CN106484766B - Searching method and device based on artificial intelligence - Google Patents
Searching method and device based on artificial intelligence Download PDFInfo
- Publication number
- CN106484766B CN106484766B CN201610808298.8A CN201610808298A CN106484766B CN 106484766 B CN106484766 B CN 106484766B CN 201610808298 A CN201610808298 A CN 201610808298A CN 106484766 B CN106484766 B CN 106484766B
- Authority
- CN
- China
- Prior art keywords
- search result
- historical
- feature vector
- training
- statistical nature
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application discloses searching methods and device based on artificial intelligence.One specific embodiment of this method includes: to receive the searching request of the search type comprising active user's input, and determine the corresponding search result of search type；Based on the associated historical operation information of operation carried out with user to search result and statistical nature information associated with the statistical nature of search result, search result is ranked up；Search result after sequence is pushed to active user.It realizes for the individualized feature of search result to be combined from different sort algorithms based on ranking results of the statistical nature to search result and search result is ranked up, on the basis of being based on statistical feature to search result progress whole sequence by different sort algorithms, the result integrally to sort is adjusted based on individualized feature, so that ranking results while relatively accurately reflecting integrated demand, meet user to the individual demand of search result.
Description
Technical field
This application involves computer fields, and in particular to search field, more particularly to the searching method based on artificial intelligence
And device.
Background technique
The fast development of artificial intelligence technology (Artificial Intelligence, abbreviation AI) is the daily work of people
Make and life is provided convenience.Artificial intelligence be research, develop intelligent theory for simulating, extending and extend people, method,
One new technological sciences of technology and application system.Artificial intelligence is a branch of computer science, it attempts to understand intelligence
The essence of energy, and a kind of new intelligence machine that can be made a response in such a way that human intelligence is similar is produced, which grinds
Study carefully including robot, language identification, image recognition, natural language processing and expert system etc..Artificial intelligence is melted more and more
Enter into application, can analyze the personalized demand of user in conjunction with the application of artificial intelligence, by the desired answer of user
Such as it is expected that the search result obtained feeds back to user with more forward order.
User is when website (such as group buying websites) scans for, generally according to the statistical feature pair of such as pageview
Search result (such as group buying voucher) is ranked up.However, since such as user for not accounting for user wishes to buy browsing again
The personalized demand of lower group buying voucher is measured, the ranking results for causing the user of different demands to obtain are same ranking results,
User is unable to satisfy to the individual demand of search result.
Invention information
This application provides a kind of searching method and device based on artificial intelligence, for solving above-mentioned background technology part
?.
In a first aspect, being used this method comprises: receiving comprising current this application provides the searching method based on artificial intelligence
The searching request of the search type of family input, and determine the corresponding search result of search type；Based on user to search result into
The associated historical operation information of operation gone and statistical nature information associated with the statistical nature of search result, to searching
Hitch fruit is ranked up；Search result after sequence is pushed to active user.
Second aspect, this application provides the searcher based on artificial intelligence, which includes: receiving unit, configuration
For receiving the searching request of the search type comprising active user's input, and the corresponding search result of determining search type；Sequence
Unit is configured to based on the associated historical operation information of the operation that carried out with user to search result and and search result
The associated statistical nature information of statistical nature, search result is ranked up；Push unit will be configured to after sorting
Search result is pushed to active user.
Searching method and device provided by the present application based on artificial intelligence, by receiving searching comprising active user's input
Cable-styled searching request, and determine the corresponding search result of search type；Based on the operation carried out with user to search result
Associated historical operation information and statistical nature information associated with the statistical nature of search result carry out search result
Sequence；Search result after sequence is pushed to active user.It realizes the individualized feature of search result and different rows
Sequence algorithm is combined based on ranking results of the statistical nature to search result and is ranked up to search result, by different
On the basis of sort algorithm is based on statistics feature to the whole sequence of search result progress, arranged based on individualized feature adjustment entirety
Sequence as a result, so that ranking results while relatively accurately reflecting integrated demand, meet user to search result
Individual demand.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the application's is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 shows the exemplary system frame that can be applied to the searching method or device based on artificial intelligence of the application
Structure；
Fig. 2 shows the flow charts according to one embodiment of the searching method based on artificial intelligence of the application；
Fig. 3 shows the flow chart of another embodiment of the searching method based on artificial intelligence according to the application；
Fig. 4 shows an exemplary architecture figure of the searching method based on artificial intelligence suitable for the application；
Fig. 5 shows the structural schematic diagram of one embodiment of the searcher based on artificial intelligence according to the application；
Fig. 6 shows the computer system for being suitable for the searcher based on artificial intelligence for being used to realize the embodiment of the present application
Structural schematic diagram.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to related invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 shows the example that can be applied to the embodiment of the searching method or device based on artificial intelligence of the application
Property system architecture 100.
As shown in Figure 1, system architecture 100 may include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 to provide the medium of transmission link.Network 104 can be with
Including various connection types, such as wired, wireless transmission link or fiber optic cables etc..
User can be used terminal device 101,102,103 and be interacted by network 104 with server 105, to receive or send out
Send message etc..Various communication applications can be installed on terminal device 101,102,103, such as purchase by group that class application, searching class answers
With, map class application etc..
Terminal device 101,102,103 can be with display screen and support the various electronic equipments of network communication, packet
Include but be not limited to smart phone, tablet computer, E-book reader, MP3 player (Moving Picture Experts
Group Audio Layer III, dynamic image expert's compression standard audio level 3), MP4 (Moving Picture
Experts Group Audio Layer IV, dynamic image expert's compression standard audio level 4) it is player, on knee portable
Computer and desktop computer etc..
Server 105 can receive the searching request of the transmission of terminal device 101,102,103, to the search result found out
It is ranked up, the search result after sequence is sent to terminal device 101,102,103.
Referring to FIG. 2, it illustrates the streams according to one embodiment of the searching method based on artificial intelligence of the application
Journey 200.It should be noted that the searching method based on artificial intelligence provided by the embodiment of the present application can be by the clothes in Fig. 1
Business device 105 executes.Method includes the following steps:
Step 201, the searching request of the search type comprising active user's input is received, and determines that search type is corresponding and searches
Hitch fruit.
In the present embodiment, active user can refer to the user of current input search type.For example, active user is vertical
It can input include searching for dining room type first in the input frame that group buying websites provide when class website group buying websites scan for
It is cable-styled.At this point it is possible to the searching request of the search type comprising user's input be received, it is then possible to determine that search type is corresponding
Search result, i.e., the group buying voucher in multiple dining rooms.
Step 202, it is based on historical operation information and statistical nature information, search result is ranked up.
In the present embodiment, the searching request that reception includes the search type of active user's input is being received by step 201,
It, can be associated according to the operation carried out with user to search result and after determining the corresponding search result of search type
Historical operation information and statistical nature information associated with the statistical nature of search result, are ranked up search result.
In some optional implementations of the present embodiment, search result includes: the Internet resources of vertical class website, behaviour
Work includes: browse operation, clicking operation, purchase operation, comment operation, and statistical nature information includes: pageview, click volume, purchase
The amount of buying, comment amount.
In the present embodiment, search result can be the Internet resources of vertical class website.For example, when user is at vertical class station
When the group buying voucher in point group buying websites search dining room, search result can be the group buying voucher of multiple dining room businessmans.
In the present embodiment, the operation that user carries out search result can include but is not limited to: browse operation clicks behaviour
Make, purchase operation, comment operation.The statistical nature information of search result can include but is not limited to: pageview, click volume, purchase
The amount of buying, comment amount.
By taking group buying voucher of the user in vertical class website group buying websites search dining room as an example, the corresponding historical operation of search result
Information can be a set.Group buying voucher with user in vertical class website group buying websites search dining room, search result are multiple
For purchasing by group volume, the corresponding historical operation information of each group buying voucher can be a set, may include user couple in set
The mark of the operation carried out of one group buying voucher sequentially, operation may include following one or more: browse, click, buying,
Comment.For example, a group buying voucher when the user clicks, has purchased a group buying voucher, is commented on, user thinks to buy the group again
Purchase certificate has again tapped on group buying voucher, and it sequentially includes: to click behaviour in the set that the historical operation information of user, which can be a set,
It makes a check mark, buy operation mark, comment operation mark, clicking operation mark.The statistical nature information of historical search result can be with
For the pageview of group buying voucher, click volume, purchase volume, comment amount.
It in the present embodiment, can be special by the statistics of the historical operation information of user and the statistical nature for indicating search result
Reference ceases the element as search results ranking, is ranked up to search result.For example, historical operation information and system can be directed to
Different weights is arranged in meter characteristic information, for the parameters in each operation and statistical nature information of historical operation information
Different rights is set, search result is ranked up.
Step 203, the search result after sequence is pushed to active user.
In the present embodiment, by step 202 be based on historical operation information and statistical nature information, to search result into
After row sequence, the search result after sequence can be pushed to active user.
Referring to FIG. 3, it illustrates according to another embodiment of the searching method based on artificial intelligence of the application
Process 300.It should be noted that the searching method based on artificial intelligence provided by the embodiment of the present application can be by Fig. 1
Server 105 executes.Method includes the following steps:
Step 301, the searching request of the search type comprising active user's input is received, and determines that search type is corresponding and searches
Hitch fruit.
In the present embodiment, when active user scans for, search can be inputted in the input area that website provides first
Formula.For example, user is when vertical class website group buying websites scan for, it can be defeated in the input frame that group buying websites provide first
Entering includes the search type of dining room type.At this point it is possible to receive the searching request of the search type comprising active user's input, so
Afterwards, the corresponding search result of search type, i.e., the group buying voucher in multiple dining rooms can be determined.
Step 302, historical operation information and statistical nature information are based on using default machine learning model, to search result
It is ranked up.
In the present embodiment, before the searching request for receiving the search type comprising active user's input by step 301,
Machine learning model can be pre-created, default machine learning model can be gbrank model.It is then possible to gbrank mould
Type is trained.Statistical nature information using the gbrank model after training based on historical operation information and search result is right
Search result is ranked up.
In the present embodiment, gbrank model can be trained in the following ways: obtains multiple historical search knots
The statistical nature information of the corresponding historical operation information of fruit and historical search result；Each historical operation information and statistics is special
Reference breath is combined, and obtains multiple training samples；Multiple training samples are combined two-by-two in a manner of pairwise, are obtained
Multiple training samples pair；Generate respectively training sample to each of the corresponding personalized training feature vector of training sample,
Personalized training feature vector includes: the component for indicating historical operation information, the component for indicating statistical nature information；It generates respectively
Training sample to each of the corresponding whole training feature vector of training sample, whole training feature vector include at least one
A component, wherein representation in components predetermined order model obtained after being ranked up to historical search result with historical search result
The associated parameter in position, the corresponding predetermined order model of each component；The each of training sample centering is generated respectively
The training input vector of the corresponding gbrank model of a training sample, training input vector include: personalized training feature vector,
Whole training feature vector；Based on training input vector, gbrank model is trained.
In the present embodiment, the corresponding historical operation information of available multiple historical search results and each history are searched
The statistical nature information of hitch fruit.
By taking group buying voucher of the user in vertical class website group buying websites search dining room as an example, historical search result can be history
The group buying voucher searched out in search.One historical search result can correspond to a historical operation information.Historical search result pair
The historical operation information answered can be a set, may include user in set and purchase by group to one searched out in historical search
The mark of the operation carried out of certificate sequentially, operation may include following one or more: browsing is clicked, purchase, comment.Example
Such as, a group buying voucher when the user clicks, has purchased a group buying voucher, is commented on, user thinks to buy the group buying voucher again again
Secondary to click group buying voucher, it sequentially includes: clicking operation mark in the set that the historical operation information of user, which can be a set,
Know, purchase operation mark, comment operation mark, clicking operation mark.The statistical nature information of historical search result can be group
Purchase pageview, click volume, purchase volume, the comment amount of certificate.
In the present embodiment, the corresponding historical operation information of multiple historical search results and each historical search are being obtained
As a result after statistical nature information, the corresponding historical operation information of historical search result and the historical operation can be believed respectively
The statistical nature information for ceasing corresponding group buying voucher is combined, and obtains multiple training samples.It is wrapped simultaneously in each training sample
Statistical nature information containing a historical search result corresponding historical operation information and the historical search result.
After obtaining multiple training samples, multiple training samples can be carried out by group using pairwise mode two-by-two
It closes, obtains training sample pair.Pairwise mode can be equivalent to selects a sample first in the sample, then selects one
It is a that compared to the sample, more preferably sample forms training sample pair.
For example, a group buying voucher when the user clicks, is only browsed and do not bought, user again taps on group buying voucher merger
When having bought group buying voucher, the historical operation information of user can be a set, sequentially include: in the set clicking operation mark,
Clicking operation mark, purchase operation mark.In another example a group buying voucher when the user clicks, has purchased a group buying voucher, carry out
Comment, user think that buying the group buying voucher again has again tapped on group buying voucher, and the historical operation information of user can be a set,
It sequentially include: clicking operation mark, purchase operation mark, comment operation mark, clicking operation mark in the set.
Can by comprising being identified by clicking operation, clicking operation mark, the historical operation information that forms of purchase operation mark
Sample with comprising identified, buy by clicking operation operation mark, comment operation mark, clicking operation mark form history behaviour
Make the sample composition training sample pair of information.In the training sample pair, comprising by clicking operation mark, purchase operation mark,
Comment on operation mark, the sample for the historical operation information that clicking operation mark forms is better than comprising being identified, being clicked by clicking operation
The sample for the historical operation information that operation mark, purchase operation mark form.
In the present embodiment, multiple training samples are being combined in a manner of pairwise two-by-two, are obtaining training sample
To the corresponding personalized training feature vector of training sample centering each training sample and whole special later, can be generated respectively
It levies in vector.
In the present embodiment, can by each training sample the corresponding historical operation information of historical search result and
The statistical nature information of historical search result is utilized respectively vector expression, will indicate that the corresponding historical operation of historical search result is believed
The vector sum of breath indicates the vector of the statistical nature information of historical search result as the corresponding personalized instruction of historical search result
The component for practicing feature vector obtains the corresponding personalized training feature vector of multiple historical search results.
In the present embodiment, multiple components be may include in the corresponding whole training feature vector of training sample, each
Component obtains related to the position of historical search result after being ranked up for a predetermined order model to historical search result
The parameter of connection.For example, the parameter that smart order models obtain after being ranked up to multiple historical search results is instruction after sequence
The score of the position of historical search result, it is each after being ranked up using smart order models to multiple historical search results
The corresponding score of a historical search result.The parameter that thick order models obtain after being ranked up to historical search result is instruction
The gear in the location of historical search result section is being ranked up multiple historical search results using thick order models
Afterwards, the corresponding gear of each historical search result, the gear indicate that historical search result is gone through in thick order models to multiple
The location of historical search result section after history search result is ranked up.
In the present embodiment, the corresponding personalized training feature vector of each historical search result and whole is being generated respectively
It, can be respectively by the corresponding personalized training feature vector of each historical search result and entirety after body training feature vector
Component of the training feature vector as the training input vector of the corresponding gbrank model of historical search result, obtains each and goes through
The training input vector of the corresponding gbrank model of history search result.It is then possible to by multiple trained input vectors with
Pairwise mode is sequentially inputted in gbrank model.In input each time, input two separately includes property one by one
The training input vector of training feature vector is special with comprising a personalized training corresponding better than the property training feature vector
The training input vector of the corresponding vector of personalized training characteristics of sign, is trained gbrank model.
Group buying voucher with user in vertical class website group buying websites search dining room, gbrank model can be according to whole training
Feature vector study is ranked up group buying voucher from entirety, meanwhile, learn user's out according to personalized training feature vector
Which kind of operation is more likely to purchase group buying voucher, to adjust the whole position for being ranked up middle part group buying voucher to group buying voucher.
In the present embodiment, the gbrank model pair after being trained to gbrank model, after can use training
The corresponding search result of search type of active user's input is ranked up.I.e. using the gbrank model after training to passing through step
The corresponding search result of search type of the 301 active user's inputs received is ranked up.
In the present embodiment, the corresponding search result of search type of active user's input can be carried out in the following ways
Sequence: generating the corresponding individualized feature vector of each search result in search result respectively, and individualized feature vector includes:
Indicate the component of the component of the corresponding historical operation information of search result, the statistical nature information for indicating search result；It gives birth to respectively
At the corresponding global feature vector of each search result in search result, global feature vector includes at least one component,
In, ginseng associated with the position of search result that representation in components predetermined order model obtains after being ranked up to search result
Number, the corresponding predetermined order model of each component；It is corresponding that each search result in search result is generated respectively
The input vector of gbrank model, input vector include: individualized feature vector, global feature vector；Utilize gbrank model
Based on input vector, exported as a result, exporting the position that result is the search result after sequence.
In the present embodiment, the corresponding search of search type of each active user's input determined by step 301
As a result corresponding historical operation information can be a set.With user's purchasing by group in vertical class website group buying websites search dining room
Certificate, search result be it is multiple purchase by group volume for, the corresponding historical operation information of each group buying voucher can for one set, set
In may include the mark of the operation carried out of user sequentially to a group buying voucher, operation may include with the next item down or more
: browsing is clicked, purchase, comment.For example, a group buying voucher when the user clicks, has purchased a group buying voucher, is commented on,
User thinks that buying the group buying voucher again has again tapped on group buying voucher, and the historical operation information of user can be a set, the collection
It sequentially include: clicking operation mark, purchase operation mark, comment operation mark, clicking operation mark in conjunction.Historical search result
Statistical nature information can be group buying voucher pageview, click volume, purchase volume, comment amount.
It in the present embodiment, can be by the statistics of the corresponding historical operation information of each search result and search result spy
Reference breath is utilized respectively vector expression, and the vector sum for indicating the corresponding historical operation information of search result is indicated search result
Component of the vector of statistical nature information as the corresponding personalized training feature vector of search result, obtains multiple search results
Corresponding personalization training feature vector.
In the present embodiment, the corresponding search of search type of each active user's input determined by step 301
As a result multiple components be may include in corresponding global feature vector, each component is that a predetermined order model ties search
The parameter associated with the position of search result that fruit obtains after being ranked up.For example, smart order models are to multiple search results
The parameter obtained after being ranked up is the score of instruction position of search result after sequence, in the smart order models of utilization to multiple
After search result is ranked up, the corresponding score of each search result.Thick order models are ranked up search result
The parameter obtained afterwards is the gear for indicating the location of search result section, in the thick order models of utilization to multiple search results
After being ranked up, the corresponding gear of each search result, the gear indicates that search result is searched in thick order models to multiple
The location of search result section after hitch fruit is ranked up.
In the present embodiment, the corresponding personalized training feature vector of each search result and whole instruction are being generated respectively
It, can be respectively by the corresponding personalized training feature vector of each search result and whole training characteristics after practicing feature vector
Component of the vector as the input vector of the corresponding gbrank model of search result, it is corresponding to obtain each search result
The input vector of gbrank model.It is then possible to which multiple input vectors are input in gbrank model, output result is obtained.
Export result be each search result using gbrank model sequence after the location of, the position can using input to
Amount is indicated.
Gbrank model can carry out search result from entirety according to the corresponding whole training feature vector of search result
Sequence determines which kind of operation of user is more likely to purchase and purchases by group according to the corresponding personalized training feature vector of search result
Certificate, to adjust the whole position for being ranked up middle partial search results to search result.
With user vertical class website group buying websites search dining room group buying voucher, search result be it is multiple purchase by group volume for,
Gbrank model on the whole can carry out multiple group buying vouchers according to the corresponding whole training feature vector of each group buying voucher
Sequence.Meanwhile determining which kind of operation of user is more likely to purchase according to the corresponding personalized training feature vector of each group buying voucher
Group buying voucher is bought, to adjust the whole position that multiple group buying vouchers are ranked up with middle part group buying voucher.For example, when according to whole training
When feature vector is integrally ranked up, the group buying voucher of the pageview of the group buying voucher in a dining room due to being less than another dining room it is clear
The amount of looking at, after coming to another dining room.But it, can be true according to the corresponding component of historical operation information in individualized feature vector
Determine user once bought the dining room group buying voucher and and commented on the group buying voucher, at this point it is possible to purchasing by group the family dining room
The position of certificate is adjusted to before the position of the group buying voucher in another dining room.
Step 303, the search result after sequence is pushed to active user.
In the present embodiment, historical operation information and statistics are being based on using default machine learning model by step 302
Search result after sequence after being ranked up to search result, can be pushed to active user by characteristic information.
Referring to FIG. 4, it illustrates an exemplary framves of the searching method based on artificial intelligence for being suitable for the application
Composition.
In fig. 4 it is shown that unified order models, smart order models, thick order models, other order models.
The group buying voucher in dining room is searched in vertical class website group buying websites with user, includes that multiple group buying vouchers are in search result
Example, unified order models can be gbrank model.Gbrank model can according to global feature vector from entirety to group buying voucher
It is ranked up, meanwhile, according to individualized feature vector, to adjust the whole position for being ranked up rear part group buying voucher to group buying voucher
It sets, obtains output result.
The corresponding individualized feature vector of each group buying voucher can be generated respectively.The corresponding personalization of each group buying voucher
Feature vector includes: the component of the statistical nature information of the component for indicating historical operation information, expression group buying voucher.Meanwhile it can be with
The corresponding global feature vector of each group buying voucher is generated respectively, and global feature vector includes multiple components.Component can be essence
Order models, thick order models, other order models are obtained after being ranked up based on the statistical nature information of group buying voucher to group buying voucher
The parameter arrived.The parameter that smart order models obtain after being ranked up to multiple group buying vouchers is the position of instruction group buying voucher after sequence
Score, after being ranked up using smart order models to multiple group buying vouchers, the corresponding score of each group buying voucher.Thick row
The parameter that sequence model obtains after being ranked up to group buying voucher is the gear for indicating the location of group buying voucher section, is utilizing thick row
After sequence model is ranked up multiple group buying vouchers, the corresponding gear of each group buying voucher, the gear indicates that group buying voucher is slightly being arranged
The location of group buying voucher section after sequence model sorts to multiple group buying voucher rows.
After generating the corresponding property feature vector of each group buying voucher and global feature vector respectively, it can be generated every
The input vector of the corresponding gbrank model of one group buying voucher.The input vector packet of the corresponding gbrank model of each group buying voucher
Containing the corresponding property feature vector of group buying voucher and global feature vector.
Gbrank model can be based on multiple input vectors, be arranged from entirety group buying voucher according to global feature vector
Sequence, meanwhile, it is obtained according to individualized feature vector to adjust the whole position for being ranked up rear part group buying voucher to group buying voucher
Export result.Output result is the position of group buying voucher after sequence, which can be indicated using input vector.
Referring to FIG. 5, it illustrates the knots according to one embodiment of the searcher based on artificial intelligence of the application
Structure schematic diagram.The Installation practice is corresponding with embodiment of the method shown in Fig. 2.
As shown in figure 5, the searcher 500 based on artificial intelligence of the present embodiment includes: receiving unit 501, sequence is single
Member 502, push unit 503.Wherein, receiving unit 501 is configured to receive the search of the search type comprising active user's input
Request, and determine the corresponding search result of search type；Sequencing unit 502 is configured to be based on carrying out search result with user
The associated historical operation information of operation crossed and statistical nature information associated with the statistical nature of search result, to search
As a result it is ranked up；Push unit 503 is configured to the search result after sequence being pushed to active user.
In some optional implementations of the present embodiment, search result includes: the Internet resources of vertical class website, behaviour
Work includes: browse operation, clicking operation, purchase operation, comment operation, and statistical nature information includes: pageview, click volume, purchase
The amount of buying, comment amount.
In some optional implementations of the present embodiment, sequencing unit 502 includes: that individualized feature vector generates son
Unit is configured to generate the corresponding individualized feature vector of each search result, individualized feature in search result respectively
Vector includes: point of the statistical nature information of the component for indicating the corresponding historical operation information of search result, expression search result
Amount；Global feature vector generates subelement, is configured to generate the corresponding entirety of each search result in search result respectively
Feature vector, global feature vector include at least one component, wherein representation in components predetermined order model carries out search result
Parameter associated with search result position that is being obtained after sequence, the corresponding predetermined order model of each component；Input
Vector generates subelement, is configured to generate the corresponding default machine learning model of each search result in search result respectively
Input vector, input vector includes: individualized feature vector, global feature vector；Subelement is exported, is configured to using pre-
If machine learning model is based on input vector, exported as a result, exporting the position that result is the search result after sequence.
In some optional implementations of the present embodiment, device 500 further include: training unit (not shown), configuration
For when default machine learning model is gbrank model, including the searching request of the search type of active user's input in reception
Before, the statistical nature information of multiple historical search results corresponding historical operation information and historical search result is obtained；It will be every
One historical operation information and statistical nature information are combined, and obtain multiple training samples；It will be multiple in a manner of pairwise
Training sample is combined two-by-two, obtains multiple training samples pair；Respectively generate training sample to each of training sample
Corresponding personalization training feature vector, personalized training feature vector include: the component for indicating historical operation information, indicate system
Count the component of characteristic information；Generate respectively training sample to each of the corresponding whole training feature vector of training sample,
Global feature vector includes at least one component, wherein representation in components predetermined order model is ranked up historical search result
The parameter associated with the position of historical search result obtained afterwards, the corresponding predetermined order model of each component；Respectively
Generate training sample to each of the corresponding gbrank model of training sample training input vector, training input vector packet
It includes: personalized training feature vector, whole training feature vector；Based on training input vector, gbrank model is trained.
In some optional implementations of the present embodiment, predetermined order model includes: smart order models, slightly sort mould
Type, the parameter that smart order models obtain after being ranked up to search result or historical search result are that knot is searched in instruction after sequence
The score of the position of fruit or historical search result, thick order models obtain after being ranked up to search result or historical search result
Parameter be indicate the location of search result or historical search result section gear.
Fig. 6 shows the computer system for being suitable for the searcher based on artificial intelligence for being used to realize the embodiment of the present application
Structural schematic diagram.
As shown in fig. 6, computer system 600 includes central processing unit (CPU) 601, it can be read-only according to being stored in
Program in memory (ROM) 602 or be loaded into the program in random access storage device (RAM) 603 from storage section 608 and
Execute various movements appropriate and processing.In RAM603, also it is stored with system 600 and operates required various programs and data.
CPU601, ROM602 and RAM603 are connected with each other by bus 604.Input/output (I/O) interface 605 is also connected to bus
604。
I/O interface 605 is connected to lower component: the importation 606 including keyboard, mouse etc.；It is penetrated including such as cathode
The output par, c 607 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.；Storage section 608 including hard disk etc.；
And the communications portion 609 of the network interface card including LAN card, modem etc..Communications portion 609 via such as because
The network of spy's net executes communication process.Driver 610 is also connected to I/O interface 605 as needed.Detachable media 611, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 610, in order to read from thereon
Computer program be mounted into storage section 608 as needed.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be tangibly embodied in machine readable
Computer program on medium, the computer program include the program code for method shown in execution flow chart.At this
In the embodiment of sample, which can be downloaded and installed from network by communications portion 609, and/or from removable
Medium 611 is unloaded to be mounted.
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the application, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it wants
It is noted that the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart, Ke Yiyong
The dedicated hardware based system of defined functions or operations is executed to realize, or can be referred to specialized hardware and computer
The combination of order is realized.
As on the other hand, present invention also provides a kind of nonvolatile computer storage media, the non-volatile calculating
Machine storage medium can be nonvolatile computer storage media included in equipment described in above-described embodiment；It is also possible to
Individualism, without the nonvolatile computer storage media in supplying terminal.Above-mentioned nonvolatile computer storage media is deposited
One or more program is contained, when one or more of programs are executed by an equipment, so that the equipment: receiving
The searching request of search type comprising active user's input, and determine the corresponding search result of described search formula；Based on with
The associated historical operation information of operation that family carried out described search result and with the statistical nature phase of described search result
Associated statistical nature information, is ranked up described search result；Search result after sequence is pushed to the current use
Family.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from the inventive concept, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination is closed and other technical solutions of formation.Such as features described above and (but being not limited to) disclosed herein have it is similar
The technical characteristic of function is replaced mutually and the technical solution that is formed.
Claims (8)
1. a kind of searching method based on artificial intelligence, which is characterized in that the described method includes:
The searching request of the search type comprising active user's input is received, and determines the corresponding search result of described search formula；
Based on the associated historical operation information of described search result was carried out with user operation and with described search result
The associated statistical nature information of statistical nature, described search result is ranked up；
Described search result after sequence is pushed to the active user；
Wherein, it is described based on the associated historical operation information of described search result was carried out with user operation and with it is described
The associated statistical nature information of the statistical nature of search result, is ranked up described search result, comprising:
Respectively generate described search result in the corresponding individualized feature vector of each search result, the individualized feature to
Amount includes: the component of the statistical nature information of the component for indicating the corresponding historical operation information of search result, expression search result；
The corresponding global feature vector of each search result in described search result, the global feature vector packet are generated respectively
Include at least one component, wherein the representation in components predetermined order model obtained after being ranked up to described search result with
The associated parameter in the position of search result, the corresponding predetermined order model of each component；
The input vector of the corresponding default machine learning model of each search result in described search result is generated respectively, it is described
Input vector includes: individualized feature vector, global feature vector；
It is based on the input vector using default machine learning model, is exported as a result, the output result is after sorting
The position of search result.
2. the method according to claim 1, wherein described search result includes: the network money of vertical class website
Source, the operation include: browse operation, clicking operation, purchase operation, comment operation, and the statistical nature information includes: browsing
Amount, click volume, purchase volume, comment amount.
3. the method according to claim 1, wherein the default machine learning model be gbrank model,
Before the searching request for receiving the search type comprising active user's input, the method also includes:
Obtain the statistical nature information of multiple historical search results corresponding historical operation information and historical search result；
Each described historical operation information and the statistical nature information are combined, multiple training samples are obtained；
Multiple training samples are combined two-by-two in a manner of pairwise, obtain multiple training samples pair；
Generate respectively the training sample to each of the corresponding personalized training feature vector of training sample, the individual character
Changing training feature vector includes: the component for indicating the historical operation information, the component for indicating the statistical nature information；
Generate respectively the training sample to each of the corresponding whole training feature vector of training sample, the whole instruction
Practice feature vector include at least one component, wherein the representation in components predetermined order model to the historical search result into
The parameter associated with the position of historical search result obtained after row sequence, the corresponding predetermined order mould of each component
Type；
Generate respectively the training sample to each of the corresponding gbrank model of training sample training input vector, institute
Stating trained input vector includes: personalized training feature vector, whole training feature vector；
Based on the trained input vector, the gbrank model is trained.
4. according to the method described in claim 3, it is characterized in that, the predetermined order model includes: smart order models, thick row
Sequence model, the parameter that smart order models obtain after being ranked up to search result or historical search result are that instruction is searched after sequence
The score of the position of hitch fruit or historical search result, after thick order models are ranked up search result or historical search result
Obtained parameter is the gear for indicating the location of search result or historical search result section.
5. a kind of searcher based on artificial intelligence, which is characterized in that described device includes:
Receiving unit is configured to receive the searching request of the search type comprising active user's input, and determines described search
The corresponding search result of formula；
Sequencing unit is configured to based on the associated historical operation information of the operation that carried out with user to described search result
With statistical nature information associated with the statistical nature of described search result, described search result is ranked up；
Push unit is configured to the described search result after sequence being pushed to the active user；
Wherein, the sequencing unit includes:
Individualized feature vector generates subelement, and it is corresponding to be configured to generate each search result in described search result respectively
Individualized feature vector, the individualized feature vector include: the component for indicating the corresponding historical operation information of search result,
Indicate the component of the statistical nature information of search result；
Global feature vector generates subelement, and it is corresponding to be configured to generate each search result in described search result respectively
Global feature vector, the global feature vector include at least one component, wherein the representation in components predetermined order model pair
The parameter associated with the position of search result that described search result obtains after being ranked up, each component corresponding one pre-
If order models；
Input vector generates subelement, and it is corresponding default to be configured to generate each search result in described search result respectively
The input vector of machine learning model, the input vector include: individualized feature vector, global feature vector；
Subelement is exported, is configured to be based on the input vector using default machine learning model, be exported as a result, described
Output result is the position of the search result after sequence.
6. device according to claim 5, which is characterized in that described search result includes: the network money of vertical class website
Source, the operation include: browse operation, clicking operation, purchase operation, comment operation, and the statistical nature information includes: browsing
Amount, click volume, purchase volume, comment amount.
7. device according to claim 5, which is characterized in that described device further include:
Training unit is configured to input receiving comprising active user when default machine learning model is gbrank model
Search type searching request before, obtain the corresponding historical operation information of multiple historical search results and historical search result
Statistical nature information；Each described historical operation information and the statistical nature information are combined, multiple training are obtained
Sample；Multiple training samples are combined two-by-two in a manner of pairwise, obtain multiple training samples pair；Described in generating respectively
Training sample to each of the corresponding personalized training feature vector of training sample, the personalization training feature vector packet
It includes: indicating the component of the historical operation information, indicates the component of the statistical nature information；The training sample is generated respectively
To each of the corresponding whole training feature vector of training sample, the entirety training feature vector includes at least one point
Amount, wherein the representation in components predetermined order model obtains after being ranked up to the historical search result and historical search
As a result the associated parameter in position, the corresponding predetermined order model of each component；The training sample pair is generated respectively
Each of the corresponding gbrank model of training sample training input vector, the trained input vector includes: personalization
Training feature vector, whole training feature vector；Based on the trained input vector, the gbrank model is trained.
8. device according to claim 7, which is characterized in that the predetermined order model includes: smart order models, thick row
Sequence model, the parameter that smart order models obtain after being ranked up to search result or historical search result are that instruction is searched after sequence
The score of the position of hitch fruit or historical search result, after thick order models are ranked up search result or historical search result
Obtained parameter is the gear for indicating the location of search result or historical search result section.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610808298.8A CN106484766B (en) | 2016-09-07 | 2016-09-07 | Searching method and device based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610808298.8A CN106484766B (en) | 2016-09-07 | 2016-09-07 | Searching method and device based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106484766A CN106484766A (en) | 2017-03-08 |
| CN106484766B true CN106484766B (en) | 2019-10-22 |
Family
ID=58273657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610808298.8A Active CN106484766B (en) | 2016-09-07 | 2016-09-07 | Searching method and device based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106484766B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107291805B (en) * | 2017-05-15 | 2021-06-29 | 百度在线网络技术（北京）有限公司 | Method and device for controlling information pushing |
| CN110651266B (en) * | 2017-05-27 | 2023-05-23 | 北京嘀嘀无限科技发展有限公司 | System and method for providing information for on-demand services |
| CN107862004A (en) * | 2017-10-24 | 2018-03-30 | 科大讯飞股份有限公司 | Intelligent sorting method and device, storage medium, electronic equipment |
| CN109582865A (en) * | 2018-11-19 | 2019-04-05 | 北京奇虎科技有限公司 | A kind of method and device of pushing application program |
| CN111782982A (en) * | 2019-05-20 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Method and device for sorting search results and computer-readable storage medium |
| CN111143695A (en) * | 2019-12-31 | 2020-05-12 | 腾讯科技（深圳）有限公司 | Searching method, searching device, server and storage medium |
| CN111783452B (en) * | 2020-06-30 | 2024-04-02 | 北京百度网讯科技有限公司 | Model training method, information processing method, device, equipment and storage medium |
| CN113343132B (en) * | 2021-06-30 | 2022-08-26 | 北京三快在线科技有限公司 | Model training method, information display method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004772A (en) * | 2010-11-15 | 2011-04-06 | 百度在线网络技术（北京）有限公司 | Method and equipment for sequencing search results according to terms |
| CN102591969A (en) * | 2011-12-31 | 2012-07-18 | 北京百度网讯科技有限公司 | Method for providing search results based on historical behaviors of user and sever therefor |
| CN102999588A (en) * | 2012-11-15 | 2013-03-27 | 广州华多网络科技有限公司 | Method and system for recommending multimedia applications |
| CN104951468A (en) * | 2014-03-28 | 2015-09-30 | 阿里巴巴集团控股有限公司 | Data searching and processing method and system |
| CN105630836A (en) * | 2014-11-05 | 2016-06-01 | 阿里巴巴集团控股有限公司 | Searching result sorting method and apparatus |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104077306B (en) * | 2013-03-28 | 2018-05-11 | 阿里巴巴集团控股有限公司 | The result ordering method and system of a kind of search engine |
-
2016
- 2016-09-07 CN CN201610808298.8A patent/CN106484766B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004772A (en) * | 2010-11-15 | 2011-04-06 | 百度在线网络技术（北京）有限公司 | Method and equipment for sequencing search results according to terms |
| CN102591969A (en) * | 2011-12-31 | 2012-07-18 | 北京百度网讯科技有限公司 | Method for providing search results based on historical behaviors of user and sever therefor |
| CN102999588A (en) * | 2012-11-15 | 2013-03-27 | 广州华多网络科技有限公司 | Method and system for recommending multimedia applications |
| CN104951468A (en) * | 2014-03-28 | 2015-09-30 | 阿里巴巴集团控股有限公司 | Data searching and processing method and system |
| CN105630836A (en) * | 2014-11-05 | 2016-06-01 | 阿里巴巴集团控股有限公司 | Searching result sorting method and apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106484766A (en) | 2017-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106484766B (en) | Searching method and device based on artificial intelligence | |
| CN107766940A (en) | Method and apparatus for generation model | |
| CN109460513A (en) | Method and apparatus for generating clicking rate prediction model | |
| CN107105031A (en) | Information-pushing method and device | |
| CN109460514A (en) | Method and apparatus for pushed information | |
| CN110008397B (en) | Recommendation model training method and device | |
| CN107679217A (en) | Association method for extracting content and device based on data mining | |
| CN107832468A (en) | Demand recognition methods and device | |
| CN109976997A (en) | Test method and device | |
| CN108121800A (en) | Information generating method and device based on artificial intelligence | |
| CN109492160A (en) | Method and apparatus for pushed information | |
| CN110457476A (en) | Method and apparatus for generating disaggregated model | |
| CN107145485A (en) | Method and apparatus for compressing topic model | |
| CN109313587A (en) | Method for enabling data analysis service at service layer | |
| CN108140055A (en) | Trigger application message | |
| CN110473042B (en) | Method and device for acquiring information | |
| CN108121814A (en) | Search results ranking model generating method and device | |
| CN107315833A (en) | Method and apparatus of the retrieval with downloading based on application program | |
| CN108268450A (en) | For generating the method and apparatus of information | |
| CN109075987A (en) | Optimize digital assembly analysis system | |
| CN110263255A (en) | Acquisition methods, system, server and the storage medium of customer attribute information | |
| CN104331429B (en) | The method and device of multiple features dimension quantization is carried out to network object | |
| CN109978594A (en) | Order processing method, apparatus and medium | |
| CN108509442A (en) | Searching method and device, server and computer readable storage medium | |
| CN116204709A (en) | Data processing method and related device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |