CN108241631A - For the method and apparatus of pushed information - Google Patents
For the method and apparatus of pushed information Download PDFInfo
- Publication number
- CN108241631A CN108241631A CN201611206433.8A CN201611206433A CN108241631A CN 108241631 A CN108241631 A CN 108241631A CN 201611206433 A CN201611206433 A CN 201611206433A CN 108241631 A CN108241631 A CN 108241631A
- Authority
- CN
- China
- Prior art keywords
- sentence
- pushed
- identification information
- user
- trained
- 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.)
- Granted
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
- G06F16/951—Indexing; Web crawling techniques
Abstract
This application discloses the method and apparatus for pushed information.One specific embodiment of the method includes：The use of depth series model trained is in advance the sentence formation sequence feature respectively to be pushed in the sentence set to be pushed being previously obtained；Input feature vector using the contextual information between the respectively sequence signature of sentence to be pushed and word that respectively sentence to be pushed is included as conditional random field models trained in advance carries out sequence labelling, so as to the identification information in each sentence to be pushed；The identification information of obtained sentence respectively to be pushed with the identification information in the identification information set for the user being previously obtained is matched, and sentence to be pushed is chosen from above-mentioned sentence set to be pushed according to matching result and is pushed to terminal.The embodiment can accurately identify the identification information in sentence to be pushed, the legal risk that user is made effectively to avoid due to promoting the identification information included in sentence and bring.
Description
Technical field
This application involves field of computer technology, and in particular to Internet technical field, more particularly, to pushed information
Method and apparatus.
Background technology
Search engine marketing (Search Engine Marketing, SEM) is exactly the side that search engine is used according to user
Marketing message is passed to target user by formula as far as possible using the chance of user search information.With the quick hair of Internet technology
Exhibition, more enterprises begin through search engine marketing to do business promotion, and in extension process, enterprise is sought by search engine
The keyword or sentence of pin system setting will directly affect promotion effect.
In order to which enterprise is made more accurately and comprehensively to describe promoted business, promotion effect, search engine marketing system are promoted
System can be enterprise's recommended keywords or sentence for selection by the user and buy.However, if enterprise is had purchased comprising other enterprises
The keyword or sentence of identification information (such as brand name, company name etc.), often bring legal risk, while also can to enterprise
The search experience of the network user is reduced, therefore, when to enterprise's recommended keywords or sentence, should avoid marking containing other enterprises
The keyword or sentence for knowing information are pushed to enterprise.
Invention content
The purpose of the application is to propose a kind of method and apparatus for pushed information, to solve background above technology department
Divide the technical issues of mentioning.
In a first aspect, this application provides a kind of method for pushed information, including：Use depth sequence trained in advance
Row model is the sentence formation sequence feature respectively to be pushed in the sentence set to be pushed being previously obtained；Use respectively sentence to be pushed
Sequence signature and word that respectively sentence to be pushed is included between contextual information as condition random trained in advance
The input feature vector of field model carries out sequence labelling, so as to the identification information in each sentence to be pushed；It respectively waits to push away by what is obtained
The identification information of sentence is sent to be matched, and according to matching with the identification information in the identification information set for the user being previously obtained
As a result it chooses sentence to be pushed from above-mentioned sentence set to be pushed and is pushed to terminal.
In some embodiments, the identification information set of above-mentioned user obtains in the following manner：Use above-mentioned depth sequence
Row model is each sentence formation sequence feature in preset at least one sentence of above-mentioned user；Use the sequence of each sentence
Contextual information between the word that feature and each sentence are included carries out sequence as the input feature vector of above-mentioned condition random field
Row mark so as to obtain the identification information in preset at least one sentence of above-mentioned user, and is believed using obtained mark
Breath composition identification information set.
In some embodiments, the above-mentioned identification information by obtained sentence respectively to be pushed and the above-mentioned user being previously obtained
Identification information set in identification information matched, and chosen and treated from above-mentioned sentence set to be pushed according to matching result
Push sentence is pushed to terminal, including：For each sentence to be pushed in above-mentioned sentence set to be pushed, following walk is performed
Suddenly：The identification information of the obtained sentence to be pushed and each identification information in the identification information set of above-mentioned user are compared
Compared with；If the identification information of the sentence to be pushed is differed with the identification information in the identification information set of above-mentioned user, from
The sentence to be pushed is deleted in above-mentioned sentence set to be pushed；Remaining sentence to be pushed in above-mentioned sentence set to be pushed is pushed away
Give terminal.
In some embodiments, above-mentioned sentence set to be pushed obtains in the following manner：End is passed through according to above-mentioned user
Preset at least one sentence is held to obtain sentence set to be pushed.
In some embodiments, it is above-mentioned to obtain waiting to push away by preset at least one sentence of terminal according to above-mentioned user
Sentence set is sent, including：It is retrieved in preset sentence set using above-mentioned at least one sentence set by user,
And form sentence set to be pushed using the sentence that retrieval obtains.
In some embodiments, training obtains above-mentioned depth series model in the following manner：It chooses and uses from information bank
In the sample sentence of model training；Using the sample sentence training generation of selection for the depth sequence mould of formation sequence feature
Type.
In some embodiments, training obtains above-mentioned condition random field models in the following manner：Use above-mentioned depth sequence
Row model is each trained sentence formation sequence feature in training sentence set, wherein, each instruction in above-mentioned trained sentence set
Practice sentence to be marked in advance；Between the word included using the sequence signature and each trained sentence of each trained sentence
Contextual information training generation is used for the conditional random field models of sequence labelling.
Second aspect, this application provides a kind of device for pushed information, including：Generation unit, it is pre- for using
First trained depth series model is the sentence formation sequence feature respectively to be pushed in the sentence set to be pushed being previously obtained；Mark
Note unit, for use the respectively sequence signature of sentence to be pushed and word that respectively sentence to be pushed is included between context
Information carries out sequence labelling as the input feature vector of conditional random field models trained in advance, so as in each sentence to be pushed
Identification information；Push unit, for by the identification information of obtained sentence respectively to be pushed and the mark of user that is previously obtained
Identification information in information aggregate is matched, and chooses language to be pushed from above-mentioned sentence set to be pushed according to matching result
Sentence is pushed to terminal.
In some embodiments, above device further includes identification information set acquiring unit, and above-mentioned identification information set obtains
Unit is taken to be used for：It is generated using above-mentioned depth series model for each sentence in preset at least one sentence of above-mentioned user
Sequence signature；Contextual information between the word included using the sequence signature of each sentence and each sentence is as above-mentioned item
The input feature vector of part random field carries out sequence labelling, so as to obtain the mark in preset at least one sentence of above-mentioned user
Information, and use obtained identification information composition identification information set.
In some embodiments, above-mentioned push unit is further used for：For each in above-mentioned sentence set to be pushed
Sentence to be pushed performs following steps：By the identification information of the obtained sentence to be pushed and the identification information collection of above-mentioned user
Each identification information in conjunction is compared；If in the identification information set of the identification information of the sentence to be pushed and above-mentioned user
Identification information differ, then delete the sentence to be pushed from above-mentioned sentence set to be pushed；By above-mentioned sentence collection to be pushed
Remaining sentence to be pushed is pushed to terminal in conjunction.
In some embodiments, above device further includes sentence set acquiring unit to be pushed, above-mentioned sentence collection to be pushed
Acquiring unit is closed to be used for：Sentence set to be pushed is obtained by preset at least one sentence of terminal according to above-mentioned user.
In some embodiments, above-mentioned sentence set acquiring unit to be pushed is further used for：It is set using above-mentioned user
At least one sentence retrieved in preset sentence set, and language to be pushed is formed using the obtained sentence of retrieval
Sentence set.
In some embodiments, training obtains above-mentioned depth series model in the following manner：It chooses and uses from information bank
In the sample sentence of model training；Using the sample sentence training generation of selection for the depth sequence mould of formation sequence feature
Type.
In some embodiments, training obtains above-mentioned condition random field models in the following manner：Use above-mentioned depth sequence
Row model is each trained sentence formation sequence feature in training sentence set, wherein, each instruction in above-mentioned trained sentence set
Practice sentence to be marked in advance；Between the word included using the sequence signature and each trained sentence of each trained sentence
Contextual information training generation is used for the conditional random field models of sequence labelling.
The method and apparatus for pushed information that the application provides are sentence set to be pushed using depth series model
In sentence formation sequence feature respectively to be pushed, the then sentence institute to be pushed using the respectively sequence signature of sentence to be pushed and respectively
Comprising word between contextual information carry out sequence labelling as the input feature vector of conditional random field models, it is each so as to obtain
Identification information in sentence to be pushed, finally by the identification information of obtained sentence respectively to be pushed and the identification information set of user
In identification information matched, and sentence to be pushed is chosen from sentence set to be pushed according to matching result and is pushed to end
End so as to accurately identify the identification information in sentence to be pushed, makes user effectively avoid and is wrapped due to promoting in sentence
The identification information contained and the legal risk brought.
Description of the drawings
By reading the detailed description made to non-limiting example made with reference to the following drawings, the application's is other
Feature, objects and advantages will become more apparent upon：
Fig. 1 is that this application can be applied to exemplary system architecture figures therein；
Fig. 2 is the flow chart according to one embodiment of the method for pushed information of the application；
Fig. 3 is the schematic diagram according to an application scenarios of the method for pushed information of the application；
Fig. 4 is the structure diagram according to one embodiment of the device for pushed information of the application；
Fig. 5 is adapted for the structural representation for realizing the terminal device of the embodiment of the present application or the computer system of server
Figure.
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, illustrated only in attached drawing and invent relevant part with related.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the 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 implementation of the method for pushed information that can apply the application or the device for pushed information
The exemplary system architecture 100 of example.
As shown in Figure 1, system architecture 100 can include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 provide communication link medium.Network 104 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be interacted with using terminal equipment 101,102,103 by network 104 with server 105, to receive or send out
Send message etc..Various telecommunication customer end applications can be installed, such as web browser should on terminal device 101,102,103
With, shopping class apply, searching class application etc..
Terminal device 101,102,103 can be there is display screen and keyword or sentence selection and purchase various electricity
Sub- equipment, including but not limited to smart mobile 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) player, knee
Mo(u)ld top half pocket computer and desktop computer etc..
Server 105 can be to provide the server of various services, and for example, terminal device 101,102,103, which pushes, waits to push away
Send the background server of sentence.Background server, which can choose sentence to be pushed from sentence set to be pushed and be pushed to terminal, to be set
Standby 101,102,103.
It should be noted that the method for pushed information that the embodiment of the present application is provided generally is held by server 105
Row, correspondingly, is generally positioned at for the device of pushed information in server 105.
It should be understood that the number of the terminal device, network and server in Fig. 1 is only schematical.According to realization need
Will, can have any number of terminal device, network and server.
With continued reference to Fig. 2, the flow of one embodiment of the method for pushed information according to the application is shown
200.The method for pushed information, includes the following steps：
Step 201, the use of depth series model trained is in advance respectively to be treated in the sentence set to be pushed being previously obtained
Push sentence formation sequence feature.
In the present embodiment, electronic equipment (such as the service shown in FIG. 1 for the operation of the method for pushed information thereon
Device 105) can to use depth series model be sentence formation sequence feature respectively to be pushed in sentence set to be pushed, wherein,
Above-mentioned depth series model can be various engineerings that advance training obtains, for generating the sequence signature of sentence to be pushed
Practise model, such as LSTM trained in advance (Long Short-Term Memory, shot and long term memory) model, RNN
(Recurrent neural Network, Recognition with Recurrent Neural Network) model etc..For each sentence to be pushed, above-mentioned electronics
Equipment can be special so as to generate the corresponding sequence of the sentence to be pushed using the sentence to be pushed as the input of depth series model
Sign, when pushing sentence and being made of multiple words or word, also by multiple words in the corresponding sequence signature of the sentence to be pushed
Or the sequence composition of word, which can be the form of digital vectors.Herein, above-mentioned electronic equipment can be beforehand through
Various modes obtain sentence set to be pushed, and above-mentioned sentence set to be pushed includes at least one for for terminal device push
Sentence to be pushed.For example, above-mentioned electronic equipment can be the enterprise customer of business promotion according to search engine marketing is used
The business to be promoted obtains sentence set to be pushed, for example, using other enterprises that same business is managed with the enterprise customer
The sentence that industry user setting is crossed forms sentence set to be pushed.
In some optional realization methods of the present embodiment, above-mentioned depth series model can train in the following manner
It obtains：Above-mentioned electronic equipment is other for training the electronic equipment of depth series model, it is possible, firstly, to be selected from information bank
The sample sentence in model training is taken, can include doing business promotion by using search engine marketing in above- mentioned information library
Enterprise customer material (such as keyword of enterprise customer's setting, sentence, intention, promote region etc.)；Then, using choosing
The sample sentence training generation taken is used for the depth series model of formation sequence feature, for example, for training LSTM models, it is right
In each sample sentence, first by sequence of the sample sentence segmentation for word or word, then, which is included
Word or the sequence inputting of word be trained to LSTM models, so as to obtain the depth sequence mould for formation sequence feature
Type.
In some optional realization methods of the present embodiment, above-mentioned sentence set to be pushed can obtain in the following manner
：Above-mentioned electronic equipment obtains sentence set to be pushed according to above-mentioned user by preset at least one sentence of terminal.
For example, above-mentioned electronic equipment can use it is existing open up word tool to user by preset at least one sentence of terminal into
Row is expanded, so as to obtain sentence set to be pushed.Herein, user can refer to do business promotion using search engine marketing
Enterprise customer, above-mentioned at least one sentence can refer to enterprise customer's setting, sentence for business promotion.
It is above-mentioned that preset at least one sentence of terminal is passed through according to above-mentioned user in some optional realization methods
Sentence set to be pushed is obtained, can be specifically included：Using above-mentioned at least one sentence set by user in preset language
It is retrieved in sentence set, and sentence set to be pushed is formed using the sentence that retrieval obtains.Due to enterprise customer's the scope of one's knowledge and
The limitation of experience, at least one sentence of setting tend not to accurately and comprehensively describe promoted business, therefore can be with base
In at least one sentence of enterprise customer's setting, from preset sentence set retrieval obtain more multiple statement and form to wait to push
Sentence set.The sentence set can be the industry statement library that (such as the mode manually set) obtains by various modes, row
The sentence of largely energy accurate description the sector business can be included in industry statement library.
Step 202, using between the respectively sequence signature of sentence to be pushed and word that respectively sentence to be pushed is included
Contextual information carries out sequence labelling as the input feature vector of conditional random field models trained in advance, so as to respectively be waited to push
Identification information in sentence.
In the present embodiment, above-mentioned electronic equipment can use the sequence signature of sentence respectively to be pushed that step 201 generates
And respectively the contextual information between word that sentence to be pushed is included as condition random field trained in advance
The input feature vector of (conditional random field, CRF) model carries out sequence labelling, that is, use condition random field mould
Type is treated the corresponding sequence signature of push sentence and is labeled, and marks out whether each word that sentence to be pushed is included is mark
Information, so as to the identification information in each sentence to be pushed.Contextual information between the word that sentence to be pushed is included
It can refer to the contextual information for describing between each word included in sentence to be pushed, for example, sentence to be pushed
For " ABC housekeeping ", the word that the contextual information of the sentence can be used for behind words of description " ABC " is " housekeeping ".Herein,
Identification information can refer to the information such as brand name, company name.It should be noted that mark may not included in some sentences to be pushed
Know information.At this time, it is only necessary to identify the identification information in the sentence to be pushed comprising identification information.
In some optional realization methods of the present embodiment, above-mentioned condition random field models can instruct in the following manner
It gets：Above-mentioned electronic equipment or other be used for training condition random field models electronic equipment, it is possible, firstly, to using above-mentioned
Depth series model is each trained sentence formation sequence feature in training sentence set, wherein, in above-mentioned trained sentence set
Each trained sentence marked in advance, for example, can manually be marked to each trained sentence, so as to mark out trained language
Identification information in sentence, such as training sentence are " elder sister's household services ", then can brand name " elder sister " be labeled as brand, word
" housekeeping ", " service " are labeled as generic word；It is then possible to the sequence signature and each trained sentence using each trained sentence are wrapped
Contextual information training generation between the word contained is used for the conditional random field models of sequence labelling.
Step 203, by the identification information of obtained sentence respectively to be pushed and the identification information set of user being previously obtained
In identification information matched, and sentence to be pushed is chosen from sentence set to be pushed according to matching result and is pushed to end
End.
In the present embodiment, above-mentioned electronic equipment can be previously obtained the identification information set of user, and step 202 is obtained
To the identification information of sentence respectively to be pushed matched with the identification information in the identification information set of user, it is last according to
Sentence to be pushed is chosen from sentence set to be pushed be pushed to terminal with result.Herein, user can refer to using search
Engine is marketed come the enterprise customer for doing business promotion, and enterprise customer can be used to retouch by the way that search engine marketing account setup is multiple
The sentence for treating promotion business is stated, the identification information that the sentence that enterprise customer is set includes can form the mark of the enterprise customer
Information aggregate.Above-mentioned terminal can refer to terminal used in enterprise customer.
In some optional realization methods of the present embodiment, the identification information set of above-mentioned user can be by with lower section
Formula obtains：First, it is above-mentioned user preset at least one that above-mentioned electronic equipment, which can use above-mentioned depth series model,
Each sentence formation sequence feature in sentence, wherein, above-mentioned preset at least one sentence of user can refer to that enterprise uses
Family by search engine marketing account setup, for describe treat at least one sentence of promotion business, above-mentioned at least one sentence
In can include brand name, company name, enterprise name etc. identification informations, for example, the enterprise A for selling fresh flower gather around there are three brand " A1 ",
" A2 " and " A3 " can pass through search engine marketing account setup " express delivery of A fresh flowers ", " express delivery of A1 fresh flowers ", " A2 fresh flowers ", " A3
Fresh flower is how " sentences such as " fresh flower express delivery which good "；Then, above-mentioned electronic equipment can use each sentence sequence signature with
And the contextual information between the word that is included of each sentence carries out sequence labelling as the input feature vector of above-mentioned condition random field,
So as to obtain the identification information in preset at least one sentence of above-mentioned user, and use obtained identification information composition mark
Know information aggregate.
In some optional realization methods of the present embodiment, above-mentioned steps 203 can specifically include：Firstly, for upper
Each sentence to be pushed in sentence set to be pushed is stated, above-mentioned electronic equipment can perform following steps：This obtained is treated
The identification information of push sentence is compared with each identification information in the identification information set of above-mentioned user；If this waits to push
The identification information of sentence is differed with the identification information in the identification information set of above-mentioned user, then from above-mentioned sentence collection to be pushed
The sentence to be pushed is deleted in conjunction；Then, above-mentioned electronic equipment can be waited to push by remaining in above-mentioned sentence set to be pushed
Sentence is pushed to terminal.
With continued reference to Fig. 3, Fig. 3 is a signal according to the application scenarios of the method for pushed information of the present embodiment
Figure.In the application scenarios of Fig. 3, the enterprise customer A for selling fresh flower is gathered around there are three brand " A1 ", " A2 " and " A3 ", and enterprise customer A leads to
Cross search engine marketing systems buying for the sentence " express delivery of A fresh flowers " of business promotion, " express delivery of A1 fresh flowers ", " A2 fresh flowers ",
" how is A3 fresh flowers " " fresh flower express delivery which good ", first, server is using depth series model in sentence set to be pushed
Sentence formation sequence feature respectively to be pushed, wherein, sentence set to be pushed be bought according to enterprise customer A, for business
What the sentence of popularization obtained；Later, server is using the respectively sequence signature of sentence to be pushed and respectively sentence to be pushed is included
Word between contextual information as conditional random field models input feature vector carry out sequence labelling, so as to respectively be waited to push away
Send the identification information in sentence；Finally, by the identification information of obtained sentence respectively to be pushed and the identification information collection of enterprise customer A
Identification information " A ", " A1 ", " A2 " and " A3 " in conjunction is matched, if the identification information of sentence to be pushed and enterprise customer A
Identification information set in identification information differ, then delete the sentence to be pushed, and will be remaining in sentence set to be pushed
Sentence to be pushed " how is the express delivery of A fresh flowers ", " fresh flower express delivery ", " flowers express delivery ", " online booking is sent to for 1-2 hours " push away
Terminal used in enterprise customer A is given, it, will be as shown in figure 3, enterprise customer passes through a little so that enterprise customer selects and buys
It hits button 301 and can buy and selected sentence.
The method that above-described embodiment of the application provides can be with by using depth series model and conditional random field models
It accurately identifies the identification information of sentence push, waits to push away with enterprise customer mismatch identification information so as to eliminate to include
Sentence is sent, enterprise customer is made effectively to avoid due to promoting the law works wind for not being consistent identification information and bringing included in sentence
Danger, while improve the search experience of the network user.
With further reference to Fig. 4, as the realization to method shown in above-mentioned each figure, this application provides one kind for pushing letter
One embodiment of the device of breath, the device embodiment is corresponding with embodiment of the method shown in Fig. 2, which can specifically answer
For in various electronic equipments.
As shown in figure 4, include described in the present embodiment for the device 400 of pushed information：Generation unit 401, mark are single
Member 402 and push unit 403.Wherein, generation unit 401 is used to be previously obtained using depth series model trained in advance
Sentence formation sequence feature respectively to be pushed in sentence set to be pushed；Unit 402 is marked to be used for using respectively sentence to be pushed
Contextual information between sequence signature and the word that respectively sentence to be pushed is included is as condition random field trained in advance
The input feature vector of model carries out sequence labelling, so as to the identification information in each sentence to be pushed；Push unit 403 is used for will
The identification information of obtained sentence respectively to be pushed and the identification information progress in the identification information set of user being previously obtained
Match, and sentence to be pushed is chosen from above-mentioned sentence set to be pushed according to matching result and is pushed to terminal.
In the present embodiment, the specific processing of generation unit 401, mark unit 402 and push unit 403 can be with reference chart
The detailed description of 2 corresponding embodiment steps 201, step 202 and step 203, details are not described herein.
In some optional realization methods of the present embodiment, above device further includes identification information set acquiring unit
(not shown), above-mentioned identification information set acquiring unit are used for：It is preset using above-mentioned depth series model for above-mentioned user
At least one sentence in each sentence formation sequence feature；The word included using the sequence signature and each sentence of each sentence
Contextual information between language carries out sequence labelling as the input feature vector of above-mentioned condition random field, pre- so as to obtain above-mentioned user
Identification information at least one sentence first set, and use obtained identification information composition identification information set.The realization
Mode can refer to the detailed description of corresponding realization method in above-mentioned Fig. 2 corresponding embodiments, and details are not described herein.
In some optional realization methods of the present embodiment, above-mentioned push unit 403 is further used for：It is treated for above-mentioned
Each sentence to be pushed in sentence set is pushed, performs following steps：By the identification information of the obtained sentence to be pushed with
Each identification information in the identification information set of above-mentioned user is compared；If the identification information of the sentence to be pushed with it is above-mentioned
Identification information in the identification information set of user differs, then deletes the language to be pushed from above-mentioned sentence set to be pushed
Sentence；Remaining sentence to be pushed in above-mentioned sentence set to be pushed is pushed to terminal.The realization method can refer to above-mentioned Fig. 2 pairs
The detailed description of corresponding realization method in embodiment is answered, details are not described herein.
In some optional realization methods of the present embodiment, above device further includes sentence set acquiring unit to be pushed
(not shown), above-mentioned sentence set acquiring unit to be pushed are used for：Terminal preset at least one is passed through according to above-mentioned user
Sentence obtains sentence set to be pushed.The realization method can refer to the detailed of corresponding realization method in above-mentioned Fig. 2 corresponding embodiments
Thin description, details are not described herein.
In some optional realization methods of the present embodiment, above-mentioned sentence set acquiring unit to be pushed further is used
In：It is retrieved, and use is retrieved in preset sentence set using above-mentioned at least one sentence set by user
To sentence form sentence set to be pushed.The realization method can refer to corresponding realization method in above-mentioned Fig. 2 corresponding embodiments
Detailed description, details are not described herein.
In some optional realization methods of the present embodiment, above-mentioned depth series model is trained in the following manner
It arrives：The sample sentence for model training is chosen from information bank；It generates to generate sequence using the sample sentence training of selection
The depth series model of row feature.The realization method can refer to the detailed of corresponding realization method in above-mentioned Fig. 2 corresponding embodiments and retouch
It states, details are not described herein.
In some optional realization methods of the present embodiment, above-mentioned condition random field models are trained in the following manner
It arrives：The use of above-mentioned depth series model is each trained sentence formation sequence feature in training sentence set, wherein, above-mentioned training
Each trained sentence in sentence set is marked in advance；Sequence signature and each trained sentence institute using each trained sentence
Comprising word between contextual information training generation for sequence labelling conditional random field models.The realization method can join
The detailed description for stating corresponding realization method in Fig. 2 corresponding embodiments is admitted to, details are not described herein.
Below with reference to Fig. 5, it illustrates suitable for being used for realizing the calculating of the terminal device of the embodiment of the present application or server
The structure diagram of machine system 500.
As shown in figure 5, computer system 500 includes central processing unit (CPU) 501, it can be read-only according to being stored in
Program in memory (ROM) 502 or be loaded into program in random access storage device (RAM) 503 from storage section 508 and
Perform various appropriate actions and processing.In RAM 503, also it is stored with system 500 and operates required various programs and data.
CPU 501, ROM 502 and RAM 503 are connected with each other by bus 504.Input/output (I/O) interface 505 is also connected to always
Line 504.
I/O interfaces 505 are connected to lower component：Importation 506 including keyboard, mouse etc.；It is penetrated including such as cathode
The output par, c 507 of spool (CRT), liquid crystal display (LCD) etc. and loud speaker etc.；Storage section 508 including hard disk etc.；
And the communications portion 509 of the network interface card including LAN card, modem etc..Communications portion 509 via such as because
The network of spy's net performs communication process.Driver 510 is also according to needing to be connected to I/O interfaces 505.Detachable media 511, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on driver 510, as needed in order to be read from thereon
Computer program be mounted into storage section 508 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, it is machine readable including being tangibly embodied in
Computer program on medium, the computer program are included for the program code of the method shown in execution flow chart.At this
In the embodiment of sample, which can be downloaded and installed from network by communications portion 509 and/or from removable
Medium 511 is unloaded to be mounted.When the computer program is performed by central processing unit (CPU) 501, perform in the present processes
The above-mentioned function of limiting.
Flow chart and block diagram in attached drawing, it is illustrated that according to the system of the various embodiments of the application, method and computer journey
Architectural framework in the cards, function and the operation of sequence product.In this regard, each box in flow chart or block diagram can generation
The part of one module of table, program segment or code, a part for the module, program segment or code include one or more
The executable instruction of logic function as defined in being used to implement.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also be occurred with being different from the sequence marked in attached drawing.For example, two boxes succeedingly represented are practical
On can perform substantially in parallel, they can also be performed in the opposite order sometimes, this is depended on the functions involved.Also
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 systems of functions or operations as defined in execution is realized or can be referred to specialized hardware and computer
The combination of order is realized.
Being described in unit involved in the embodiment of the present application can be realized by way of software, can also be by hard
The mode of part is realized.Described unit can also be set in the processor, for example, can be described as：A kind of processor packet
Include generation unit, mark unit and push unit.Wherein, the title of these units is not formed under certain conditions to the unit
The restriction of itself, for example, generation unit is also described as " being previously obtained using depth series model trained in advance
The unit of sentence formation sequence feature respectively to be pushed in sentence set to be pushed ".
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 device described in above-described embodiment；Can also be
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 performed by an equipment so that the equipment：It uses
Trained depth series model is the sentence formation sequence feature respectively to be pushed in the sentence set to be pushed being previously obtained in advance；
Using the contextual information between the respectively sequence signature of sentence to be pushed and word that respectively sentence to be pushed is included as pre-
The input feature vector of first trained conditional random field models carries out sequence labelling, so as to which the mark in each sentence push is believed
Breath；By the identification information in the identification information of obtained sentence respectively to be pushed and the identification information set of user that is previously obtained into
Row matching, and sentence to be pushed is chosen from above-mentioned sentence set to be pushed according to matching result and is pushed to terminal.
The preferred embodiment and the explanation to institute's application technology principle that above description is only the application.People in the art
Member should be appreciated that invention scope involved in the application, however it is not limited to the technology that the specific combination of above-mentioned technical characteristic forms
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
The other technical solutions for arbitrarily combining and being formed.Such as features described above has similar work(with (but not limited to) disclosed herein
The technical solution that the technical characteristic of energy is replaced mutually and formed.
Claims (14)
- A kind of 1. method for pushed information, which is characterized in that the method includes：The use of depth series model trained is in advance sentence respectively the to be pushed generation in the sentence set to be pushed being previously obtained Sequence signature；Made using the contextual information between the respectively sequence signature of sentence to be pushed and word that respectively sentence to be pushed is included Input feature vector for conditional random field models trained in advance carries out sequence labelling, so as to the mark in each sentence to be pushed Information；By the identification information in the identification information of obtained sentence respectively to be pushed and the identification information set of user being previously obtained It is matched, and sentence to be pushed is chosen from the sentence set to be pushed according to matching result and is pushed to terminal.
- 2. according to the method described in claim 1, it is characterized in that, the identification information set of the user obtains in the following manner ：It is special for each sentence formation sequence in preset at least one sentence of the user using the depth series model Sign；Contextual information between the word included using the sequence signature of each sentence and each sentence as the condition with The input feature vector on airport carries out sequence labelling, so as to obtain the letter of the mark in preset at least one sentence of the user Breath, and use obtained identification information composition identification information set.
- 3. according to the method described in claim 1, it is characterized in that, the identification information by obtained sentence respectively to be pushed with Identification information in the identification information set for the user being previously obtained is matched, and waits to push away from described according to matching result It send and sentence to be pushed is chosen in sentence set is pushed to terminal, including：For each sentence to be pushed in the sentence set to be pushed, following steps are performed：The language to be pushed that will be obtained The identification information of sentence is compared with each identification information in the identification information set of the user；If the sentence to be pushed Identification information is differed with the identification information in the identification information set of the user, then is deleted from the sentence set to be pushed Except the sentence to be pushed；Remaining sentence to be pushed in the sentence set to be pushed is pushed to terminal.
- 4. according to the method described in claim 1, it is characterized in that, the sentence set to be pushed obtains in the following manner：Sentence set to be pushed is obtained by preset at least one sentence of terminal according to the user.
- It is 5. according to the method described in claim 4, it is characterized in that, described preset extremely by terminal according to the user A few sentence obtains sentence set to be pushed, including：It is retrieved, and use is retrieved in preset sentence set using at least one sentence set by user To sentence form sentence set to be pushed.
- 6. according to the method described in claim 1, it is characterized in that, the depth series model is trained in the following manner It arrives：The sample sentence for model training is chosen from information bank；Using the sample sentence training generation of selection for the depth series model of formation sequence feature.
- 7. according to the method described in claim 6, it is characterized in that, the conditional random field models are trained in the following manner It arrives：The use of the depth series model is each trained sentence formation sequence feature in training sentence set, wherein, the instruction Each trained sentence practiced in sentence set is marked in advance；Contextual information training life between the word included using the sequence signature and each trained sentence of each trained sentence Into the conditional random field models for sequence labelling.
- 8. a kind of device for pushed information, which is characterized in that described device includes：Generation unit is respectively to be treated in the sentence set to be pushed being previously obtained for using depth series model trained in advance Push sentence formation sequence feature；Mark unit, for use the respectively sequence signature of sentence to be pushed and word that respectively sentence to be pushed is included between Contextual information carries out sequence labelling as the input feature vector of conditional random field models trained in advance, so as to respectively be waited to push Identification information in sentence；Push unit, for by the identification information of obtained sentence respectively to be pushed and the identification information set of user that is previously obtained In identification information matched, and sentence to be pushed is chosen from the sentence set to be pushed according to matching result and is pushed to Terminal.
- 9. device according to claim 8, which is characterized in that described device further includes identification information set acquiring unit, The identification information set acquiring unit is used for：It is special for each sentence formation sequence in preset at least one sentence of the user using the depth series model Sign；Contextual information between the word included using the sequence signature of each sentence and each sentence as the condition with The input feature vector on airport carries out sequence labelling, so as to obtain the letter of the mark in preset at least one sentence of the user Breath, and use obtained identification information composition identification information set.
- 10. device according to claim 8, which is characterized in that the push unit is further used for：For each sentence to be pushed in the sentence set to be pushed, following steps are performed：The language to be pushed that will be obtained The identification information of sentence is compared with each identification information in the identification information set of the user；If the sentence to be pushed Identification information is differed with the identification information in the identification information set of the user, then is deleted from the sentence set to be pushed Except the sentence to be pushed；Remaining sentence to be pushed in the sentence set to be pushed is pushed to terminal.
- 11. device according to claim 8, which is characterized in that described device further includes sentence set to be pushed and obtains list Member, the sentence set acquiring unit to be pushed are used for：Sentence set to be pushed is obtained by preset at least one sentence of terminal according to the user.
- 12. according to the devices described in claim 11, which is characterized in that the sentence set acquiring unit to be pushed further is used In：It is retrieved, and use is retrieved in preset sentence set using at least one sentence set by user To sentence form sentence set to be pushed.
- 13. device according to claim 8, which is characterized in that the depth series model is trained in the following manner It arrives：The sample sentence for model training is chosen from information bank；Using the sample sentence training generation of selection for the depth series model of formation sequence feature.
- 14. device according to claim 8, which is characterized in that the conditional random field models are trained in the following manner It obtains：The use of the depth series model is each trained sentence formation sequence feature in training sentence set, wherein, the instruction Each trained sentence practiced in sentence set is marked in advance；Contextual information training life between the word included using the sequence signature and each trained sentence of each trained sentence Into the conditional random field models for sequence labelling.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611206433.8A CN108241631B (en) | 2016-12-23 | 2016-12-23 | Method and device for pushing information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611206433.8A CN108241631B (en) | 2016-12-23 | 2016-12-23 | Method and device for pushing information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108241631A true CN108241631A (en) | 2018-07-03 |
| CN108241631B CN108241631B (en) | 2022-09-30 |
Family
ID=62703479
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611206433.8A Active CN108241631B (en) | 2016-12-23 | 2016-12-23 | Method and device for pushing information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108241631B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544496A (en) * | 2012-07-12 | 2014-01-29 | 同济大学 | Method for recognizing robot scenes on basis of space and time information fusion |
| CN103778142A (en) * | 2012-10-23 | 2014-05-07 | 南开大学 | Conditional random fields (CRF) based acronym expansion explanation recognition method |
| CN103778407A (en) * | 2012-10-23 | 2014-05-07 | 南开大学 | Gesture recognition algorithm based on conditional random fields under transfer learning framework |
| CN103988158A (en) * | 2011-12-12 | 2014-08-13 | 国际商业机器公司 | System, device, and method for processing of three-dimensional time series data |
| CN104573054A (en) * | 2015-01-21 | 2015-04-29 | 杭州朗和科技有限公司 | Information pushing method and equipment |
| CN104598510A (en) * | 2014-10-16 | 2015-05-06 | 苏州大学 | Event trigger word recognition method and device |
| CN104679885A (en) * | 2015-03-17 | 2015-06-03 | 北京理工大学 | User search string organization name recognition method based on semantic feature model |
| US20150169549A1 (en) * | 2013-12-13 | 2015-06-18 | Google Inc. | Cross-lingual discriminative learning of sequence models with posterior regularization |
| CN105404632A (en) * | 2014-09-15 | 2016-03-16 | 深港产学研基地 | Deep neural network based biomedical text serialization labeling system and method |
| CN105701155A (en) * | 2015-12-30 | 2016-06-22 | 百度在线网络技术（北京）有限公司 | Information push method and the device |
| CN105868193A (en) * | 2015-01-19 | 2016-08-17 | 富士通株式会社 | Device and method used to detect product relevant information in electronic text |
| CN105976056A (en) * | 2016-05-03 | 2016-09-28 | 成都数联铭品科技有限公司 | Information extraction system based on bidirectional RNN |
-
2016
- 2016-12-23 CN CN201611206433.8A patent/CN108241631B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103988158A (en) * | 2011-12-12 | 2014-08-13 | 国际商业机器公司 | System, device, and method for processing of three-dimensional time series data |
| CN103544496A (en) * | 2012-07-12 | 2014-01-29 | 同济大学 | Method for recognizing robot scenes on basis of space and time information fusion |
| CN103778142A (en) * | 2012-10-23 | 2014-05-07 | 南开大学 | Conditional random fields (CRF) based acronym expansion explanation recognition method |
| CN103778407A (en) * | 2012-10-23 | 2014-05-07 | 南开大学 | Gesture recognition algorithm based on conditional random fields under transfer learning framework |
| US20150169549A1 (en) * | 2013-12-13 | 2015-06-18 | Google Inc. | Cross-lingual discriminative learning of sequence models with posterior regularization |
| CN105404632A (en) * | 2014-09-15 | 2016-03-16 | 深港产学研基地 | Deep neural network based biomedical text serialization labeling system and method |
| CN104598510A (en) * | 2014-10-16 | 2015-05-06 | 苏州大学 | Event trigger word recognition method and device |
| CN105868193A (en) * | 2015-01-19 | 2016-08-17 | 富士通株式会社 | Device and method used to detect product relevant information in electronic text |
| CN104573054A (en) * | 2015-01-21 | 2015-04-29 | 杭州朗和科技有限公司 | Information pushing method and equipment |
| CN104679885A (en) * | 2015-03-17 | 2015-06-03 | 北京理工大学 | User search string organization name recognition method based on semantic feature model |
| CN105701155A (en) * | 2015-12-30 | 2016-06-22 | 百度在线网络技术（北京）有限公司 | Information push method and the device |
| CN105976056A (en) * | 2016-05-03 | 2016-09-28 | 成都数联铭品科技有限公司 | Information extraction system based on bidirectional RNN |
Non-Patent Citations (3)
| Title |
|---|
| KAISHENG YAO 等: "Spoken language understanding using long short-term memory neural networks", 《 2014 IEEE SPOKEN LANGUAGE TECHNOLOGY WORKSHOP (SLT)》 * |
| 冯仓龙 等: "面向商品评价的情感要素抽取", 《沈阳航空航天大学学报》 * |
| 奚雪峰 等: "面向自然语言处理的深度学习研究", 《自动化学报》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108241631B (en) | 2022-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106845999A (en) | Risk subscribers recognition methods, device and server | |
| CN107172151A (en) | Method and apparatus for pushed information | |
| CN106874467A (en) | Method and apparatus for providing Search Results | |
| CN107644364A (en) | Object filter method and system | |
| CN106407425A (en) | A method and a device for information push based on artificial intelligence | |
| CN110020162B (en) | User identification method and device | |
| CN108268573A (en) | For the method and apparatus of pushed information | |
| CN108280200A (en) | Method and apparatus for pushed information | |
| CN109614540A (en) | Commodity retrieval, device, computer equipment and storage medium | |
| CN107943895A (en) | Information-pushing method and device | |
| CN110298716A (en) | Information-pushing method and device | |
| CN107360243A (en) | Information-pushing method and device | |
| CN110084658A (en) | The matched method and apparatus of article | |
| CN107742128A (en) | Method and apparatus for output information | |
| CN108595448A (en) | Information-pushing method and device | |
| US8972279B2 (en) | Matching audio advertisements to items on a shopping list in a mobile device | |
| CN108121814A (en) | Search results ranking model generating method and device | |
| CN107330091A (en) | Information processing method and device | |
| CN109460503A (en) | Answer input method, device, storage medium and electronic equipment | |
| CN105718571A (en) | Information pushing method and device | |
| CN108667875A (en) | information updating method and device | |
| CN110164429A (en) | Voice interactive method and device | |
| CN111666408A (en) | Method and device for screening and displaying important clauses | |
| CN108241631A (en) | For the method and apparatus of pushed information | |
| CN106355415B (en) | Processing method and device for commodity traceability identification |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |