CN109146211A - The distribution of order, the training method of model and device - Google Patents
The distribution of order, the training method of model and device Download PDFInfo
- Publication number
- CN109146211A CN109146211A CN201710457389.6A CN201710457389A CN109146211A CN 109146211 A CN109146211 A CN 109146211A CN 201710457389 A CN201710457389 A CN 201710457389A CN 109146211 A CN109146211 A CN 109146211A
- Authority
- CN
- China
- Prior art keywords
- information
- rideshare
- target
- history
- sample
- 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
Links
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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
-
- 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)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Educational Administration (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The disclosure provides a kind of distribution of order, the training method of model and device, it is related to machine learning techniques field, one specific embodiment of the method includes: acquisition target information, which includes that the information of service provider, the service provider have connect the information of the first rideshare order, the information and current real time information of the second rideshare order to be allocated;According to the target information, target signature information is obtained based on the target information;The target signature information is inputted into target linear regression model (LRM) and target depth learning model respectively;Target linear regression model (LRM) is weighted with the result that target depth learning model exports and is added, to obtain match parameter;If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given to the service provider.The embodiment makes the matching of rideshare order and service provider more reasonable, improves efficiency of service, also improves the utilization rate of Service Source.
Description
Technical field
This disclosure relates to machine learning techniques field, in particular to a kind of distribution of order, model training method and dress
It sets.
Background technique
In recent years, with the continuous development of Internet technology, by the commercial chance under line in conjunction with internet, occur new
O2O (Online To Offline, under line on online offline/line) business model, so that internet is become the flat of off-line transaction
Platform.Currently, O2O has come into the stage of high speed development, wherein the O2O service of the vehicles is that the comparison of development is successful
O2O service.By taking vehicle service as an example, currently, vehicle service may include a variety of different classifications, for example, express service, special train
Service, windward driving service, test ride service and service of cars on hire etc..Wherein, there is some type of vehicle service that can also provide
Rideshare service.When providing rideshare service, due to requiring driver to provide service to two or more passengers simultaneously, to be related to
Multiple starts of a run and multiple stroke ends, therefore, if order matching it is unreasonable, may result in problems (for example,
The problems such as excessive or response rate that detours is too low), to not only reduce efficiency of service, also reduce the utilization of Service Source
Rate.
Summary of the invention
In order to solve the above-mentioned technical problem, present disclose provides a kind of distribution of order, the training method of model and devices.
According to the first aspect of the embodiments of the present disclosure, a kind of distribution method of order is provided, comprising:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect first
The information and current real time information of the information of rideshare order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, with
Obtain match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given to the clothes
Be engaged in provider.
According to the second aspect of an embodiment of the present disclosure, a kind of training method of the distribution model of order is provided, comprising:
Obtain sample information, the sample information include in multiple history rideshare events each history rideshare event it is corresponding
Related information;
The corresponding sample attribute of each history rideshare event, the sample attribute packet are determined based on the sample information
Include positive sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained line is treated
Property regression model and deep learning model to be trained carry out parameter adjustment, it is deep to obtain target linear regression model (LRM) and target
Spend learning model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare
The first history rideshare that the information of service provider, the service provider are first connected in the history rideshare event in event is ordered
The information of single information and the second history rideshare order being followed by.
According to the third aspect of an embodiment of the present disclosure, a kind of distributor of order is provided, comprising:
First acquisition unit is configured as obtaining target information, and the target information includes the information of service provider, institute
It states service provider and has connect the information of the first rideshare order, the information of the second rideshare order to be allocated and current real-time letter
Breath;
Second acquisition unit is configured as obtaining target signature information based on the target information;
Input unit is configured as the target signature information inputting target linear regression model (LRM) and target depth respectively
Learning model;
Subelement is exported, is configured as exporting the target linear regression model (LRM) and the target deep learning model
As a result it is weighted addition, to obtain match parameter;
Allocation unit is configured as then closing described second when the match parameter is more than or equal to preset threshold
Multiply Order splitting to the service provider.
According to a fourth aspect of embodiments of the present disclosure, a kind of distributor of order is provided, comprising:
First acquisition unit is configured as obtaining sample information, and the sample information includes in multiple history rideshare events
The corresponding related information of each history rideshare event;
Determination unit is configured as determining the corresponding sample category of each history rideshare event based on the sample information
Property, the sample attribute includes positive sample attribute and negative sample attribute;
Second acquisition unit is configured as obtaining the corresponding mesh of each history rideshare event based on the sample information
Standard specimen eigen information;
Adjustment unit is configured as according to the corresponding sample attribute of each history rideshare event and target sample feature
Information, treats trained linear regression model (LRM) and deep learning model to be trained carries out parameter adjustment, linear to obtain target
Regression model and target deep learning model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare
The first history rideshare that the information of service provider, the service provider are first connected in the history rideshare event in event is ordered
The information of single information and the second history rideshare order being followed by.
According to a fifth aspect of the embodiments of the present disclosure, a kind of computer storage medium is provided, is stored in the storage medium
There is program instruction, described instruction includes:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect first
The information and current real time information of the information of rideshare order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, with
Obtain match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given to the clothes
Be engaged in provider.
According to a sixth aspect of an embodiment of the present disclosure, a kind of computer storage medium is provided, is stored in the storage medium
There is program instruction, described instruction includes:
Obtain sample information, the sample information include in multiple history rideshare events each history rideshare event it is corresponding
Related information;
The corresponding sample attribute of each history rideshare event, the sample attribute packet are determined based on the sample information
Include positive sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained line is treated
Property regression model and deep learning model to be trained carry out parameter adjustment, it is deep to obtain target linear regression model (LRM) and target
Spend learning model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare
The first history rideshare that the information of service provider, the service provider are first connected in the history rideshare event in event is ordered
The information of single information and the second history rideshare order being followed by.
According to the 7th of the embodiment of the present disclosure the aspect, a kind of electronic equipment is provided, comprising:
Processor is adapted for carrying out each instruction;And
Equipment is stored, is suitable for storing a plurality of instruction, described instruction is suitable for being loaded and being executed by processor:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect first
The information and current real time information of the information of rideshare order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, with
Obtain match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given to the clothes
Be engaged in provider.
According to the eighth aspect of the embodiment of the present disclosure, a kind of electronic equipment is provided, comprising:
Processor is adapted for carrying out each instruction;And
Equipment is stored, is suitable for storing a plurality of instruction, described instruction is suitable for being loaded and being executed by processor:
Obtain sample information, the sample information include in multiple history rideshare events each history rideshare event it is corresponding
Related information;
The corresponding sample attribute of each history rideshare event, the sample attribute packet are determined based on the sample information
Include positive sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained line is treated
Property regression model and deep learning model to be trained carry out parameter adjustment, it is deep to obtain target linear regression model (LRM) and target
Spend learning model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare
The first history rideshare that the information of service provider, the service provider are first connected in the history rideshare event in event is ordered
The information of single information and the second history rideshare order being followed by.
The technical scheme provided by this disclosed embodiment can include the following benefits:
The distribution method and device for the order that embodiment of the disclosure provides are believed by obtaining target information based on target
Breath obtains target signature information, and target signature information is inputted to target linear regression model (LRM) and target deep learning model respectively,
Target linear regression model (LRM) is weighted with the result that target deep learning model exports and is added, to obtain match parameter.Such as
The fruit match parameter is more than or equal to preset threshold, then by the second rideshare Order splitting to service provider.Wherein, the target
Information includes the information of service provider, and service provider has connect the information of the first rideshare order, and the second rideshare to be allocated is ordered
Single information and current real time information.Due to what is combined using target linear regression model (LRM) and target deep learning model
Mode determines the matching degree of rideshare order and service provider, therefore, so that the matching of rideshare order and service provider is more
Adduction reason, also improves the utilization rate of Service Source.
The training method and device of the distribution model of the order provided by the above embodiment of the disclosure, by obtaining sample letter
Breath, which includes the corresponding related information of each history rideshare event in multiple history rideshare events, and is based on the sample
This information determines the corresponding sample attribute of each history rideshare event.Each history rideshare event pair is obtained based on the sample information
The target sample characteristic information answered is treated according to the corresponding sample attribute of each history rideshare event and target sample characteristic information
Trained linear regression model (LRM) and deep learning model to be trained carry out parameter adjustment, to obtain target linear regression model (LRM)
And target deep learning model.To obtain the model that can be used for rideshare Order splitting, so that rideshare order and service
The matching of provider is more reasonable, helps to improve the utilization rate of Service Source.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure
Example, and together with specification for explaining the principles of this disclosure.
Fig. 1 is the exemplary system architecture schematic diagram using the embodiment of the present disclosure;
Fig. 2 is a kind of disclosure flow chart of the distribution method of order shown according to an exemplary embodiment;
Fig. 3 is a kind of disclosure schematic diagram of a scenario of the distribution of order shown according to an exemplary embodiment;
Fig. 4 is the flow chart of the distribution method of the disclosure another order shown according to an exemplary embodiment;
Fig. 5 is a kind of disclosure process of the training method of the distribution model of order shown according to an exemplary embodiment
Figure;
Fig. 6 is a kind of disclosure distributor block diagram of order shown according to an exemplary embodiment;
Fig. 7 is a kind of disclosure training device block diagram of the distribution model of order shown according to an exemplary embodiment.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
It is only to be not intended to be limiting the disclosure merely for for the purpose of describing particular embodiments in the term that the disclosure uses.
The "an" of the singular used in disclosure and the accompanying claims book, " described " and "the" are also intended to including majority
Form, unless the context clearly indicates other meaning.It is also understood that term "and/or" used herein refers to and wraps
It may be combined containing one or more associated any or all of project listed.
It will be appreciated that though various information, but this may be described using term first, second, third, etc. in the disclosure
A little information should not necessarily be limited by these terms.These terms are only used to for same type of information being distinguished from each other out.For example, not departing from
In the case where disclosure range, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as
One information.Depending on context, word as used in this " if " can be construed to " ... when " or " when ...
When " or " in response to determination ".
Referring to Fig. 1, for using the exemplary system architecture schematic diagram of the embodiment of the present disclosure:
As shown in Figure 1, system architecture 100 may include terminal device, for example, diagram terminal device 101,102,103,
Network 104 and server 105.It should be understood that the number or type of terminal device, network and server in Fig. 1 are only to show
Meaning property.According to needs are realized, the terminal device, network and server of arbitrary number or type can have.
Network 104 between terminal device, server for providing the medium of communication link.Network 104 may include each
Kind connection type, such as wired, wireless communication link or fiber optic cables etc..
Terminal device 101,102,103 can be interacted by network 104 with server, with receive or send request or
Information etc..Terminal device 101,102,103 can be various electronic equipments, including but not limited to smart phone, tablet computer, intelligence
It can wearable device and personal digital assistant etc..
Server 105 can be to provide the server of various services.Server can store the data received,
The processing such as analysis can also send control command or request etc. to terminal device or other servers.Server can respond
In user service request and service is provided.It is appreciated that a server can provide one or more services, same clothes
Business can also be provided by multiple servers.
The disclosure is described in detail below in conjunction with specific embodiments.
As shown in Fig. 2, Fig. 2 is a kind of flow chart of the distribution method of order shown according to an exemplary embodiment, it should
Method can be applied in server.Method includes the following steps:
In step 201, target information is obtained, which includes the information of service provider, the service provider
The information of the first rideshare order, the information and current real time information of the second rideshare order to be allocated are connect.
In the present embodiment, related service can be the rideshare service of the vehicles (for example, vehicle rideshare service
Deng), related scene can be to provide the service provider of vehicles Composite service to have been attached to a rideshare order, just
In the scene for waiting another rideshare order to be received.For example, being directed to the service of vehicle rideshare, service provider, which can be, carries passenger
Driver.First rideshare order is the rideshare order that service provider is connected to first, and the second rideshare order is rideshare to be allocated
Order.
In the present embodiment, the information of service provider may include that various can characterize service provider personal characteristics
Information.By taking the service of vehicle rideshare as an example, service provider is to provide the driver of vehicle rideshare service, and the information of service provider can
To include but is not limited to the id information of the driver, the gender information of the driver, the age information of the driver, the driver's is serviced
Divide information, the star information of the driver, the vehicle model information of the driver driving, the current location information etc. of the driver.
In the present embodiment, the information of the first rideshare order may include the various information for including in the first rideshare order,
For example, the information of the first rideshare order can include but is not limited to the corresponding start of a run of the first rideshare order and stroke end
Location information, the first rideshare order corresponding bill moment, corresponding user information of the first rideshare order etc..Second rideshare
The information of order may include the various information for including in the second rideshare order, for example, the information of the second rideshare order can wrap
The location information of the corresponding start of a run of the second rideshare order and stroke end is included but is not limited to, the second rideshare order is corresponding
Bill moment, corresponding user information of the second rideshare order etc..Wherein, user information can include but is not limited to the user's
Id information, the user of the user draw a portrait information (for example, gender information, age information, preference information, occupational information etc.) etc.
Deng.
In the present embodiment, current real time information can include but is not limited to current Weather information, current time
Information (for example, time information, week information, the date information of the Gregorian calendar, the date information of the lunar calendar, festival information etc.), current
Traffic related information etc..
In step 202, target signature information is obtained based on target information.
In the present embodiment, target information can be primarily based on and obtain initial characteristics information, initial characteristics information can wrap
Mark class initial characteristics information and non-identifying class initial characteristics information are included, then, respectively to above-mentioned mark class initial characteristics information
Processing is optimized with non-identifying class initial characteristics information, to obtain target signature information.
In the present embodiment, initial characteristics information may include the first initial characteristics information and the second initial characteristics letter
Breath, the first initial characteristics information are that can be directly based upon characteristic information obtained from target information, and the second initial characteristics information is
Characteristic information obtained from needing to estimate target information.Specifically, first can be directly extracted from target information
Initial characteristics information, also, being estimated based on target information (can use preset algorithm, strategy or model be estimated
Survey), to obtain the second initial characteristics information.
In the present embodiment, the first initial characteristics information may include following one or more: the gender of service provider
Characteristic information, the age characteristics information of service provider, the service of service provider divide characteristic information, the star of service provider
Characteristic information, the vehicle characteristic information that service provider drives, the present position characteristic information of service provider, weather are special
Reference breath, temporal characteristics information.
In the present embodiment, the second initial characteristics information may include following one or more: the first rideshare order is corresponding
Original stroke estimated distance characteristic information;The corresponding original stroke estimated distance characteristic information of second rideshare order;After rideshare
The corresponding stroke estimated distance characteristic information of first rideshare order;The corresponding stroke estimated distance of the second rideshare order is special after rideshare
Reference breath;First rideshare order with the second rideshare order is corresponding after rideshare multiplies estimated distance characteristic information altogether;First rideshare
Order and the second rideshare order are corresponding after rideshare to multiply estimation temporal characteristics information altogether;First rideshare order it is corresponding estimation around
Road distance feature information;Second rideshare order is corresponding to estimate the distance feature information that detours;The corresponding estimation of first rideshare order
Detour temporal characteristics information;Second rideshare order is corresponding to estimate the temporal characteristics information that detours;First rideshare order is corresponding to be estimated
Count the characteristic information of the ratio between the distance original stroke estimated distance corresponding with the first rideshare order that detours;Second rideshare order is corresponding
Estimation detour the characteristic information of the ratio between distance original stroke estimated distance corresponding with the second rideshare order;Second rideshare order
Moment characteristic information is welcomed the emperor in corresponding estimation;Distance feature information is welcomed the emperor in the corresponding estimation of second rideshare order;Second rideshare is ordered
Single corresponding estimation welcomes the emperor distance and estimates the characteristic information of the ratio between stroke distances.
In the present embodiment, initial characteristics information may include two classes, one kind for mark class initial characteristics information (for example,
The characteristic information of ID class), another kind of is non-identifying class initial characteristics information (for example, characteristic information of non-ID class).To mark class
After initial characteristics information and non-identifying class initial characteristics information optimize processing, available target signature information, target spy
Reference breath may include first object characteristic information, the second target signature information and third target signature information.
It in the present embodiment, can be initially special to mark class initial characteristics information and non-identifying class respectively in the following way
Reference breath optimizes processing: in the target integrated model that non-identifying class initial characteristics information input is trained in advance, object set
It can be by (each leaf node on a leaf node of non-identifying class initial characteristics information MAP to every decision tree at model
A corresponding weighted value), using these leaf nodes as destination node, using the corresponding weighted value of destination node as characteristic value,
Available non-identifying class initial characteristics information integrates the character representation of the model space as the defeated of target integrated model in target
Out as a result, and the output result of target integrated model is normalized, to obtain first object characteristic information.It will be nonstandard
Know class initial characteristics information to be normalized, to obtain the second target signature information.Will mark class initial characteristics information into
Row sliding-model control, and be normalized, to obtain third target signature information.Wherein, target integrated model can be
Any reasonable integrated model, including but not limited to XGBoost (Extreme Gradient Boosting, in extreme gradient
Rise) model, the disclosure to not limiting in this respect.
In step 203, target signature information is inputted to target linear regression model (LRM) and target deep learning model respectively.
In the present embodiment, target linear regression model (LRM) can be any reasonable linear regression model (LRM), target depth
Practising model (e.g., deep neural network CNN etc.) can be any reasonable deep learning model, and the disclosure is to unlimited in this respect
It is fixed.
In step 204, the result that target linear regression model (LRM) and target deep learning model export is weighted phase
Add, to obtain match parameter.
In the present embodiment, target signature information can be inputted to target linear regression model (LRM) and target depth study respectively
Target linear regression model (LRM) is weighted with the result that target deep learning model exports and is added by model, to obtain matching ginseng
Number.The match parameter can characterize the matching degree of the second rideshare order and service provider.If the match parameter be greater than or
Equal to preset threshold, then illustrate that the second rideshare order matches with service provider.
In step 205, if match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given to the clothes
Be engaged in provider.
Fig. 3 is a kind of disclosure schematic diagram of a scenario of the distribution of order shown according to an exemplary embodiment, such as Fig. 3 institute
Show, it is possible, firstly, to obtain initial characteristics information according to target information, then, it is initially special that initial characteristics information is divided into mark class
Reference breath and non-identifying class initial characteristics information.By non-identifying class initial characteristics information input to target integrated model, and to mesh
The result of mark integrated model output is normalized, and obtains first object characteristic information.Non-identifying class initial characteristics are believed
Breath is directly normalized, and obtains the second target signature information.Mark class initial characteristics information is subjected to sliding-model control,
And be normalized, obtain third target signature information.By first object characteristic information, the second target signature information and
Three target signature informations are used as target signature information, input target linear regression model (LRM) and target deep learning model respectively.
Target linear regression model (LRM) is weighted with the result that target deep learning model exports and is added, to obtain match parameter.
The second rideshare order can be allocated based on match parameter.
The distribution method of the order provided by the above embodiment of the disclosure is based on target information by obtaining target information
Target signature information is obtained, target signature information is inputted to target linear regression model (LRM) and target deep learning model respectively, it will
Target linear regression model (LRM) is weighted with the result that target deep learning model exports and is added, to obtain match parameter.If
The match parameter is more than or equal to preset threshold, then by the second rideshare Order splitting to service provider.Wherein, which believes
Breath includes the information of service provider, and service provider has connect the information of the first rideshare order, the second rideshare order to be allocated
Information and current real time information.Due to the side combined using target linear regression model (LRM) and target deep learning model
Formula determines the matching degree of rideshare order and service provider, therefore, so that the matching of rideshare order and service provider is more
Rationally, the utilization rate of Service Source is also improved.
Fig. 4 is the flow chart of the distribution method of another order shown according to an exemplary embodiment, and the embodiment is detailed
It carefully describes according to the process for obtaining target signature information based on target information, this method can be applied in server.The party
Method may comprise steps of:
In step 401, target information is obtained, which includes the information of service provider, the service provider
The information of the first rideshare order, the information and current real time information of the second rideshare order to be allocated are connect.
In step 402, mark class initial characteristics information and non-identifying class initial characteristics information are obtained according to target information.
In step 403, place is optimized to mark class initial characteristics information and non-identifying class initial characteristics information respectively
Reason, to obtain target signature information.
In step 404, target signature information is inputted to target linear regression model (LRM) and target deep learning model respectively.
In step 405, the result that target linear regression model (LRM) and target deep learning model export is weighted phase
Add, to obtain match parameter.
In a step 406, if the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given
Above-mentioned service provider.
It should be noted that no longer going to live in the household of one's in-laws on getting married in above-mentioned Fig. 4 embodiment for the step identical with Fig. 2 embodiment
It states, related content can be found in Fig. 2 embodiment.
The distribution method of the order provided by the above embodiment of the disclosure, by obtaining target information, according to target information
Mark class initial characteristics information and non-identifying class initial characteristics information is obtained, respectively to mark class initial characteristics information and non-identifying
Class initial characteristics information optimizes processing, to obtain target signature information.It is linear that target signature information is inputted to target respectively
Regression model and target deep learning model carry out the result that target linear regression model (LRM) and target deep learning model export
Weighting summation, to obtain match parameter.If the match parameter is more than or equal to preset threshold, by the second rideshare order point
Dispensing service provider.Wherein, which includes the information of service provider, and service provider has connect the first rideshare order
Information, the information and current real time information of the second rideshare order to be allocated.Due to respectively to mark class initial characteristics
Processing is optimized in information and non-identifying class initial characteristics information, to obtain target signature information, then linear using target
The mode that regression model and target deep learning model combine, determines the matching degree of rideshare order and service provider, because
This improves efficiency of service further such that the matching of rideshare order and service provider is more reasonable, also improves service money
The utilization rate in source.
As shown in figure 5, Fig. 5 is a kind of training method of the distribution model of order shown according to an exemplary embodiment
Flow chart, this method can be applied in server.Method includes the following steps:
In step 501, sample information is obtained, which includes each history rideshare in multiple history rideshare events
The corresponding related information of event.
In the present embodiment, for any history rideshare event, which be can wrap
Include the corresponding real time information of history rideshare event, the information of service provider, the service provider in the history rideshare event
The information for the first history rideshare order being first connected in the history rideshare event and the second history rideshare order being followed by
Information.
In the present embodiment, the information of service provider may include that various can characterize service provider personal characteristics
Information.By taking the service of vehicle rideshare as an example, service provider is to provide the driver of vehicle rideshare service, and the information of service provider can
To include but is not limited to the id information of the driver, the gender information of the driver, the age information of the driver, the driver's is serviced
Divide information, the star information of the driver, the vehicle model information of the driver driving, location information of the driver etc..
In the present embodiment, the information of the first history rideshare order may include include each in the first history rideshare order
Kind information, for example, the information of the first history rideshare order can include but is not limited to the corresponding stroke of the first history rideshare order
The location information of starting point and stroke end, the first history rideshare order corresponding bill moment, the first history rideshare order pair
User information answered etc..The information of second history rideshare order may include the various letters for including in the second history rideshare order
Breath, for example, the information of the second history rideshare order can include but is not limited to the corresponding start of a run of the second history rideshare order
And the location information of stroke end, at the second history rideshare order corresponding bill moment, the second history rideshare order is corresponding
User information etc..Wherein, user information can include but is not limited to the id information of the user, user's portrait information of the user
(for example, gender information, age information, preference information, occupational information etc.) etc..
In the present embodiment, which can include but is not limited to the history rideshare thing
Weather information when part occurs, temporal information is (for example, time information, week information, the date information of the Gregorian calendar, the date of the lunar calendar
Information, festival information etc.), traffic related information etc..
In step 502, the corresponding sample attribute of each history rideshare event, the sample category are determined based on the sample information
Property includes positive sample attribute and negative sample attribute.
In the present embodiment, it can determine that each history is closed according to the evaluation information and response situation of order in sample information
Multiply the corresponding sample attribute of event.For example, order evaluation is preferable, the corresponding sample of the successful history rideshare event of order response
Attribute can be positive sample attribute.Order evaluation is poor, the corresponding sample attribute of history rideshare event of order answer failed
Can be negative sample attribute.The disclosure does not limit the specific division mode of positive and negative sample attribute.
In step 503, the corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information.
In the present embodiment, the sample information can be primarily based on and obtain the corresponding initial sample of each history rideshare event
Characteristic information, initial sample characteristics information may include the mark initial sample characteristics information of class and the initial sample characteristics of non-identifying class
Then information respectively optimizes the initial sample characteristics information of above-mentioned mark class and the initial sample characteristics information of non-identifying class
Processing, to obtain target sample characteristic information.
In the present embodiment, for any history rideshare event, the corresponding initial sample characteristics letter of the history rideshare event
Breath may include the corresponding first initial sample characteristics information of the history rideshare event and corresponding second initial sample characteristics
Information.First initial sample characteristics information is that can be directly based upon characteristic information obtained from sample information, the second initial sample
Characteristic information is characteristic information obtained from needing to estimate sample information.It specifically, can be straight from sample information
It connects and extracts the first initial sample characteristics information, also, estimated based on sample information and (preset algorithm, strategy can be used
Or model is estimated), to obtain the second initial sample characteristics information.
In the present embodiment, which may include following one or more: the history rideshare
The sex character information of the corresponding service provider of event, the age characteristics letter of the corresponding service provider of history rideshare event
Breath, the service of the corresponding service provider of history rideshare event divide characteristic information, which mentions
The star characteristic information of supplier, the vehicle characteristic information that the corresponding service provider of the history rideshare event drives, the history are closed
Multiply the position characteristic information of the corresponding service provider of event, which should
The corresponding temporal characteristics information of history rideshare event.
In the present embodiment, which may include following one or more: the first history is closed
Multiply the corresponding original stroke estimated distance characteristic information of order;The corresponding original stroke estimated distance of second history rideshare order is special
Reference breath;The corresponding stroke estimated distance characteristic information of first history rideshare order after rideshare;The second history rideshare is ordered after rideshare
Single corresponding stroke estimated distance characteristic information;First rideshare order with the second rideshare order is corresponding after rideshare multiplies estimation altogether
Distance feature information;First rideshare order and the second rideshare order are corresponding after rideshare to multiply estimation temporal characteristics information altogether;The
One history rideshare order is corresponding to estimate the distance feature information that detours;The corresponding estimation of second history rideshare order detours apart from special
Reference breath;First history rideshare order is corresponding to estimate the temporal characteristics information that detours;The corresponding estimation of second history rideshare order
Detour temporal characteristics information;First history rideshare order is corresponding to estimate the distance original corresponding with the first history rideshare order that detours
The characteristic information of the ratio between the journey that begins estimated distance;Second history rideshare order is corresponding to estimate detour distance and the second history rideshare
The characteristic information of the ratio between the corresponding original stroke estimated distance of order;Moment spy is welcomed the emperor in the corresponding estimation of second history rideshare order
Reference breath;Distance feature information is welcomed the emperor in the corresponding estimation of second history rideshare order;The corresponding estimation of second history rideshare order
It welcomes the emperor distance and estimates the characteristic information of the ratio between stroke distances.
In the present embodiment, initial sample characteristics information may include two classes, and one kind is the initial sample characteristics letter of mark class
It ceases (for example, characteristic information of ID class), another kind of is the initial sample characteristics information of non-identifying class (for example, the feature of non-ID class is believed
Breath).It is available after optimizing processing to the mark initial sample characteristics information of class and the initial sample characteristics information of non-identifying class
Target sample characteristic information, target sample characteristic information may include first object sample characteristics information, and the second target sample is special
Reference breath and third target sample characteristic information.
It in the present embodiment, can be in the following way respectively at the beginning of the mark initial sample characteristics information of class and non-identifying class
Beginning sample characteristics information optimizes processing: the target that the initial sample characteristics information input of non-identifying class is trained in advance integrates mould
In type, and the output result of target integrated model is normalized, to obtain first object sample characteristics information.It will be non-
The mark initial sample characteristics information of class is normalized, to obtain the second target sample characteristic information.It is initial class will to be identified
Sample characteristics information carries out sliding-model control, and is normalized, to obtain third target sample characteristic information.Wherein,
Target integrated model can be any reasonable integrated model, including but not limited to XGBoost (Extreme Gradient
Boosting, extreme gradient rise) model, the disclosure to not limiting in this respect.
In step 504, it according to the corresponding sample attribute of each history rideshare event and target sample characteristic information, treats
Trained linear regression model (LRM) and deep learning model to be trained carry out parameter adjustment, with obtain target linear regression model (LRM) with
And target deep learning model.
In the present embodiment, for each history rideshare event, corresponding target sample characteristic information can be distinguished defeated
Enter linear regression model (LRM) to be trained and deep learning model to be trained, by linear regression model (LRM) to be trained and to be trained
The result of deep learning model output is weighted addition, corresponding with reference to match parameter to obtain, and is based on each history rideshare
Event is corresponding to refer to match parameter and corresponding sample attribute, treats trained linear regression model (LRM) and depth to be trained
It spends learning model and carries out parameter adjustment.
Specifically, target linear regression model (LRM) and target deep learning model can be trained in the following way:
Firstly, obtaining the target sample characteristic information of a data set, each data set includes training set and verifying collection (wherein, training
The corresponding multiple history rideshare events of collection, the corresponding multiple history rideshare events of verifying collection).Then, using the target sample of training set
Characteristic information is adjusted the parameter of current linear regression model and deep learning model.Using the target sample of verifying collection
The linear regression model (LRM) and deep learning model that characteristic information trains front are verified.Until verification result satisfaction is wanted
It asks, using current linear regression model and deep learning model as trained target linear regression model (LRM) and target depth
Learning model.
Wherein, using the target sample characteristic information of training set to current linear regression model and deep learning model
Parameter be adjusted may include: by the target sample characteristic information of training set be separately input to current linear regression model (LRM) with
It is for each history rideshare event, current linear regression model (LRM) and deep learning model is defeated and in deep learning model
Probability value (probability of sample attribute that is, the corresponding sample attribute of history rideshare event is positive) out is weighted addition, obtains
To the corresponding reference parameter of history rideshare event.Reference parameter and history rideshare thing are corresponded to according to multiple history rideshare events
The corresponding sample attribute of part, obtains ROC curve.Corresponding AUC value is obtained based on ROC curve again.If AUC value is less than or waits
In preset threshold, then the parameter of current linear regression model and deep learning model is adjusted, then repeated to working as front
The step of parameter of property regression model and deep learning model is adjusted.If AUC value is greater than preset threshold, execution pair
Linear regression model (LRM) that front is trained and deep learning model carry out the step of verifying.
Wherein, the linear regression model (LRM) and deep learning front trained using the target sample characteristic information of verifying collection
Model carry out verifying may include: by the target sample characteristic information of training set be input to the linear regression model (LRM) that front is trained with
And in deep learning model, corresponding first AUC value is obtained.The target sample characteristic information of verifying collection is input to front training
Linear regression model (LRM) and deep learning model in, obtain corresponding second AUC value.The second AUC value is subtracted with the first AUC value
A difference is obtained, if the absolute value of this difference is greater than preset threshold, is repeated to linear regression model (LRM) and deep learning
The step of parameter of model is adjusted.If the absolute value of this difference is less than preset threshold, illustrate that verification result satisfaction is wanted
It asks.
In the present embodiment, target linear regression model (LRM) can be any reasonable linear regression model (LRM), target depth
Practising model (e.g., deep neural network CNN etc.) can be any reasonable deep learning model, and the disclosure is to unlimited in this respect
It is fixed.
The training method of the distribution model of the order provided by the above embodiment of the disclosure should by obtaining sample information
Sample information includes the corresponding related information of each history rideshare event in multiple history rideshare events, and is based on the sample information
Determine the corresponding sample attribute of each history rideshare event.The corresponding mesh of each history rideshare event is obtained based on the sample information
Standard specimen eigen information is treated trained according to the corresponding sample attribute of each history rideshare event and target sample characteristic information
Linear regression model (LRM) and deep learning model to be trained carry out parameter adjustment, to obtain target linear regression model (LRM) and mesh
Mark deep learning model.To obtain the model that can be used for rideshare Order splitting, so that rideshare order and service provider
Matching it is more reasonable, help to improve the utilization rate of Service Source.
In some optional embodiments, the above method further include: according to the corresponding sample category of each history rideshare event
Property and the initial sample characteristics information of corresponding non-identifying class train target integrated model.
It in the present embodiment, can be in the following way according to the corresponding sample attribute of each history rideshare event and correspondence
The initial sample characteristics information of non-identifying class train target integrated model: firstly, obtain a data set non-identifying class at the beginning of
Beginning sample characteristics information, each data set include training set and verifying collection (wherein, training set corresponds to multiple history rideshare events,
The corresponding multiple history rideshare events of verifying collection).Then, using the initial sample characteristics information of the non-identifying class of training set to current collection
It is adjusted at the parameter of model.The integrated model that front is trained using the non-identifying class initial sample characteristics information of verifying collection
It is verified.Until verification result is met the requirements, using current integrated model as trained target integrated model.
It should be noted that although describing the operation of method of disclosure in the accompanying drawings with particular order, this is not required that
Or hint must execute these operations in this particular order, or have to carry out operation shown in whole and be just able to achieve the phase
The result of prestige.On the contrary, the step of describing in flow chart can change and execute sequence.Additionally or alternatively, it is convenient to omit certain
Multiple steps are merged into a step and executed, and/or a step is decomposed into execution of multiple steps by step.
It is corresponding with the distribution of aforementioned order, the training method embodiment of model, the disclosure additionally provide order distribution,
The embodiment of the training device of model.
As shown in fig. 6, Fig. 6 is a kind of disclosure distributor block diagram of order shown according to an exemplary embodiment,
The device includes: first acquisition unit 601, second acquisition unit 602, input unit 603, output subelement 604 and distribution
Unit 605.
Wherein, first acquisition unit 601 are configured as obtaining target information, which includes service provider
Information, the service provider have connect the information of the first rideshare order, the information and current reality of the second rideshare order to be allocated
When information.
In the present embodiment, related service can be the rideshare service of the vehicles (for example, vehicle rideshare service
Deng), related scene can be to provide the service provider of vehicles Composite service to have been attached to a rideshare order, just
In the scene for waiting another rideshare order to be received.For example, being directed to the service of vehicle rideshare, service provider, which can be, carries passenger
Driver.First rideshare order is the rideshare order that service provider is connected to first, and the second rideshare order is rideshare to be allocated
Order.
In the present embodiment, the information of service provider may include that various can characterize service provider personal characteristics
Information.By taking the service of vehicle rideshare as an example, service provider is to provide the driver of vehicle rideshare service, and the information of service provider can
To include but is not limited to the id information of the driver, the gender information of the driver, the age information of the driver, the driver's is serviced
Divide information, the star information of the driver, the vehicle model information of the driver driving, the current location information etc. of the driver.
In the present embodiment, the information of the first rideshare order may include the various information for including in the first rideshare order,
For example, the information of the first rideshare order can include but is not limited to the corresponding start of a run of the first rideshare order and stroke end
Location information, the first rideshare order corresponding bill moment, corresponding user information of the first rideshare order etc..Second rideshare
The information of order may include the various information for including in the second rideshare order, for example, the information of the second rideshare order can wrap
The location information of the corresponding start of a run of the second rideshare order and stroke end is included but is not limited to, the second rideshare order is corresponding
Bill moment, corresponding user information of the second rideshare order etc..Wherein, user information can include but is not limited to the user's
Id information, the user of the user draw a portrait information (for example, gender information, age information, preference information, occupational information etc.) etc.
Deng.
In the present embodiment, current real time information can include but is not limited to current Weather information, current time
Information (for example, time information, week information, the date information of the Gregorian calendar, the date information of the lunar calendar, festival information etc.), current
Traffic related information etc..
Second acquisition unit 602 is configured as obtaining target signature information based on target information.
In the present embodiment, target information can be primarily based on and obtain initial characteristics information, initial characteristics information can wrap
Mark class initial characteristics information and non-identifying class initial characteristics information are included, then, respectively to above-mentioned mark class initial characteristics information
Processing is optimized with non-identifying class initial characteristics information, to obtain target signature information.
In the present embodiment, initial characteristics information may include the first initial characteristics information and the second initial characteristics letter
Breath, the first initial characteristics information are that can be directly based upon characteristic information obtained from target information, and the second initial characteristics information is
Characteristic information obtained from needing to estimate target information.Specifically, first can be directly extracted from target information
Initial characteristics information, also, being estimated based on target information (can use preset algorithm, strategy or model be estimated
Survey), to obtain the second initial characteristics information.
In the present embodiment, the first initial characteristics information may include following one or more: the gender of service provider
Characteristic information, the age characteristics information of service provider, the service of service provider divide characteristic information, the star of service provider
Characteristic information, the vehicle characteristic information that service provider drives, the present position characteristic information of service provider, weather are special
Reference breath, temporal characteristics information.
In the present embodiment, the second initial characteristics information may include following one or more: the first rideshare order is corresponding
Original stroke estimated distance characteristic information;The corresponding original stroke estimated distance characteristic information of second rideshare order;After rideshare
The corresponding stroke estimated distance characteristic information of first rideshare order;The corresponding stroke estimated distance of the second rideshare order is special after rideshare
Reference breath;First rideshare order with the second rideshare order is corresponding after rideshare multiplies estimated distance characteristic information altogether;First rideshare
Order and the second rideshare order are corresponding after rideshare to multiply estimation temporal characteristics information altogether;First rideshare order it is corresponding estimation around
Road distance feature information;Second rideshare order is corresponding to estimate the distance feature information that detours;The corresponding estimation of first rideshare order
Detour temporal characteristics information;Second rideshare order is corresponding to estimate the temporal characteristics information that detours;First rideshare order is corresponding to be estimated
Count the characteristic information of the ratio between the distance original stroke estimated distance corresponding with the first rideshare order that detours;Second rideshare order is corresponding
Estimation detour the characteristic information of the ratio between distance original stroke estimated distance corresponding with the second rideshare order;Second rideshare order
Moment characteristic information is welcomed the emperor in corresponding estimation;Distance feature information is welcomed the emperor in the corresponding estimation of second rideshare order;Second rideshare is ordered
Single corresponding estimation welcomes the emperor distance and estimates the characteristic information of the ratio between stroke distances.
In the present embodiment, initial characteristics information may include two classes, one kind for mark class initial characteristics information (for example,
The characteristic information of ID class), another kind of is non-identifying class initial characteristics information (for example, characteristic information of non-ID class).To mark class
After initial characteristics information and non-identifying class initial characteristics information optimize processing, available target signature information, target spy
Reference breath may include first object characteristic information, the second target signature information and third target signature information.
It in the present embodiment, can be initially special to mark class initial characteristics information and non-identifying class respectively in the following way
Reference breath optimizes processing: in the target integrated model that non-identifying class initial characteristics information input is trained in advance, object set
It can be by (each leaf node on a leaf node of non-identifying class initial characteristics information MAP to every decision tree at model
A corresponding weighted value), using these leaf nodes as destination node, using the corresponding weighted value of destination node as characteristic value,
Available non-identifying class initial characteristics information integrates the character representation of the model space as the defeated of target integrated model in target
Out as a result, and the output result of target integrated model is normalized, to obtain first object characteristic information.It will be nonstandard
Know class initial characteristics information to be normalized, to obtain the second target signature information.Will mark class initial characteristics information into
Row sliding-model control, and be normalized, to obtain third target signature information.Wherein, target integrated model can be
Any reasonable integrated model, including but not limited to XGBoost (Extreme Gradient Boosting, in extreme gradient
Rise) model, the disclosure to not limiting in this respect.
Input unit 603 is configured as inputting target signature information into target linear regression model (LRM) and target depth respectively
Learning model.
In the present embodiment, target linear regression model (LRM) can be any reasonable linear regression model (LRM), target depth
Practising model (e.g., deep neural network CNN etc.) can be any reasonable deep learning model, and the disclosure is to unlimited in this respect
It is fixed.
Subelement 604 is exported, the result for exporting target linear regression model (LRM) and target deep learning model is configured as
It is weighted addition, to obtain match parameter.
In the present embodiment, target signature information can be inputted to target linear regression model (LRM) and target depth study respectively
Target linear regression model (LRM) is weighted with the result that target deep learning model exports and is added by model, to obtain matching ginseng
Number.The match parameter can characterize the matching degree of the second rideshare order and service provider.If the match parameter be greater than or
Equal to preset threshold, then illustrate that the second rideshare order matches with service provider.
Allocation unit 605 is configured as when match parameter is more than or equal to preset threshold, by the second rideshare order point
Service provider described in dispensing.
The distributor of the order provided by the above embodiment of the disclosure is based on target information by obtaining target information
Target signature information is obtained, target signature information is inputted to target linear regression model (LRM) and target deep learning model respectively, it will
Target linear regression model (LRM) is weighted with the result that target deep learning model exports and is added, to obtain match parameter.If
The match parameter is more than or equal to preset threshold, then by the second rideshare Order splitting to service provider.Wherein, which believes
Breath includes the information of service provider, and service provider has connect the information of the first rideshare order, the second rideshare order to be allocated
Information and current real time information.Due to the side combined using target linear regression model (LRM) and target deep learning model
Formula determines the matching degree of rideshare order and service provider, therefore, so that the matching of rideshare order and service provider is more
Rationally, the utilization rate of Service Source is also improved.
In some optional embodiments, second acquisition unit 602 may include: to obtain subelement and processing subelement
(not shown).
Wherein, subelement is obtained, is configured as obtaining mark class initial characteristics information and non-identifying class according to target information
Initial characteristics information.
Subelement is handled, is configured to carry out mark class initial characteristics information and non-identifying class initial characteristics information
Optimization processing, to obtain target signature information.
The distributor of the order provided by the above embodiment of the disclosure, by obtaining target information, according to target information
Mark class initial characteristics information and non-identifying class initial characteristics information is obtained, respectively to mark class initial characteristics information and non-identifying
Class initial characteristics information optimizes processing, to obtain target signature information.It is linear that target signature information is inputted to target respectively
Regression model and target deep learning model carry out the result that target linear regression model (LRM) and target deep learning model export
Weighting summation, to obtain match parameter.If the match parameter is more than or equal to preset threshold, by the second rideshare order point
Dispensing service provider.Wherein, which includes the information of service provider, and service provider has connect the first rideshare order
Information, the information and current real time information of the second rideshare order to be allocated.Due to respectively to mark class initial characteristics
Processing is optimized in information and non-identifying class initial characteristics information, to obtain target signature information, then linear using target
The mode that regression model and target deep learning model combine, determines the matching degree of rideshare order and service provider, because
This improves efficiency of service further such that the matching of rideshare order and service provider is more reasonable, also improves service money
The utilization rate in source.
In other optional embodiments, target signature information may include first object characteristic information, the second target
Characteristic information and third target signature information.
Processing subelement is configured for: the target integrated model that non-identifying class initial characteristics information input is trained in advance
In, and the output result of target integrated model is normalized, to obtain first object characteristic information.By non-identifying class
Initial characteristics information is normalized, to obtain the second target signature information.Will mark class initial characteristics information carry out from
Dispersion processing, and be normalized, to obtain third target signature information.
In other optional embodiments, the information of the first rideshare order may include: that the first rideshare order is corresponding
The location information of start of a run and stroke end;The first rideshare order corresponding bill moment.
The information of second rideshare order may include: the position of the corresponding start of a run of the second rideshare order and stroke end
Confidence breath;The second rideshare order corresponding bill moment.
It should be appreciated that above-mentioned apparatus can be preset in the server, can also be loaded by modes such as downloadings
In server.Corresponding units in above-mentioned apparatus can be cooperated with the unit in server to realize the distribution side of order
Case.
As shown in fig. 7, Fig. 7 is a kind of disclosure training of the distribution model of order shown according to an exemplary embodiment
Device block diagram, the device include: first acquisition unit 701, determination unit 702, second acquisition unit 703 and adjustment unit
704。
Wherein, first acquisition unit 701 are configured as obtaining sample information, which includes multiple history rideshares
The corresponding related information of each history rideshare event in event.
In the present embodiment, for any history rideshare event, which be can wrap
Include the corresponding real time information of history rideshare event, the information of service provider, the service provider in the history rideshare event
The information for the first history rideshare order being first connected in the history rideshare event and the second history rideshare order being followed by
Information.
In the present embodiment, the information of service provider may include that various can characterize service provider personal characteristics
Information.By taking the service of vehicle rideshare as an example, service provider is to provide the driver of vehicle rideshare service, and the information of service provider can
To include but is not limited to the id information of the driver, the gender information of the driver, the age information of the driver, the driver's is serviced
Divide information, the star information of the driver, the vehicle model information of the driver driving, location information of the driver etc..
In the present embodiment, the information of the first history rideshare order may include include each in the first history rideshare order
Kind information, for example, the information of the first history rideshare order can include but is not limited to the corresponding stroke of the first history rideshare order
The location information of starting point and stroke end, the first history rideshare order corresponding bill moment, the first history rideshare order pair
User information answered etc..The information of second history rideshare order may include the various letters for including in the second history rideshare order
Breath, for example, the information of the second history rideshare order can include but is not limited to the corresponding start of a run of the second history rideshare order
And the location information of stroke end, at the second history rideshare order corresponding bill moment, the second history rideshare order is corresponding
User information etc..Wherein, user information can include but is not limited to the id information of the user, user's portrait information of the user
(for example, gender information, age information, preference information, occupational information etc.) etc..
In the present embodiment, which can include but is not limited to the history rideshare thing
Weather information when part occurs, temporal information is (for example, time information, week information, the date information of the Gregorian calendar, the date of the lunar calendar
Information, festival information etc.), traffic related information etc..
Determination unit 702 is configured as determining the corresponding sample attribute of each history rideshare event based on sample information, should
Sample attribute includes positive sample attribute and negative sample attribute.
In the present embodiment, it can determine that each history is closed according to the evaluation information and response situation of order in sample information
Multiply the corresponding sample attribute of event.For example, order evaluation is preferable, the corresponding sample of the successful history rideshare event of order response
Attribute can be positive sample attribute.Order evaluation is poor, the corresponding sample attribute of history rideshare event of order answer failed
Can be negative sample attribute.The disclosure does not limit the specific division mode of positive and negative sample attribute.
Second acquisition unit 703 is configured as obtaining the corresponding target sample of each history rideshare event based on sample information
Eigen information.
In the present embodiment, the sample information can be primarily based on and obtain the corresponding initial sample of each history rideshare event
Characteristic information, initial sample characteristics information may include the mark initial sample characteristics information of class and the initial sample characteristics of non-identifying class
Then information respectively optimizes the initial sample characteristics information of above-mentioned mark class and the initial sample characteristics information of non-identifying class
Processing, to obtain target sample characteristic information.
In the present embodiment, for any history rideshare event, the corresponding initial sample characteristics letter of the history rideshare event
Breath may include the corresponding first initial sample characteristics information of the history rideshare event and corresponding second initial sample characteristics
Information.First initial sample characteristics information is that can be directly based upon characteristic information obtained from sample information, the second initial sample
Characteristic information is characteristic information obtained from needing to estimate sample information.It specifically, can be straight from sample information
It connects and extracts the first initial sample characteristics information, also, estimated based on sample information and (preset algorithm, strategy can be used
Or model is estimated), to obtain the second initial sample characteristics information.
In the present embodiment, which may include following one or more: the history rideshare
The sex character information of the corresponding service provider of event, the age characteristics letter of the corresponding service provider of history rideshare event
Breath, the service of the corresponding service provider of history rideshare event divide characteristic information, which mentions
The star characteristic information of supplier, the vehicle characteristic information that the corresponding service provider of the history rideshare event drives, the history are closed
Multiply the position characteristic information of the corresponding service provider of event, which should
The corresponding temporal characteristics information of history rideshare event.
In the present embodiment, which may include following one or more: the first history is closed
Multiply the corresponding original stroke estimated distance characteristic information of order;The corresponding original stroke estimated distance of second history rideshare order is special
Reference breath;The corresponding stroke estimated distance characteristic information of first history rideshare order after rideshare;The second history rideshare is ordered after rideshare
Single corresponding stroke estimated distance characteristic information;First rideshare order with the second rideshare order is corresponding after rideshare multiplies estimation altogether
Distance feature information;First rideshare order and the second rideshare order are corresponding after rideshare to multiply estimation temporal characteristics information altogether;The
One history rideshare order is corresponding to estimate the distance feature information that detours;The corresponding estimation of second history rideshare order detours apart from special
Reference breath;First history rideshare order is corresponding to estimate the temporal characteristics information that detours;The corresponding estimation of second history rideshare order
Detour temporal characteristics information;First history rideshare order is corresponding to estimate the distance original corresponding with the first history rideshare order that detours
The characteristic information of the ratio between the journey that begins estimated distance;Second history rideshare order is corresponding to estimate detour distance and the second history rideshare
The characteristic information of the ratio between the corresponding original stroke estimated distance of order;Moment spy is welcomed the emperor in the corresponding estimation of second history rideshare order
Reference breath;Distance feature information is welcomed the emperor in the corresponding estimation of second history rideshare order;The corresponding estimation of second history rideshare order
It welcomes the emperor distance and estimates the characteristic information of the ratio between stroke distances.
In the present embodiment, initial sample characteristics information may include two classes, and one kind is the initial sample characteristics letter of mark class
It ceases (for example, characteristic information of ID class), another kind of is the initial sample characteristics information of non-identifying class (for example, the feature of non-ID class is believed
Breath).It is available after optimizing processing to the mark initial sample characteristics information of class and the initial sample characteristics information of non-identifying class
Target sample characteristic information, target sample characteristic information may include first object sample characteristics information, and the second target sample is special
Reference breath and third target sample characteristic information.
It in the present embodiment, can be in the following way respectively at the beginning of the mark initial sample characteristics information of class and non-identifying class
Beginning sample characteristics information optimizes processing: the target that the initial sample characteristics information input of non-identifying class is trained in advance integrates mould
In type, and the output result of target integrated model is normalized, to obtain first object sample characteristics information.It will be non-
The mark initial sample characteristics information of class is normalized, to obtain the second target sample characteristic information.It is initial class will to be identified
Sample characteristics information carries out sliding-model control, and is normalized, to obtain third target sample characteristic information.Wherein,
Target integrated model can be any reasonable integrated model, including but not limited to XGBoost (Extreme Gradient
Boosting, extreme gradient rise) model, the disclosure to not limiting in this respect.
Adjustment unit 704 is configured as according to the corresponding sample attribute of each history rideshare event and target sample feature
Information, treats trained linear regression model (LRM) and deep learning model to be trained carries out parameter adjustment, linear to obtain target
Regression model and target deep learning model.
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare
The first history rideshare that the information of service provider, the service provider are first connected in the history rideshare event in event is ordered
The information of single information and the second history rideshare order being followed by.
In the present embodiment, for each history rideshare event, corresponding target sample characteristic information can be distinguished defeated
Enter linear regression model (LRM) to be trained and deep learning model to be trained, by linear regression model (LRM) to be trained and to be trained
The result of deep learning model output is weighted addition, corresponding with reference to match parameter to obtain, and is based on each history rideshare
Event is corresponding to refer to match parameter and corresponding sample attribute, treats trained linear regression model (LRM) and depth to be trained
It spends learning model and carries out parameter adjustment.
Specifically, target linear regression model (LRM) and target deep learning model can be trained in the following way:
Firstly, obtaining the target sample characteristic information of a data set, each data set includes training set and verifying collection (wherein, training
The corresponding multiple history rideshare events of collection, the corresponding multiple history rideshare events of verifying collection).Then, using the target sample of training set
Characteristic information is adjusted the parameter of current linear regression model and deep learning model.Using the target sample of verifying collection
The linear regression model (LRM) and deep learning model that characteristic information trains front are verified.Until verification result satisfaction is wanted
It asks, using current linear regression model and deep learning model as trained target linear regression model (LRM) and target depth
Learning model.
Wherein, using the target sample characteristic information of training set to current linear regression model and deep learning model
Parameter be adjusted may include: by the target sample characteristic information of training set be separately input to current linear regression model (LRM) with
It is for each history rideshare event, current linear regression model (LRM) and deep learning model is defeated and in deep learning model
Probability value (probability of sample attribute that is, the corresponding sample attribute of history rideshare event is positive) out is weighted addition, obtains
To the corresponding reference parameter of history rideshare event.Reference parameter and history rideshare thing are corresponded to according to multiple history rideshare events
The corresponding sample attribute of part, obtains ROC curve.Corresponding AUC value is obtained based on ROC curve again.If AUC value is less than or waits
In preset threshold, then the parameter of current linear regression model and deep learning model is adjusted, then repeated to working as front
The step of parameter of property regression model and deep learning model is adjusted.If AUC value is greater than preset threshold, execution pair
Linear regression model (LRM) that front is trained and deep learning model carry out the step of verifying.
Wherein, the linear regression model (LRM) and deep learning front trained using the target sample characteristic information of verifying collection
Model carry out verifying may include: by the target sample characteristic information of training set be input to the linear regression model (LRM) that front is trained with
And in deep learning model, corresponding first AUC value is obtained.The target sample characteristic information of verifying collection is input to front training
Linear regression model (LRM) and deep learning model in, obtain corresponding second AUC value.The second AUC value is subtracted with the first AUC value
A difference is obtained, if the absolute value of this difference is greater than preset threshold, is repeated to linear regression model (LRM) and deep learning
The step of parameter of model is adjusted.If the absolute value of this difference is less than preset threshold, illustrate that verification result satisfaction is wanted
It asks.
In the present embodiment, target linear regression model (LRM) can be any reasonable linear regression model (LRM), target depth
Practising model (e.g., deep neural network CNN etc.) can be any reasonable deep learning model, and the disclosure is to unlimited in this respect
It is fixed.
The training method of the distribution model of the order provided by the above embodiment of the disclosure should by obtaining sample information
Sample information includes the corresponding related information of each history rideshare event in multiple history rideshare events, and is based on the sample information
Determine the corresponding sample attribute of each history rideshare event.The corresponding mesh of each history rideshare event is obtained based on the sample information
Standard specimen eigen information is treated trained according to the corresponding sample attribute of each history rideshare event and target sample characteristic information
Linear regression model (LRM) and deep learning model to be trained carry out parameter adjustment, to obtain target linear regression model (LRM) and mesh
Mark deep learning model.To obtain the model that can be used for rideshare Order splitting, so that rideshare order and service provider
Matching it is more reasonable, help to improve the utilization rate of Service Source.
In other optional embodiments, adjustment unit 704 is configured for: it is directed to each history rideshare event, it will
Corresponding target sample characteristic information inputs linear regression model (LRM) to be trained and deep learning model to be trained respectively.It will be to
The result of trained linear regression model (LRM) and deep learning model output to be trained is weighted addition, to obtain corresponding ginseng
Examine match parameter.It is corresponding with reference to match parameter and corresponding sample attribute based on each history rideshare event, treat training
Linear regression model (LRM) and deep learning model to be trained carry out parameter adjustment.
In other optional embodiments, for any history rideshare event, second acquisition unit can be by as follows
Mode is based on sample information and obtains the corresponding target sample characteristic information of history rideshare event: according to the history in sample information
The corresponding related information of rideshare event obtains the initial sample characteristics information of the corresponding mark class of history rideshare event and non-identifying
The initial sample characteristics information of class.The mark initial sample characteristics information of class and the initial sample characteristics information of non-identifying class are carried out respectively
Optimization processing, to obtain target sample characteristic information.
In other optional embodiments, target sample characteristic information may include first object sample characteristics information,
Second target sample characteristic information and third target sample characteristic information.
Second acquisition unit 703 can be in the following way respectively to the mark initial sample characteristics information of class and non-identifying class
Initial sample characteristics information optimizes processing, to obtain target sample characteristic information: the initial sample characteristics of non-identifying class are believed
In breath input target integrated model trained in advance, and the output result of target integrated model is normalized, with
To first object sample characteristics information.The initial sample characteristics information of non-identifying class is normalized, to obtain the second mesh
Standard specimen eigen information.The initial sample characteristics information of class will be identified and carry out sliding-model control, and be normalized, to obtain
Third target sample characteristic information.
In other optional embodiments, which can also include: training unit (not shown).Wherein, it instructs
Practice unit, is configured as according to the corresponding sample attribute of each history rideshare event and the corresponding initial sample characteristics of non-identifying class
Information trains target integrated model.
It in the present embodiment, can be in the following way according to the corresponding sample attribute of each history rideshare event and correspondence
The initial sample characteristics information of non-identifying class train target integrated model: firstly, obtain a data set non-identifying class at the beginning of
Beginning sample characteristics information, each data set include training set and verifying collection (wherein, training set corresponds to multiple history rideshare events,
The corresponding multiple history rideshare events of verifying collection).Then, using the initial sample characteristics information of the non-identifying class of training set to current collection
It is adjusted at the parameter of model.The integrated model that front is trained using the non-identifying class initial sample characteristics information of verifying collection
It is verified.Until verification result is met the requirements, using current integrated model as trained target integrated model.
In other optional embodiments, the information of the first history rideshare order may include: that the first history rideshare is ordered
The location information of single corresponding start of a run and stroke end;The first history rideshare order corresponding bill moment.
Second history rideshare order can include: the corresponding start of a run of the second history rideshare order and stroke with information
The location information of terminal;The second history rideshare order corresponding bill moment.
It should be appreciated that above-mentioned apparatus can be preset in the server, can also be loaded by modes such as downloadings
In server.Corresponding units in above-mentioned apparatus can be cooperated with the unit in server to realize the distribution model of order
Training program.
For device embodiment, since it corresponds essentially to embodiment of the method, so related place is referring to method reality
Apply the part explanation of example.The apparatus embodiments described above are merely exemplary, wherein described be used as separation unit
The unit of explanation may or may not be physically separated, and component shown as a unit can be or can also be with
It is not physical unit, it can it is in one place, or may be distributed over multiple network units.It can be according to actual
The purpose for needing to select some or all of the modules therein to realize disclosure scheme.Those of ordinary skill in the art are not paying
Out in the case where creative work, it can understand and implement.
It is (including but unlimited that the storage medium for wherein including program code in one or more can be used in the embodiment of the present disclosure
In magnetic disk storage, CD-ROM, optical memory etc.) on the form of computer program product implemented.
Correspondingly, the embodiment of the present disclosure also provides a kind of computer storage medium, program is stored in the storage medium and is referred to
It enables, which includes:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect first
The information and current real time information of the information of rideshare order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, with
Obtain match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is given to the clothes
Be engaged in provider.
Correspondingly, the embodiment of the present disclosure also provides a kind of computer storage medium, program is stored in the storage medium and is referred to
It enables, which includes:
Obtain sample information, the sample information include in multiple history rideshare events each history rideshare event it is corresponding
Related information;
The corresponding sample attribute of each history rideshare event, the sample attribute packet are determined based on the sample information
Include positive sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained line is treated
Property regression model and deep learning model to be trained carry out parameter adjustment, it is deep to obtain target linear regression model (LRM) and target
Spend learning model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare
The first history rideshare that the information of service provider, the service provider are first connected in the history rideshare event in event is ordered
The information of single information and the second history rideshare order being followed by.
Being described in unit module involved in the embodiment of the present disclosure can be realized by way of software, can also be led to
The mode of hardware is crossed to realize.Described unit module also can be set in the processor, for example, can be described as: a kind of
Processor includes acquiring unit, determination unit and allocation unit.Wherein, the title of these unit modules is under certain conditions simultaneously
The restriction to the unit module itself is not constituted, for example, determination unit is also described as " for determining the second rideshare order
With the whether matched unit of service provider ".
As on the other hand, the disclosure additionally provides a kind of computer readable storage medium, the computer-readable storage medium
Matter can be computer readable storage medium included in device described in above-described embodiment;It is also possible to individualism, not
The computer readable storage medium being fitted into terminal or server.The computer-readable recording medium storage has one or one
A procedure above, the program be used to execute by one or more than one processor be described in the order of the disclosure distribution,
The training method of model.
Computer-usable storage medium includes permanent and non-permanent, removable and non-removable media, can be by appointing
What method or technique realizes that information stores.Information can be computer readable instructions, data structure, the module of program or other
Data.The example of the storage medium of computer includes but is not limited to: phase change memory (PRAM), static random access memory
(SRAM), dynamic random access memory (DRAM), other kinds of random access memory (RAM), read-only memory
(ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory techniques, CD-ROM are read-only
Memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape magnetic disk storage or
Other magnetic storage devices or any other non-transmission medium, can be used for storage can be accessed by a computing device information.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to its of the disclosure
Its embodiment.The disclosure is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by following
Claim is pointed out.
It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present disclosure is only limited by the accompanying claims.
Claims (24)
1. a kind of distribution method of order, which is characterized in that the described method includes:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect the first rideshare
The information and current real time information of the information of order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, to obtain
Match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is mentioned to the service
Supplier.
2. the method according to claim 1, wherein described obtain target signature letter based on the target information
Breath, comprising:
Mark class initial characteristics information and non-identifying class initial characteristics information are obtained according to the target information;
Processing is optimized to the mark class initial characteristics information and non-identifying class initial characteristics information respectively, to obtain target
Characteristic information.
3. according to the method described in claim 2, it is characterized in that, the target signature information includes first object feature letter
Breath, the second target signature information and third target signature information;
It is described that processing is optimized to the mark class initial characteristics information and non-identifying class initial characteristics information respectively, to obtain
Target signature information, comprising:
In the target integrated model that the non-identifying class initial characteristics information input is trained in advance, and mould is integrated to the target
The output result of type is normalized, to obtain first object characteristic information;
The non-identifying class initial characteristics information is normalized, to obtain the second target signature information;
The mark class initial characteristics information is subjected to sliding-model control, and is normalized, it is special to obtain third target
Reference breath.
4. method according to claim 1 to 3, which is characterized in that the information of the first rideshare order includes:
The location information of the corresponding start of a run of the first rideshare order and stroke end;The corresponding hair of the first rideshare order
Single moment;
The information of the second rideshare order includes: the position of the corresponding start of a run of the second rideshare order and stroke end
Confidence breath;The second rideshare order corresponding bill moment.
5. a kind of training method of the distribution model of order, which is characterized in that the described method includes:
Sample information is obtained, the sample information includes the corresponding association of each history rideshare event in multiple history rideshare events
Information;
Determine that the corresponding sample attribute of each history rideshare event, the sample attribute include just based on the sample information
Sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained linear return is treated
Model and deep learning model to be trained is returned to carry out parameter adjustment, to obtain target linear regression model (LRM) and target depth
Practise model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare event
The first history rideshare order that the information of middle service provider, the service provider are first connected in the history rideshare event
The information of information and the second history rideshare order being followed by.
6. according to the method described in claim 5, it is characterized in that, described according to the corresponding sample of each history rideshare event
This attribute and target sample characteristic information, treat trained linear regression model (LRM) and deep learning model to be trained is joined
Number adjustment, comprising:
For each history rideshare event, corresponding target sample characteristic information is inputted into linear regression model (LRM) to be trained respectively
And deep learning model to be trained;
The result of the linear regression model (LRM) to be trained and deep learning model output to be trained is weighted addition, with
It obtains corresponding with reference to match parameter;
It is corresponding with reference to match parameter and corresponding sample attribute based on each history rideshare event, treat trained linear time
Model and deep learning model to be trained is returned to carry out parameter adjustment.
7. according to the method described in claim 5, it is characterized in that, being directed to any history rideshare event, in the following way base
The corresponding target sample characteristic information of history rideshare event is obtained in the sample information:
According to the corresponding related information of history rideshare event in the sample information, the corresponding mark of history rideshare event is obtained
Know the initial sample characteristics information of class and the initial sample characteristics information of non-identifying class;
Processing is optimized to the initial sample characteristics information of mark class and the initial sample characteristics information of non-identifying class respectively, with
Obtain target sample characteristic information.
8. the method according to the description of claim 7 is characterized in that the target sample characteristic information includes first object sample
Characteristic information, the second target sample characteristic information and third target sample characteristic information;
It is described that place is optimized to the initial sample characteristics information of mark class and the initial sample characteristics information of non-identifying class respectively
Reason, to obtain target sample characteristic information, comprising:
In the target integrated model that the non-identifying initial sample characteristics information input of class is trained in advance, and to the object set
It is normalized at the output result of model, to obtain first object sample characteristics information;
The non-identifying initial sample characteristics information of class is normalized, to obtain the second target sample characteristic information;
The initial sample characteristics information of the mark class is subjected to sliding-model control, and is normalized, to obtain third mesh
Standard specimen eigen information.
9. according to the method described in claim 8, it is characterized in that, the method also includes:
According to each corresponding sample attribute of history rideshare event and the initial sample characteristics information instruction of corresponding non-identifying class
Practise the target integrated model.
10. according to the method any in claim 5-9, which is characterized in that the information of the first history rideshare order
It include: the location information of the corresponding start of a run of the first history rideshare order and stroke end;First history is closed
Multiply the order corresponding bill moment;
The information of the second history rideshare order includes: the corresponding start of a run of the second history rideshare order and stroke
The location information of terminal;The second history rideshare order corresponding bill moment.
11. a kind of distributor of order, which is characterized in that described device includes:
First acquisition unit is configured as obtaining target information, and the target information includes the information of service provider, the clothes
Business provider has connect the information of the first rideshare order, the information and current real time information of the second rideshare order to be allocated;
Second acquisition unit is configured as obtaining target signature information based on the target information;
Input unit is configured as inputting the target signature information into target linear regression model (LRM) and target depth study respectively
Model;
Subelement is exported, the result for exporting the target linear regression model (LRM) and the target deep learning model is configured as
It is weighted addition, to obtain match parameter;
Allocation unit is configured as when the match parameter is more than or equal to preset threshold, by the second rideshare order
Distribute to the service provider.
12. device according to claim 11, which is characterized in that the second acquisition unit includes:
Subelement is obtained, is configured as obtaining mark class initial characteristics information according to the target information and non-identifying class is initially special
Reference breath;
Subelement is handled, is configured to carry out the mark class initial characteristics information and non-identifying class initial characteristics information
Optimization processing, to obtain target signature information.
13. device according to claim 12, which is characterized in that the target signature information includes first object feature letter
Breath, the second target signature information and third target signature information;
The processing subelement is configured for:
In the target integrated model that the non-identifying class initial characteristics information input is trained in advance, and mould is integrated to the target
The output result of type is normalized, to obtain first object characteristic information;
The non-identifying class initial characteristics information is normalized, to obtain the second target signature information;
The mark class initial characteristics information is subjected to sliding-model control, and is normalized, it is special to obtain third target
Reference breath.
14. any device in 1-13 according to claim 1, which is characterized in that the packet of the first rideshare order
It includes: the location information of the corresponding start of a run of the first rideshare order and stroke end;The first rideshare order is corresponding
The bill moment;
The information of the second rideshare order includes: the position of the corresponding start of a run of the second rideshare order and stroke end
Confidence breath;The second rideshare order corresponding bill moment.
15. a kind of training device of the distribution model of order, which is characterized in that described device includes:
First acquisition unit is configured as obtaining sample information, and the sample information includes each in multiple history rideshare events
The corresponding related information of history rideshare event;
Determination unit is configured as determining the corresponding sample attribute of each history rideshare event based on the sample information,
The sample attribute includes positive sample attribute and negative sample attribute;
Second acquisition unit is configured as obtaining the corresponding target sample of each history rideshare event based on the sample information
Eigen information;
Adjustment unit is configured as according to the corresponding sample attribute of each history rideshare event and target sample feature letter
Breath, treats trained linear regression model (LRM) and deep learning model to be trained carries out parameter adjustment, is linearly returned with obtaining target
Return model and target deep learning model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare event
The first history rideshare order that the information of middle service provider, the service provider are first connected in the history rideshare event
The information of information and the second history rideshare order being followed by.
16. device according to claim 15, which is characterized in that the adjustment unit is configured for:
For each history rideshare event, corresponding target sample characteristic information is inputted into linear regression model (LRM) to be trained respectively
And deep learning model to be trained;
The result of the linear regression model (LRM) to be trained and deep learning model output to be trained is weighted addition, with
It obtains corresponding with reference to match parameter;
It is corresponding with reference to match parameter and corresponding sample attribute based on each history rideshare event, treat trained linear time
Model and deep learning model to be trained is returned to carry out parameter adjustment.
17. device according to claim 15, which is characterized in that be directed to any history rideshare event, described second obtains
Unit is based on the sample information in the following way and obtains the corresponding target sample characteristic information of history rideshare event:
According to the corresponding related information of history rideshare event in the sample information, the corresponding mark of history rideshare event is obtained
Know the initial sample characteristics information of class and the initial sample characteristics information of non-identifying class;
Processing is optimized to the initial sample characteristics information of mark class and the initial sample characteristics information of non-identifying class respectively, with
Obtain target sample characteristic information.
18. device according to claim 17, which is characterized in that the target sample characteristic information includes first object sample
Eigen information, the second target sample characteristic information and third target sample characteristic information;
The second acquisition unit is in the following way respectively at the beginning of the initial sample characteristics information of mark class and non-identifying class
Beginning sample characteristics information optimizes processing, to obtain target sample characteristic information:
In the target integrated model that the non-identifying initial sample characteristics information input of class is trained in advance, and to the object set
It is normalized at the output result of model, to obtain first object sample characteristics information;
The non-identifying initial sample characteristics information of class is normalized, to obtain the second target sample characteristic information;
The initial sample characteristics information of the mark class is subjected to sliding-model control, and is normalized, to obtain third mesh
Standard specimen eigen information.
19. device according to claim 18, which is characterized in that described device further include:
Training unit is configured as according at the beginning of each corresponding sample attribute of history rideshare event and corresponding non-identifying class
Beginning sample characteristics information trains the target integrated model.
20. any device in 5-19 according to claim 1, which is characterized in that the letter of the first history rideshare order
Breath includes: the location information of the corresponding start of a run of the first history rideshare order and stroke end;First history
The rideshare order corresponding bill moment;
The information of the second history rideshare order includes: the corresponding start of a run of the second history rideshare order and stroke
The location information of terminal;The second history rideshare order corresponding bill moment.
21. a kind of computer storage medium, program instruction is stored in the storage medium, which is characterized in that described instruction packet
It includes:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect the first rideshare
The information and current real time information of the information of order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, to obtain
Match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is mentioned to the service
Supplier.
22. a kind of computer storage medium, program instruction is stored in the storage medium, which is characterized in that described instruction packet
It includes:
Sample information is obtained, the sample information includes the corresponding association of each history rideshare event in multiple history rideshare events
Information;
Determine that the corresponding sample attribute of each history rideshare event, the sample attribute include just based on the sample information
Sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained linear return is treated
Model and deep learning model to be trained is returned to carry out parameter adjustment, to obtain target linear regression model (LRM) and target depth
Practise model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare event
The first history rideshare order that the information of middle service provider, the service provider are first connected in the history rideshare event
The information of information and the second history rideshare order being followed by.
23. a kind of electronic equipment, comprising:
Processor is adapted for carrying out each instruction;And
Equipment is stored, is suitable for storing a plurality of instruction, described instruction is suitable for being loaded and being executed by processor:
Target information is obtained, the target information includes that the information of service provider, the service provider have connect the first rideshare
The information and current real time information of the information of order, the second rideshare order to be allocated;
Target signature information is obtained based on the target information;
The target signature information is inputted into target linear regression model (LRM) and target deep learning model respectively;
The target linear regression model (LRM) is weighted with the result that the target deep learning model exports and is added, to obtain
Match parameter;
If the match parameter is more than or equal to preset threshold, the second rideshare Order splitting is mentioned to the service
Supplier.
24. a kind of electronic equipment, comprising:
Processor is adapted for carrying out each instruction;And
Equipment is stored, is suitable for storing a plurality of instruction, described instruction is suitable for being loaded and being executed by processor:
Sample information is obtained, the sample information includes the corresponding association of each history rideshare event in multiple history rideshare events
Information;
Determine that the corresponding sample attribute of each history rideshare event, the sample attribute include just based on the sample information
Sample attribute and negative sample attribute;
The corresponding target sample characteristic information of each history rideshare event is obtained based on the sample information;
According to the corresponding sample attribute of each history rideshare event and target sample characteristic information, trained linear return is treated
Model and deep learning model to be trained is returned to carry out parameter adjustment, to obtain target linear regression model (LRM) and target depth
Practise model;
Wherein, for any history rideshare event, corresponding related information includes corresponding real time information, the history rideshare event
The first history rideshare order that the information of middle service provider, the service provider are first connected in the history rideshare event
The information of information and the second history rideshare order being followed by.
Priority Applications (10)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710457389.6A CN109146211A (en) | 2017-06-16 | 2017-06-16 | The distribution of order, the training method of model and device |
SG11201811535RA SG11201811535RA (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
EP18815095.7A EP3459025A4 (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
AU2018282300A AU2018282300B2 (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
CA3028215A CA3028215C (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
CN201880002585.7A CN109478275B (en) | 2017-06-16 | 2018-06-15 | System and method for distributing service requests |
PCT/CN2018/091534 WO2018228541A1 (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
JP2018566885A JP6797943B2 (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
CA3072656A CA3072656A1 (en) | 2017-06-16 | 2018-06-15 | Systems and methods for allocating service requests |
US16/232,044 US11631027B2 (en) | 2017-06-16 | 2018-12-25 | Systems and methods for allocating service requests |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710457389.6A CN109146211A (en) | 2017-06-16 | 2017-06-16 | The distribution of order, the training method of model and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109146211A true CN109146211A (en) | 2019-01-04 |
Family
ID=64830499
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710457389.6A Pending CN109146211A (en) | 2017-06-16 | 2017-06-16 | The distribution of order, the training method of model and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109146211A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110163655A (en) * | 2019-04-15 | 2019-08-23 | 中国平安人寿保险股份有限公司 | Distribution method of attending a banquet, device, equipment and storage medium based on gradient boosted tree |
CN111027710A (en) * | 2019-12-06 | 2020-04-17 | 浙江大搜车软件技术有限公司 | Resource clue allocation method, device, computer equipment and storage medium |
WO2020155837A1 (en) * | 2019-01-31 | 2020-08-06 | 阿里巴巴集团控股有限公司 | Ride-hailing information display method, apparatus, system, and device |
CN111861497A (en) * | 2019-04-15 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Data processing method and device, computer equipment and storage medium |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101355714A (en) * | 2007-07-24 | 2009-01-28 | 梁宇杰 | System and method for real time pooling vehicle |
US20100280884A1 (en) * | 2009-04-30 | 2010-11-04 | Uri Levine | Automated carpool matching |
US20110184977A1 (en) * | 2008-09-27 | 2011-07-28 | Jiachun Du | Recommendation method and system based on collaborative filtering |
CN103325046A (en) * | 2012-03-19 | 2013-09-25 | 上海博路信息技术有限公司 | Real-time car-sharing system based on HTML5 |
CN104599002A (en) * | 2015-02-13 | 2015-05-06 | 北京嘀嘀无限科技发展有限公司 | Order value predicting method and equipment |
CN105117777A (en) * | 2015-07-28 | 2015-12-02 | 北京嘀嘀无限科技发展有限公司 | Order distributing method and apparatus |
CN105894359A (en) * | 2016-03-31 | 2016-08-24 | 百度在线网络技术(北京)有限公司 | Order pushing method, device and system |
CN106803197A (en) * | 2015-11-26 | 2017-06-06 | 滴滴(中国)科技有限公司 | Spell single method and apparatus |
-
2017
- 2017-06-16 CN CN201710457389.6A patent/CN109146211A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101355714A (en) * | 2007-07-24 | 2009-01-28 | 梁宇杰 | System and method for real time pooling vehicle |
US20110184977A1 (en) * | 2008-09-27 | 2011-07-28 | Jiachun Du | Recommendation method and system based on collaborative filtering |
US20100280884A1 (en) * | 2009-04-30 | 2010-11-04 | Uri Levine | Automated carpool matching |
CN103325046A (en) * | 2012-03-19 | 2013-09-25 | 上海博路信息技术有限公司 | Real-time car-sharing system based on HTML5 |
CN104599002A (en) * | 2015-02-13 | 2015-05-06 | 北京嘀嘀无限科技发展有限公司 | Order value predicting method and equipment |
CN105117777A (en) * | 2015-07-28 | 2015-12-02 | 北京嘀嘀无限科技发展有限公司 | Order distributing method and apparatus |
CN106803197A (en) * | 2015-11-26 | 2017-06-06 | 滴滴(中国)科技有限公司 | Spell single method and apparatus |
CN105894359A (en) * | 2016-03-31 | 2016-08-24 | 百度在线网络技术(北京)有限公司 | Order pushing method, device and system |
Non-Patent Citations (1)
Title |
---|
张亦楠: "出租车合乘模式下的智能匹配问题的研究与实现", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020155837A1 (en) * | 2019-01-31 | 2020-08-06 | 阿里巴巴集团控股有限公司 | Ride-hailing information display method, apparatus, system, and device |
TWI784195B (en) * | 2019-01-31 | 2022-11-21 | 開曼群島商創新先進技術有限公司 | Car-calling information display method, device, system and equipment |
CN110163655A (en) * | 2019-04-15 | 2019-08-23 | 中国平安人寿保险股份有限公司 | Distribution method of attending a banquet, device, equipment and storage medium based on gradient boosted tree |
CN111861497A (en) * | 2019-04-15 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Data processing method and device, computer equipment and storage medium |
CN110163655B (en) * | 2019-04-15 | 2024-03-05 | 中国平安人寿保险股份有限公司 | Agent distribution method, device and equipment based on gradient lifting tree and storage medium |
CN111027710A (en) * | 2019-12-06 | 2020-04-17 | 浙江大搜车软件技术有限公司 | Resource clue allocation method, device, computer equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110390108B (en) | Task type interaction method and system based on deep reinforcement learning | |
Chen et al. | DNNOff: offloading DNN-based intelligent IoT applications in mobile edge computing | |
CN109146211A (en) | The distribution of order, the training method of model and device | |
CN106791235B (en) | A kind of method, apparatus and system selecting service agent | |
CN104346366B (en) | Extend the method and apparatus of test data | |
CN108734317A (en) | Net about vehicle order information processing method and processing device | |
CN108734293A (en) | Task management system, method and apparatus | |
CN111143525A (en) | Vehicle information acquisition method and device and intelligent vehicle moving system | |
CN107516090A (en) | Integrated face identification method and system | |
CN107741976A (en) | Intelligent response method, apparatus, medium and electronic equipment | |
CN110750523A (en) | Data annotation method, system, computer equipment and storage medium | |
JP7207309B2 (en) | Program, information processing method, and information processing apparatus | |
CN109313540A (en) | The two stages training of spoken dialogue system | |
CN108090645A (en) | For the method and device of coordination service resource | |
CN109895781A (en) | Method for controlling a vehicle and device | |
CN109766418A (en) | Method and apparatus for output information | |
CN109947989A (en) | Method and apparatus for handling video | |
CN109146109A (en) | The distribution of order, the training method of model and device | |
CN110290169A (en) | A kind of Service Order response method and device | |
CN103095784B (en) | A kind of cloud user mapped system and method | |
CN111159344A (en) | Robot response method, device, equipment and storage medium | |
CN110390577A (en) | The distribution method and device of order | |
CN110390406A (en) | Reserve the distribution method and device of order | |
CN109992659A (en) | Method and apparatus for text sequence | |
CN109978380A (en) | A kind of workflow processing method and device |
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 | ||
AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20220401 |