction model.
In the process of model training, the value of cross entropy loss needs to be gradually reduced, and a basic prediction model obtained when the value of cross entropy loss is minimum is used as a prediction model.
After the prediction model is obtained, the service demand prediction result of the user to be predicted in the future preset time period can be predicted based on the prediction model.
The processing mode of the prediction model on the order feature vector sequence is similar to that of the basic prediction model on the sample order feature vector in the model training process, and is not described herein again.
According to the method and the device, the order information of a plurality of orders of the user to be predicted in the preset historical time period is obtained, the order characteristic vector sequence of the user to be predicted is built, and the business demand influence factor of the user to be predicted in the future preset time period is determined with higher accuracy according to the order characteristic vector sequence and a pre-trained business demand prediction model.
Fig. 13 is a block diagram illustrating a business requirement influencing factor determining apparatus according to some embodiments of the present application, where the functions implemented by the business requirement influencing factor determining apparatus correspond to the steps executed by the above method. The apparatus may be understood as the server or the processor of the server, or may be understood as a component that is independent of the server or the processor and implements the functions of the present application under the control of the server, as shown in fig. 13, the service requirement influencing factor determining apparatus includes:
the system comprises an acquisition module, a prediction module and a prediction module, wherein the acquisition module is used for acquiring order information of a plurality of orders of a user to be predicted within a preset historical time period;
the construction module is used for constructing an order feature vector sequence of the user to be predicted according to the order information of the orders, wherein the order feature vector sequence comprises order feature vectors corresponding to the orders respectively;
and the determining module is used for determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model.
In an optional implementation manner, the constructing module is configured to construct the order feature vector sequence of the user to be predicted according to the order information of the multiple orders by using the following steps, including:
for each order, determining characteristic values of the order under a plurality of order characteristics according to order information of the order;
according to the characteristic values of the order under the characteristics of the orders, constructing an order characteristic vector corresponding to the order;
and according to the order characteristic vectors corresponding to the orders respectively, constructing the order characteristic vector sequence according to the sequence of order generation time.
In an optional implementation manner, the determining module is configured to determine, based on the order feature vector sequence and a pre-trained prediction model, influence factor information of a business demand of the user to be predicted in a future preset time period by using the following steps:
inputting the order characteristic vector sequence into the prediction model to obtain a prediction result corresponding to the order characteristic vector sequence;
determining a second contribution value corresponding to each order feature based on a prediction result corresponding to the order feature vector sequence, elements in each feature vector sequence and reference values corresponding to the elements in the feature vector sequence;
and determining influence factor information of the user to be predicted in the order characteristic dimension based on the second contribution value.
In an optional embodiment, the determining module is configured to determine, based on the second contribution value, influence factor information of the user to be predicted in an order feature dimension by using the following steps, including:
determining at least one target order characteristic from the order characteristics according to the magnitude of the second contribution value;
and determining the influence factor information of the user to be predicted in order feature dimensions based on the influence factor classification corresponding to the target order features.
In an alternative embodiment, the classification of the influencing factors includes: the system is sensitive to one or more of price, few vehicles, waiting time, driver service, potential safety hazard and vehicle type and vehicle condition.
In an alternative embodiment, the second contribution value integrated gradsi(x) Satisfies the following formula:
Figure BDA0001895419220000401
wherein x isiRepresenting the ith element in the order feature vector; x is the number ofi' represents a reference value corresponding to the ith element in the order feature vector; f represents a prediction function corresponding to the prediction model; x' represents a feature vector composed of reference values of the features of each order; x represents an order feature vector in the sequence of order feature vectors.
In an optional implementation, the building module is further configured to:
according to the user information of the user to be predicted, constructing a user characteristic vector of the user to be predicted;
the determining module is used for determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model by adopting the following steps:
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model.
In an optional implementation manner, the determining module is configured to determine, based on the order feature vector sequence, the user feature vector input, and the pre-trained prediction model, influence factor information of a business demand of the user to be predicted in a future preset time period by using the following steps:
inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, obtaining attention distribution weights corresponding to the order feature vectors from a target network layer of the prediction model, and taking the attention distribution weights as first contribution values of orders corresponding to the order feature vectors to silence the user to be predicted;
and determining influence factor information of the user to be predicted in the order dimension based on the first contribution value.
In an optional implementation manner, the building module is configured to build a user feature vector of the user to be predicted according to the user information by adopting the following steps:
determining characteristic values of the user to be predicted under various user characteristics according to the user information;
and constructing a user characteristic vector of the user to be predicted according to the characteristic values of the user to be predicted under the various user characteristics.
In an alternative embodiment, the user features include a plurality of the following features:
age, gender, occupation, complaint factors of the service requester.
In an alternative embodiment, the order characteristics include a plurality of the following characteristics:
the order starting point distance, the queuing waiting time, the order price, the order duration, the driving receiving distance, the pre-evaluation value and the service provider information.
In an optional embodiment, the service provider information comprises at least one of the following information:
age, gender, rating level, service score, service equipment level, service equipment cleanliness, order completion rate, complained factors.
In an alternative embodiment, the method further comprises: a training module for training the predictive model in the following manner:
obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and an actual business demand result of each sample user in a prediction historical time period;
for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders;
determining a service demand prediction result of each sample user in a prediction history time period based on the sample order feature vector sequence of each sample user and a basic prediction model;
and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the prediction model.
In an optional embodiment, the training module is further configured to, after obtaining sample order information of a plurality of sample orders of a plurality of sample users within a target historical time period, for each sample user, generate a user feature vector of the sample user according to the user information of the sample user;
the training module is used for determining a business demand prediction result of each sample user based on the sample order feature vector sequence and the basic prediction model of each sample user by adopting the following steps:
and inputting the sample order characteristic vector sequence and the user characteristic vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user.
In an optional implementation manner, the training module is configured to input the sample order feature vector sequence and the user feature vector of each sample user into a basic prediction model to obtain a business demand prediction result of each sample user by using the following steps:
for each sample user, inputting a sample order feature vector in a sample order feature vector sequence of the sample user into a first neural network, and acquiring a middle feature vector corresponding to each order feature vector;
inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and acquiring attention distribution weights corresponding to each intermediate feature vector;
generating a fused feature vector based on each of the intermediate feature vectors of the sample user and the attention distribution weight;
and inputting the fusion characteristic vector of the sample user into a third neural network, extracting a target characteristic vector for the fusion characteristic vector, and inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
In an optional implementation manner, the training module is configured to, for each sample user, input a sample order feature vector in a sample order feature vector sequence of the sample user into the first neural network, and obtain an intermediate feature vector corresponding to each order feature vector, where:
selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors of the sample user;
inputting the obtained characteristic vector of the previous sample order into the first neural network, wherein a target characteristic extraction layer of the first neural network is an intermediate characteristic vector output by the characteristic vector of the previous sample order;
inputting the current sample order feature vector and the intermediate feature vector of the previous sample order feature vector into the first neural network, and acquiring the intermediate feature vector corresponding to the current sample order feature vector;
and returning a sample order feature vector sequence aiming at the sample user, and selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors until the intermediate feature vectors of all the sample order feature vectors are extracted.
In an alternative embodiment, the training module is configured to input each intermediate feature vector of the sample user and the user feature vector to a second neural network, and obtain an attention allocation weight corresponding to each intermediate feature vector by:
for each intermediate feature vector, splicing the intermediate feature vector with the user feature vector to generate a spliced vector corresponding to the intermediate feature vector;
inputting the splicing vector into the second neural network to obtain the matching degree corresponding to the splicing vector;
and performing activation operation on the matching degree based on a preset activation function to obtain the attention distribution weight corresponding to the intermediate feature vector.
In an optional embodiment, the training module is configured to generate a fused feature vector based on each of the intermediate feature vectors and the attention assignment weight of the sample user by:
and carrying out weighted summation on the intermediate feature vectors according to the attention distribution weights corresponding to the intermediate feature vectors to generate the fusion feature vectors.
In an optional implementation manner, the training module is configured to train the basic prediction model according to the service demand prediction result of each sample user and the corresponding actual service demand result by using the following steps:
taking any sample user in sample users who have not completed training in the current round as a current sample user, and determining the cross entropy loss of the current sample user in the current round according to the business demand prediction result of the current sample user and the corresponding actual business demand result;
adjusting parameters of the basic prediction model according to the cross entropy loss of the current sample user in the current round;
and taking the current sample user as a sample user completing the training in the current round, and returning to the step of determining the cross entropy loss of the current sample user in the current round until all sample users complete the training in the current round.
In an optional embodiment, the training module is further configured to detect whether a current round reaches a preset number of rounds after completing the current round of training on the basic prediction model; if so, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
verifying the basic prediction model obtained in the current round by using a test set; if the cross entropy loss is not greater than the number of the test data of the preset cross entropy loss threshold value in the test set, the percentage of the total number of the test data in the test set is occupied, and the percentage is greater than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the sample user corresponding to the previous round in sequence; and if the cross entropy loss of the sample user in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample user in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as the prediction model.
In an optional implementation, the training module is configured to generate a sample order feature vector by:
for each sample order, determining characteristic values of the sample order under a plurality of order characteristics according to order information of the sample order;
detecting whether the dimension of the characteristic value corresponding to each order characteristic is larger than a preset dimension or not;
if the dimension of the characteristic value corresponding to any order characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any order characteristic to obtain a compressed characteristic value of any order characteristic;
and constructing the sample order feature vector based on the compressed feature value corresponding to any order feature and the feature values corresponding to other order features.
In an optional implementation, the training module is configured to generate a user order feature vector by:
for each sample user, determining a characteristic value of the sample user under a plurality of user characteristics according to the user information of the sample user;
detecting whether the dimensionality of the characteristic value corresponding to each user characteristic is larger than a preset dimensionality;
if the dimension of the characteristic value corresponding to any user characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any user characteristic to obtain a compressed characteristic value of any user characteristic;
and constructing the user feature vector based on the compressed feature value corresponding to any user feature and the feature values corresponding to other user features.
In an alternative embodiment, the training module is configured to perform dimension compression on the feature values by:
acquiring a characteristic embedding matrix;
determining a product of the eigenvalue and the feature embedding matrix as the compressed eigenvalue.
According to the method and the device, the order information of a plurality of orders of the user to be predicted in the preset historical time period is obtained, the order characteristic vector sequence of the user to be predicted is built, and the business demand influence factors of the user to be predicted in the future preset time period are determined with higher accuracy according to the order characteristic vector sequence and a pre-trained business demand prediction model, so that resource allocation is better performed based on the influence factors, and the service quality is improved.
The modules may be connected or in communication with each other via a wired or wireless connection. The wired connection may include a metal cable, an optical cable, a hybrid cable, etc., or any combination thereof. The wireless connection may comprise a connection over a LAN, WAN, bluetooth, ZigBee, NFC, or the like, or any combination thereof. Two or more modules may be combined into a single module, and any one module may be divided into two or more units.
As shown in fig. 2, an embodiment of the present application further provides an electronic device, including: the system comprises a processor 220, a storage medium and a bus 230, wherein the storage medium stores machine-readable instructions executable by the processor 220, when an electronic device runs, the processor 220 communicates with the storage medium through the bus 230, and the processor 220 executes the machine-readable instructions to execute the steps of the business requirement influencing factor determining method provided by the embodiment of the application.
An embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the service demand influence factor determination method as provided in the embodiment of the present application are executed.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the system and the apparatus described above may refer to corresponding processes in the method embodiments, and are not described in detail in this application. In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is merely a logical division, and there may be other divisions in actual implementation, and for example, a plurality of modules or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or modules through some communication interfaces, and may be in an electrical, mechanical or other form.
The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (48)

1. A method for determining business demand influence factors is characterized by comprising the following steps:
the method comprises the steps of obtaining order information of a plurality of orders of a user to be predicted within a preset historical time period;
according to the order information of the orders, constructing an order feature vector sequence of the user to be predicted, wherein the order feature vector sequence comprises order feature vectors corresponding to the orders;
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model.
2. The method according to claim 1, wherein constructing the order feature vector sequence of the user to be predicted according to the order information of the plurality of orders comprises:
for each order, determining characteristic values of the order under a plurality of order characteristics according to order information of the order;
according to the characteristic values of the order under the characteristics of the orders, constructing an order characteristic vector corresponding to the order;
and according to the order characteristic vectors corresponding to the orders respectively, constructing the order characteristic vector sequence according to the sequence of order generation time.
3. The method according to claim 2, wherein the determining of the influence factor information of the business demand of the user to be predicted in the future preset time period based on the order feature vector sequence and a pre-trained prediction model comprises:
inputting the order characteristic vector sequence into the prediction model to obtain a prediction result corresponding to the order characteristic vector sequence;
determining a second contribution value corresponding to each order feature based on a prediction result corresponding to the order feature vector sequence, elements in each feature vector sequence and reference values corresponding to the elements in the feature vector sequence;
and determining influence factor information of the user to be predicted in the order characteristic dimension based on the second contribution value.
4. The method according to claim 3, wherein the determining influence factor information of the user to be predicted in the order feature dimension based on the second contribution value comprises:
determining at least one target order characteristic from the order characteristics according to the magnitude of the second contribution value;
and determining the influence factor information of the user to be predicted in order feature dimensions based on the influence factor classification corresponding to the target order features.
5. The method of claim 4, wherein the classification of influencing factors comprises: the system is sensitive to one or more of price, few vehicles, waiting time, driver service, potential safety hazard and vehicle type and vehicle condition.
6. The method according to claim 3, wherein the second contribution value IntegratedGradsi(x) Satisfies the following formula:
Figure FDA0001895419210000021
wherein x isiRepresenting the ith element in the order feature vector; x'iRepresenting a reference value corresponding to the ith element in the order feature vector; f represents a prediction function corresponding to the prediction model; x' represents a feature vector composed of reference values of the features of each order; x represents an order feature vector in the sequence of order feature vectors.
7. The method of claim 1, wherein after obtaining order information of a plurality of orders of the user to be predicted within a preset historical time period, the method further comprises:
according to the user information of the user to be predicted, constructing a user characteristic vector of the user to be predicted;
the determining of the influence factor information of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence and the pre-trained prediction model comprises the following steps:
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model.
8. The method according to claim 7, wherein the determining of the influence factor information of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model comprises:
inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, obtaining attention distribution weights corresponding to the order feature vectors from a target network layer of the prediction model, and taking the attention distribution weights as first contribution values of orders corresponding to the order feature vectors to silence the user to be predicted;
and determining influence factor information of the user to be predicted in the order dimension based on the first contribution value.
9. The method according to claim 7, wherein the constructing the user feature vector of the user to be predicted according to the user information comprises:
determining characteristic values of the user to be predicted under various user characteristics according to the user information;
and constructing a user characteristic vector of the user to be predicted according to the characteristic values of the user to be predicted under the various user characteristics.
10. The method of claim 7, wherein the user characteristics include a plurality of the following characteristics:
age, gender, occupation, complaint factors of the service requester.
11. The method of claim 2, wherein the order characteristics include a plurality of the following characteristics:
the order starting point distance, the queuing waiting time, the order price, the order duration, the driving receiving distance, the pre-evaluation value and the service provider information.
12. The method of claim 11, wherein the service provider information comprises at least one of:
age, gender, rating level, service score, service equipment level, service equipment cleanliness, order completion rate, complained factors.
13. The method of claim 1, wherein the predictive model is trained by:
obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and an actual business demand result of each sample user in a prediction historical time period;
for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders;
determining a service demand prediction result of each sample user in a prediction history time period based on the sample order feature vector sequence of each sample user and a basic prediction model;
and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the prediction model.
14. The method of claim 13, wherein after obtaining sample order information for a plurality of sample orders for a plurality of sample users over a target historical time period, further comprising:
for each sample user, generating a user feature vector of the sample user according to the user information of the sample user;
the determining a service demand prediction result of each sample user based on the sample order feature vector sequence and the basic prediction model of each sample user comprises:
and inputting the sample order characteristic vector sequence and the user characteristic vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user.
15. The method according to claim 14, wherein the inputting the sample order feature vector sequence and the user feature vector of each sample user into a basic prediction model to obtain the prediction result of the business demand of each sample user comprises:
for each sample user, inputting a sample order feature vector in a sample order feature vector sequence of the sample user into a first neural network, and acquiring a middle feature vector corresponding to each order feature vector;
inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and acquiring attention distribution weights corresponding to each intermediate feature vector;
generating a fused feature vector based on each of the intermediate feature vectors of the sample user and the attention distribution weight;
and inputting the fusion characteristic vector of the sample user into a third neural network, extracting a target characteristic vector for the fusion characteristic vector, and inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
16. The method of claim 15, wherein the inputting, for each sample user, a sample order feature vector in a sample order feature vector sequence of the sample user into the first neural network, and obtaining an intermediate feature vector corresponding to each order feature vector comprises:
selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors of the sample user;
inputting the obtained characteristic vector of the previous sample order into the first neural network, wherein a target characteristic extraction layer of the first neural network is an intermediate characteristic vector output by the characteristic vector of the previous sample order;
inputting the current sample order feature vector and the intermediate feature vector of the previous sample order feature vector into the first neural network, and acquiring the intermediate feature vector corresponding to the current sample order feature vector;
and returning a sample order feature vector sequence aiming at the sample user, and selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors until the intermediate feature vectors of all the sample order feature vectors are extracted.
17. The method of claim 15, wherein inputting each intermediate feature vector of the sample user and the user feature vector to a second neural network, obtaining attention assignment weights corresponding to each intermediate feature vector, comprises:
for each intermediate feature vector, splicing the intermediate feature vector with the user feature vector to generate a spliced vector corresponding to the intermediate feature vector;
inputting the splicing vector into the second neural network to obtain the matching degree corresponding to the splicing vector;
and performing activation operation on the matching degree based on a preset activation function to obtain the attention distribution weight corresponding to the intermediate feature vector.
18. The method of claim 15, wherein generating a fused feature vector based on the respective intermediate feature vectors of the sample users and the attention-assigning weights comprises:
and carrying out weighted summation on the intermediate feature vectors according to the attention distribution weights corresponding to the intermediate feature vectors to generate the fusion feature vectors.
19. The method of claim 13, wherein the training the basic prediction model according to the traffic demand prediction result of each sample user and the corresponding actual traffic demand result comprises:
taking any sample user in sample users who have not completed training in the current round as a current sample user, and determining the cross entropy loss of the current sample user in the current round according to the business demand prediction result of the current sample user and the corresponding actual business demand result;
adjusting parameters of the basic prediction model according to the cross entropy loss of the current sample user in the current round;
and taking the current sample user as a sample user completing the training in the current round, and returning to the step of determining the cross entropy loss of the current sample user in the current round until all sample users complete the training in the current round.
20. The method of claim 19, wherein the performing the current round of training of the base predictive model further comprises:
detecting whether the number of the current wheel reaches a preset number; if so, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
verifying the basic prediction model obtained in the current round by using a test set; if the cross entropy loss is not greater than the number of the test data of the preset cross entropy loss threshold value in the test set, the percentage of the total number of the test data in the test set is occupied, and the percentage is greater than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the sample user corresponding to the previous round in sequence; and if the cross entropy loss of the sample user in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample user in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as the prediction model.
21. The method of claim 13, wherein the sample order feature vector is generated by:
for each sample order, determining characteristic values of the sample order under a plurality of order characteristics according to order information of the sample order;
detecting whether the dimension of the characteristic value corresponding to each order characteristic is larger than a preset dimension or not;
if the dimension of the characteristic value corresponding to any order characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any order characteristic to obtain a compressed characteristic value of any order characteristic;
and constructing the sample order feature vector based on the compressed feature value corresponding to any order feature and the feature values corresponding to other order features.
22. The method of claim 13, wherein the user order feature vector is generated by:
for each sample user, determining a characteristic value of the sample user under a plurality of user characteristics according to the user information of the sample user;
detecting whether the dimensionality of the characteristic value corresponding to each user characteristic is larger than a preset dimensionality;
if the dimension of the characteristic value corresponding to any user characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any user characteristic to obtain a compressed characteristic value of any user characteristic;
and constructing the user feature vector based on the compressed feature value corresponding to any user feature and the feature values corresponding to other user features.
23. The method according to claim 21 or 22, characterized in that the feature values are dimension compressed in the following way:
acquiring a characteristic embedding matrix;
determining a product of the eigenvalue and the feature embedding matrix as the compressed eigenvalue.
24. A business need influencing factor determining apparatus, the apparatus comprising:
the system comprises an acquisition module, a prediction module and a prediction module, wherein the acquisition module is used for acquiring order information of a plurality of orders of a user to be predicted within a preset historical time period;
the construction module is used for constructing an order feature vector sequence of the user to be predicted according to the order information of the orders, wherein the order feature vector sequence comprises order feature vectors corresponding to the orders respectively;
and the determining module is used for determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model.
25. The apparatus of claim 24, wherein the constructing module is configured to construct the sequence of order feature vectors of the user to be predicted according to the order information of the plurality of orders by using the following steps, including:
for each order, determining characteristic values of the order under a plurality of order characteristics according to order information of the order;
according to the characteristic values of the order under the characteristics of the orders, constructing an order characteristic vector corresponding to the order;
and according to the order characteristic vectors corresponding to the orders respectively, constructing the order characteristic vector sequence according to the sequence of order generation time.
26. The apparatus of claim 25, wherein the determining module is configured to determine the information of the influence factors of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model by using the following steps:
inputting the order characteristic vector sequence into the prediction model to obtain a prediction result corresponding to the order characteristic vector sequence;
determining a second contribution value corresponding to each order feature based on a prediction result corresponding to the order feature vector sequence, elements in each feature vector sequence and reference values corresponding to the elements in the feature vector sequence;
and determining influence factor information of the user to be predicted in the order characteristic dimension based on the second contribution value.
27. The apparatus of claim 26, wherein the determining module is configured to determine the influence factor information of the user to be predicted in the order feature dimension based on the second contribution value by using the following steps, including:
determining at least one target order characteristic from the order characteristics according to the magnitude of the second contribution value;
and determining the influence factor information of the user to be predicted in order feature dimensions based on the influence factor classification corresponding to the target order features.
28. The apparatus of claim 27, wherein the classification of influencing factors comprises: the system is sensitive to one or more of price, few vehicles, waiting time, driver service, potential safety hazard and vehicle type and vehicle condition.
29. The apparatus of claim 26, wherein the second contribution value is an integrated gradsi(x) Satisfies the following formula:
Figure FDA0001895419210000101
wherein x isiRepresenting the ith element in the order feature vector; x'iRepresenting a reference value corresponding to the ith element in the order feature vector; f represents a prediction function corresponding to the prediction model; x' represents a feature vector composed of reference values of the features of each order; x represents an order feature vector in the sequence of order feature vectors.
30. The apparatus of claim 24, wherein the build module is further configured to:
according to the user information of the user to be predicted, constructing a user characteristic vector of the user to be predicted;
the determining module is used for determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence and a pre-trained prediction model by adopting the following steps:
and determining the influence factor information of the service demand of the user to be predicted in a future preset time period based on the order feature vector sequence, the user feature vector input and the pre-trained prediction model.
31. The apparatus of claim 30, wherein the determining module is configured to determine the information about the influence factors of the service demand of the user to be predicted in the future preset time period based on the order feature vector sequence, the user feature vector input, and the pre-trained prediction model by using the following steps:
inputting the order feature vector sequence and the user feature vector into the pre-trained prediction model, obtaining attention distribution weights corresponding to the order feature vectors from a target network layer of the prediction model, and taking the attention distribution weights as first contribution values of orders corresponding to the order feature vectors to silence the user to be predicted;
and determining influence factor information of the user to be predicted in the order dimension based on the first contribution value.
32. The apparatus of claim 30, wherein the constructing module is configured to construct the user feature vector of the user to be predicted according to the user information by:
determining characteristic values of the user to be predicted under various user characteristics according to the user information;
and constructing a user characteristic vector of the user to be predicted according to the characteristic values of the user to be predicted under the various user characteristics.
33. The apparatus of claim 30, wherein the user characteristics comprise a plurality of the following characteristics:
age, gender, occupation, complaint factors of the service requester.
34. The apparatus of claim 25, wherein the order characteristics include a plurality of the following characteristics:
the order starting point distance, the queuing waiting time, the order price, the order duration, the driving receiving distance, the pre-evaluation value and the service provider information.
35. The apparatus of claim 34, wherein the service provider information comprises at least one of:
age, gender, rating level, service score, service equipment level, service equipment cleanliness, order completion rate, complained factors.
36. The apparatus of claim 24, further comprising: a training module for training the predictive model in the following manner:
obtaining sample order information of a plurality of sample orders of a plurality of sample users in a target historical time period and an actual business demand result of each sample user in a prediction historical time period;
for each sample user, generating a sample order feature vector sequence of the sample user according to sample order information corresponding to each sample order of the sample user in a target historical time period; the sample order feature vector sequence comprises sample order feature vectors corresponding to all sample orders;
determining a service demand prediction result of each sample user in a prediction history time period based on the sample order feature vector sequence of each sample user and a basic prediction model;
and training the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result to obtain the prediction model.
37. The apparatus of claim 36, wherein the training module is further configured to, after obtaining sample order information of a plurality of sample orders of a plurality of sample users within a target historical time period, generate, for each sample user, a user feature vector of the sample user according to the user information of the sample user;
the training module is used for determining a business demand prediction result of each sample user based on the sample order feature vector sequence and the basic prediction model of each sample user by adopting the following steps:
and inputting the sample order characteristic vector sequence and the user characteristic vector of each sample user into a basic prediction model to obtain a service demand prediction result of each sample user.
38. The apparatus of claim 37, wherein the training module is configured to input the sample order feature vector sequence and the user feature vector of each sample user into a basic prediction model to obtain the prediction result of the business demand of each sample user by using the following steps:
for each sample user, inputting a sample order feature vector in a sample order feature vector sequence of the sample user into a first neural network, and acquiring a middle feature vector corresponding to each order feature vector;
inputting each intermediate feature vector of the sample user and the user feature vector into a second neural network, and acquiring attention distribution weights corresponding to each intermediate feature vector;
generating a fused feature vector based on each of the intermediate feature vectors of the sample user and the attention distribution weight;
and inputting the fusion characteristic vector of the sample user into a third neural network, extracting a target characteristic vector for the fusion characteristic vector, and inputting the target characteristic vector into a classifier to obtain a service demand prediction result of the sample user.
39. The apparatus of claim 38, wherein the training module is configured to, for each sample user, input a sample order feature vector in a sample order feature vector sequence of the sample user into the first neural network, and obtain an intermediate feature vector corresponding to each order feature vector by:
selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors of the sample user;
inputting the obtained characteristic vector of the previous sample order into the first neural network, wherein a target characteristic extraction layer of the first neural network is an intermediate characteristic vector output by the characteristic vector of the previous sample order;
inputting the current sample order feature vector and the intermediate feature vector of the previous sample order feature vector into the first neural network, and acquiring the intermediate feature vector corresponding to the current sample order feature vector;
and returning a sample order feature vector sequence aiming at the sample user, and selecting a sample order feature vector from the sample order feature vector sequence as a current sample order feature vector according to the arrangement sequence of the sample order feature vectors until the intermediate feature vectors of all the sample order feature vectors are extracted.
40. The apparatus of claim 38, wherein the training module is configured to input each of the intermediate feature vectors of the sample user and the user feature vector to a second neural network, and obtain attention allocation weights corresponding to each of the intermediate feature vectors by:
for each intermediate feature vector, splicing the intermediate feature vector with the user feature vector to generate a spliced vector corresponding to the intermediate feature vector;
inputting the splicing vector into the second neural network to obtain the matching degree corresponding to the splicing vector;
and performing activation operation on the matching degree based on a preset activation function to obtain the attention distribution weight corresponding to the intermediate feature vector.
41. The apparatus of claim 38, wherein the training module is configured to generate a fused feature vector based on the respective intermediate feature vectors and the attention-assigned weights of the sample users by:
and carrying out weighted summation on the intermediate feature vectors according to the attention distribution weights corresponding to the intermediate feature vectors to generate the fusion feature vectors.
42. The apparatus of claim 36, wherein the training module is configured to train the basic prediction model according to the business demand prediction result of each sample user and the corresponding actual business demand result by:
taking any sample user in sample users who have not completed training in the current round as a current sample user, and determining the cross entropy loss of the current sample user in the current round according to the business demand prediction result of the current sample user and the corresponding actual business demand result;
adjusting parameters of the basic prediction model according to the cross entropy loss of the current sample user in the current round;
and taking the current sample user as a sample user completing the training in the current round, and returning to the step of determining the cross entropy loss of the current sample user in the current round until all sample users complete the training in the current round.
43. The apparatus according to claim 42, wherein the training module is further configured to detect whether a current round reaches a preset number of rounds after completing the current round of training of the basic prediction model; if so, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
verifying the basic prediction model obtained in the current round by using a test set; if the cross entropy loss is not greater than the number of the test data of the preset cross entropy loss threshold value in the test set, the percentage of the total number of the test data in the test set is occupied, and the percentage is greater than a preset first percentage threshold value, stopping training the basic prediction model, and taking the basic prediction model obtained in the last round of training as the prediction model;
or,
comparing the cross entropy loss of each sample user in the current round with the cross entropy loss of the sample user corresponding to the previous round in sequence; and if the cross entropy loss of the sample user in the current round is larger than the number of the sample users with the cross entropy loss of the corresponding sample user in the previous round, and the percentage of the number of all the sample users reaches a preset second percentage threshold, stopping the training of the basic prediction model, and taking the basic prediction model obtained in the previous round of training as the prediction model.
44. The apparatus of claim 36, wherein the training module is configured to generate the sample order feature vector by:
for each sample order, determining characteristic values of the sample order under a plurality of order characteristics according to order information of the sample order;
detecting whether the dimension of the characteristic value corresponding to each order characteristic is larger than a preset dimension or not;
if the dimension of the characteristic value corresponding to any order characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any order characteristic to obtain a compressed characteristic value of any order characteristic;
and constructing the sample order feature vector based on the compressed feature value corresponding to any order feature and the feature values corresponding to other order features.
45. The apparatus of claim 36, wherein the training module is configured to generate the user order feature vector by:
for each sample user, determining a characteristic value of the sample user under a plurality of user characteristics according to the user information of the sample user;
detecting whether the dimensionality of the characteristic value corresponding to each user characteristic is larger than a preset dimensionality;
if the dimension of the characteristic value corresponding to any user characteristic is larger than the preset dimension, performing dimension compression on the characteristic value of any user characteristic to obtain a compressed characteristic value of any user characteristic;
and constructing the user feature vector based on the compressed feature value corresponding to any user feature and the feature values corresponding to other user features.
46. The apparatus of claim 44 or 45, wherein the training module is configured to perform dimension compression on the eigenvalues by:
acquiring a characteristic embedding matrix;
determining a product of the eigenvalue and the feature embedding matrix as the compressed eigenvalue.
47. An electronic device, comprising: a processor, a storage medium and a bus, the storage medium storing machine-readable instructions executable by the processor, the processor and the storage medium communicating via the bus when the electronic device is operating, the processor executing the machine-readable instructions to perform the steps of the business requirement influencing factor method according to any one of claims 1 to 23.
48. A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, carries out the steps of the business need influencing factor method according to any of claims 1 to 23.
CN201811489839.0A 2018-12-06 2018-12-06 Method and device for determining business demand influence factors Pending CN111292106A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811489839.0A CN111292106A (en) 2018-12-06 2018-12-06 Method and device for determining business demand influence factors

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811489839.0A CN111292106A (en) 2018-12-06 2018-12-06 Method and device for determining business demand influence factors

Publications (1)

Publication Number Publication Date
CN111292106A true CN111292106A (en) 2020-06-16

Family

ID=71025214

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811489839.0A Pending CN111292106A (en) 2018-12-06 2018-12-06 Method and device for determining business demand influence factors

Country Status (1)

Country Link
CN (1) CN111292106A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112801690A (en) * 2020-12-31 2021-05-14 北京嘀嘀无限科技发展有限公司 Method and device for determining intervention characteristics
CN113393279A (en) * 2021-07-08 2021-09-14 北京沃东天骏信息技术有限公司 Order quantity estimation method and system
CN118264624A (en) * 2024-05-28 2024-06-28 北京长亭科技有限公司 Flow characteristic vector generation method and device

Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001134643A (en) * 1999-11-02 2001-05-18 Hitachi Ltd Device and method for predicting demand
US20020169657A1 (en) * 2000-10-27 2002-11-14 Manugistics, Inc. Supply chain demand forecasting and planning
JP2004234470A (en) * 2003-01-31 2004-08-19 Ricoh Co Ltd Demand prediction method and demand prediction program
CN102938742A (en) * 2012-11-08 2013-02-20 广东电网公司电力调度控制中心 Communication bandwidth forecasting method and device based on power business requirements
US20150058078A1 (en) * 2013-08-26 2015-02-26 Microsoft Corporation Rule to constraint translator for business application systems
US20160055494A1 (en) * 2014-08-19 2016-02-25 Boyi Ni Booking based demand forecast
CN106203682A (en) * 2016-06-29 2016-12-07 东南大学 Hotel's disposable product needing forecasting method and form ordering system
WO2017063356A1 (en) * 2015-10-14 2017-04-20 深圳市天行家科技有限公司 Designated-driving order predicting method and designated-driving transport capacity scheduling method
CN106651213A (en) * 2017-01-03 2017-05-10 百度在线网络技术（北京）有限公司 Processing method and device for service orders
CN107038492A (en) * 2016-02-04 2017-08-11 滴滴（中国）科技有限公司 Daily Order volume Forecasting Methodology and device based on Arma models
CN107220845A (en) * 2017-05-09 2017-09-29 北京小度信息科技有限公司 User purchases probabilistic forecasting/user quality and determines method, device and electronic equipment again
CN107292420A (en) * 2017-05-24 2017-10-24 国网北京市电力公司 Data processing method and device
US20170364933A1 (en) * 2014-12-09 2017-12-21 Beijing Didi Infinity Technology And Development Co., Ltd. User maintenance system and method
CN108230049A (en) * 2018-02-09 2018-06-29 新智数字科技有限公司 The Forecasting Methodology and system of order
US20180232676A1 (en) * 2017-02-10 2018-08-16 Cimpress Schweiz Gmbh Fulfillment Success Prediction for Product Procurement
CN108681857A (en) * 2018-05-18 2018-10-19 北京顺丰同城科技有限公司 A kind of dispatching order allocation method and device, computer readable storage medium
CN108681751A (en) * 2018-05-22 2018-10-19 平安科技（深圳）有限公司 Determine the method and terminal device of event influence factor
CN108876056A (en) * 2018-07-20 2018-11-23 广东工业大学 A kind of shared bicycle Demand Forecast method, apparatus, equipment and storage medium

Patent Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001134643A (en) * 1999-11-02 2001-05-18 Hitachi Ltd Device and method for predicting demand
US20020169657A1 (en) * 2000-10-27 2002-11-14 Manugistics, Inc. Supply chain demand forecasting and planning
JP2004234470A (en) * 2003-01-31 2004-08-19 Ricoh Co Ltd Demand prediction method and demand prediction program
CN102938742A (en) * 2012-11-08 2013-02-20 广东电网公司电力调度控制中心 Communication bandwidth forecasting method and device based on power business requirements
US20150058078A1 (en) * 2013-08-26 2015-02-26 Microsoft Corporation Rule to constraint translator for business application systems
US20160055494A1 (en) * 2014-08-19 2016-02-25 Boyi Ni Booking based demand forecast
US20170364933A1 (en) * 2014-12-09 2017-12-21 Beijing Didi Infinity Technology And Development Co., Ltd. User maintenance system and method
WO2017063356A1 (en) * 2015-10-14 2017-04-20 深圳市天行家科技有限公司 Designated-driving order predicting method and designated-driving transport capacity scheduling method
CN107038492A (en) * 2016-02-04 2017-08-11 滴滴（中国）科技有限公司 Daily Order volume Forecasting Methodology and device based on Arma models
CN106203682A (en) * 2016-06-29 2016-12-07 东南大学 Hotel's disposable product needing forecasting method and form ordering system
CN106651213A (en) * 2017-01-03 2017-05-10 百度在线网络技术（北京）有限公司 Processing method and device for service orders
US20180232676A1 (en) * 2017-02-10 2018-08-16 Cimpress Schweiz Gmbh Fulfillment Success Prediction for Product Procurement
CN107220845A (en) * 2017-05-09 2017-09-29 北京小度信息科技有限公司 User purchases probabilistic forecasting/user quality and determines method, device and electronic equipment again
CN107292420A (en) * 2017-05-24 2017-10-24 国网北京市电力公司 Data processing method and device
CN108230049A (en) * 2018-02-09 2018-06-29 新智数字科技有限公司 The Forecasting Methodology and system of order
CN108681857A (en) * 2018-05-18 2018-10-19 北京顺丰同城科技有限公司 A kind of dispatching order allocation method and device, computer readable storage medium
CN108681751A (en) * 2018-05-22 2018-10-19 平安科技（深圳）有限公司 Determine the method and terminal device of event influence factor
CN108876056A (en) * 2018-07-20 2018-11-23 广东工业大学 A kind of shared bicycle Demand Forecast method, apparatus, equipment and storage medium

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112801690A (en) * 2020-12-31 2021-05-14 北京嘀嘀无限科技发展有限公司 Method and device for determining intervention characteristics
CN113393279A (en) * 2021-07-08 2021-09-14 北京沃东天骏信息技术有限公司 Order quantity estimation method and system
CN118264624A (en) * 2024-05-28 2024-06-28 北京长亭科技有限公司 Flow characteristic vector generation method and device
CN118264624B (en) * 2024-05-28 2024-08-23 北京长亭科技有限公司 Flow characteristic vector generation method and device

Similar Documents

Publication Publication Date Title
CN111292105B (en) Service demand determining method and device
Chen et al. Understanding ridesplitting behavior of on-demand ride services: An ensemble learning approach
KR102272942B1 (en) Presentation systems, devices, methods and readable storage media for auto insurance expedition missions
CN105849998B (en) Electric power demand forecasting device and prediction technique, power supply system and recording medium
US20200302322A1 (en) Machine learning system
US9576250B2 (en) Method and system for simulating users in the context of a parking lot based on the automatic learning of a user choice decision function from historical data considering multiple user behavior profiles
Hancock et al. Decision field theory: Improvements to current methodology and comparisons with standard choice modelling techniques
CN111325374B (en) Method and device for predicting order cancellation probability and electronic equipment
US20240194071A1 (en) Parking area recommendation method and apparatus, and medium
CN111292106A (en) Method and device for determining business demand influence factors
CN111353094A (en) Information pushing method and device
CN110910180A (en) Information pushing method and device, electronic equipment and storage medium
Jabbari et al. How do perceptions of safety and car ownership importance affect autonomous vehicle adoption?
CN111724184A (en) Transformation probability prediction method and device
US20200311553A1 (en) Method, apparatus, and computer program product for identifying and compensating content contributors
CN111680382A (en) Grade prediction model training method, grade prediction device and electronic equipment
CN111310961A (en) Data prediction method, data prediction device, electronic equipment and computer readable storage medium
CN111461379A (en) Position prediction method and device
CN111861538A (en) Information pushing method and device, electronic equipment and storage medium
CN111369029A (en) Service selection prediction method, device, electronic equipment and storage medium
KR102474974B1 (en) Method, device and system for analyzing brand based on artificial intelligence
US20200311757A1 (en) Method, apparatus, and computer program product for identifying and compensating content contributors
CN111611353A (en) Screening method and device, electronic equipment and computer readable storage medium
CN111259229A (en) Question recommendation method and device
CN111291253A (en) Model training method, consultation recommendation method, device and electronic equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
AD01 Patent right deemed abandoned

Effective date of abandoning: 20240913

AD01 Patent right deemed abandoned