CN110442332A - Natural language is converted into the method, apparatus and computer equipment of program language - Google Patents
Natural language is converted into the method, apparatus and computer equipment of program language Download PDFInfo
- Publication number
- CN110442332A CN110442332A CN201910600648.5A CN201910600648A CN110442332A CN 110442332 A CN110442332 A CN 110442332A CN 201910600648 A CN201910600648 A CN 201910600648A CN 110442332 A CN110442332 A CN 110442332A
- Authority
- CN
- China
- Prior art keywords
- entity
- natural language
- template
- slot
- program
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000003062 neural network model Methods 0.000 claims abstract description 20
- 239000000284 extract Substances 0.000 claims abstract description 15
- 230000014509 gene expression Effects 0.000 claims description 61
- 239000013598 vector Substances 0.000 claims description 46
- 238000004364 calculation method Methods 0.000 claims description 34
- 238000004590 computer program Methods 0.000 claims description 13
- 239000000470 constituent Substances 0.000 claims description 6
- 238000005429 filling process Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000012804 iterative process Methods 0.000 claims description 5
- 230000004069 differentiation Effects 0.000 claims 1
- 238000013461 design Methods 0.000 abstract description 8
- 238000004422 calculation algorithm Methods 0.000 description 14
- 238000000605 extraction Methods 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 239000000203 mixture Substances 0.000 description 4
- 244000287680 Garcinia dulcis Species 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3347—Query execution using vector based model
-
- 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
- G06F16/903—Querying
- G06F16/90335—Query processing
- G06F16/90344—Query processing by using string matching techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/30—Creation or generation of source code
- G06F8/33—Intelligent editors
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Stored Programmes (AREA)
- Machine Translation (AREA)
Abstract
This application discloses method, apparatus and computer equipment that natural language is converted into program language, wherein the method that natural language is converted into program language, comprising: obtains the natural language sentence of user's input；All entity types for including in the natural language sentence are distinguished, and extract the corresponding entity of all entity types；Natural language template is matched according to all entity types；The natural language template is converted into program language template by the neural network model of pre-training, wherein described program language template carries slot to be filled；All entities are respectively corresponded into the slot to be filled filled into described program template, generate the corresponding program language of the natural language sentence.Making to have reserved slot to be filled in to-fill procedure template by training design, for filling the entity in natural language sentence into process template, forming the corresponding program language of natural language sentence.
Description
Technical field
This application involves computer field is arrived, the method, apparatus that natural language is converted into program language is especially related to
And computer equipment.
Background technique
In the prior art, without the paginal translation tool of program language and natural language sentence, programmer is being compiled under normal conditions
Can be potentially encountered during journey can he has conceived with natural language description implementation, but be transcribed into code
Technical difficulty is encountered in the process；And existing Internet resources, such as the website of question and answer type, the programming that programmer is submitted
Program table is made in problem, when design new task or composite task, when needing the programming content in conjunction with history, can not pass through nature
Language statement realizes the effective lookup and calling to corresponding programming content.
Summary of the invention
The main purpose of the application is the method for providing natural language and being converted into program language, it is intended to which solving can not be by certainly
Right language statement realizes the technical issues of effective lookup and calling to corresponding programming content.
The application proposes a kind of method that natural language is converted into program language, comprising:
Obtain the natural language sentence of user's input；
All entity types for including in the natural language sentence are distinguished, and it is right respectively to extract all entity types
The entity answered；
Natural language template is matched according to all entity types；
The natural language template is converted into program language template by the neural network model of pre-training, wherein described
Program language template carries slot to be filled；
All entities are respectively corresponded into the slot to be filled filled into described program template, generate the nature
The corresponding program language of language statement.
It is preferably, described that all entities are respectively corresponded into the slot to be filled filled into described program template,
The step of generating the natural language sentence corresponding program language, comprising:
It is determining with the matched first instance of the first slot according to preset matching calculation formula, wherein the first instance includes
In all entities, first slot is contained in all slots to be filled；
If so, filling the first instance into first slot；
The filling of all slots to be filled in described program template is completed according to the filling process of first slot, it is raw
At described program language.
Preferably, the step of first instance matched with the first slot determining according to preset matching calculation formula, comprising:
Pass through calculation formulaCalculate separately all entities with
Matching fractional value between first slot, wherein d(i,j),(c,d)=cos ((hi,hj`),(hc,hd`)),hiIndicate the corresponding vector of all entities, hj` indicates that first slot is corresponding
Vector, (ec,sd) it is the second instance and the second slot for having marked matching status, hcIndicate the second instance ecIt is corresponding to
Amount, hd` indicates the second slot sdCorresponding vector；
When determining the matching fractional value maximum, designated entities corresponding with first slot；
The designated entities are determined as the first instance.
Preferably, the neural network model includes coding-decoded model, described to pass through the natural language template in advance
Trained neural network model is converted into the step of program language template, comprising:
By in the hidden layer of the decoding encoding model of natural language template input pre-training, it is fitted and is fitted
State natural language template simultaneously exports, wherein the hidden layer of the decoding encoding model includes the form of the first specified quantity, described
The quantity of entity type for including in first specified quantity and the natural language template is identical；
By the hidden layer of the decoded model of the fitting state natural language template input pre-training, is inputted and calculated by iteration
The list of ordering described program language template and output are converted to, wherein described program language template is protected according to preset format
Stay the slot to be filled.
Preferably, the hidden layer of the decoded model by the fitting state natural language template input pre-training, passes through
Iteration input calculates the step of being converted to list and the output of ordering described program language template, comprising:
Using the fitting state natural language template as the first input data, it is input to the first of the hidden layer of decoded model
Gradient layer determines the corresponding first instance type of the highest vector of weight coefficient and exports；
Using the first instance type and the calculating data of the first gradient layer as the second input data, it is input to solution
Second gradient layer of the hidden layer of code model determines the weight coefficient time high corresponding second instance type of vector and exports,
Described in first gradient layer and the second gradient layer be adjacent connection in the hidden layer of the decoded model iteration layer；
According to the iterative process for inputting data into second input data by described first, successively iteration is input to described
The hidden layer of decoded model, until all trunk words of the natural language sentence are determined, wherein each trunk word point
It is not determined according to the sequence of corresponding weight coefficient, and according to the sequence of weight coefficient, forms the described program language of sequence
Say the list of template.
Preferably, described to distinguish all entity types for including in the natural language sentence, and extract all realities
The step of body type corresponding entity, comprising:
The regular expression being pre-configured according to the constituent element of described program template is called, wherein the regular expression
Type includes at least two kinds；
Judged in the natural language sentence according to the first regular expression, if include and first regular expression
The given entity type of adaptation, wherein first regular expression is contained in all regular expressions；
If so, extracting the corresponding entity of the given entity type by first regular expression.
Preferably, the entity type includes mode class entity and quantity class entity, described according to all entity class
Type matches the step of natural language template, comprising:
The corresponding entity of the mode class entity and physical quantities and the corresponding reality of the quantity class entity are obtained respectively
Body and physical quantities；
According to the corresponding entity of the mode class entity and physical quantities and the corresponding reality of the quantity class entity
Body and physical quantities, according to preset order arrange the corresponding entity of the mode class entity with and the quantity class entity it is corresponding
Entity, form the natural language template.
Present invention also provides the devices that a kind of natural language is converted into program language, comprising:
Module is obtained, for obtaining the natural language sentence of user's input；
Discriminating module for distinguishing all entity types for including in the natural language sentence, and is extracted all described
The corresponding entity of entity type；
Matching module, for matching natural language template according to all entity types；
Conversion module, for the natural language template to be converted into program language by the neural network model of pre-training
Template, wherein described program language template carries slot to be filled；
Module is filled, it is described to be filled into described program template for all entities to be respectively corresponded filling
Slot generates the corresponding program language of the natural language sentence.
Present invention also provides a kind of computer equipment, including memory and processor, the memory is stored with calculating
The step of machine program, the processor realizes the above method when executing the computer program.
Present invention also provides a kind of computer readable storage mediums, are stored thereon with computer program, the computer
The step of above-mentioned method is realized when program is executed by processor.
The application is generated by extracting the entity type in natural language sentence, and according to the entity type of extraction
Then natural language template is input in trained neural network by natural language template, natural language template is passed through quasi-
It is total to calculate and iterative calculation, it is converted into process template, and by training design make to have reserved in to-fill procedure template to be filled
Slot forms the corresponding program language of natural language sentence for filling the entity in natural language sentence into process template.
By the way that natural language sentence is converted directly into program language, realization can allow programming personnel to express expection using natural language
Operation, facilitate programmer examine and use.K- nearest neighbor algorithm is improved to complete by local matching function, is solved existing
Euclidean distance etc. in k- nearest neighbor algorithm is not suitable for the problem of matching of slot and entity in the application, by will be apart from nearest
Entity and slot composition data pair, be spliced into a long vector, by calculate two long vectors between similarity, and then
Matching degree between slot and entity to be matched.The accurate alignment for realizing slot and entity through the above way, is improved from nature
Accurate correspondence of the language statement to program language.Different entity types is identified by matching different regular expressions,
And corresponding entity is extracted, method is easy, and calculation amount is small and identifies that the accuracy for extracting entity is high.By by natural language sentence
It is sequentially inputted in each regular expression, to judge in natural language sentence whether to include reality with current regular expression matching
Body type, and if it exists, then extract the entity.
Detailed description of the invention
The natural language of one embodiment of Fig. 1 the application is converted into the method flow schematic diagram of program language；
The natural language of one embodiment of Fig. 2 the application is converted into the apparatus structure schematic diagram of program language；
The computer equipment schematic diagram of internal structure of one embodiment of Fig. 3 the application.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not
For limiting the application.
Referring to Fig.1, the method that the natural language of one embodiment of the application is converted into program language, comprising:
S1: the natural language sentence of user's input is obtained.
S2: distinguishing all entity types for including in the natural language sentence, and extracts all entity types point
Not corresponding entity.
S3: natural language template is matched according to all entity types.
S4: the natural language template is converted into program language template by the neural network model of pre-training, wherein
Described program language template carries slot to be filled.
S5: all entities being respectively corresponded and fill the slot to be filled into described program template, described in generation
The corresponding program language of natural language sentence.
The present embodiment is given birth to by extracting the entity type in natural language sentence, and according to the entity type of extraction
At natural language template, then natural language template is input in trained neural network, natural language template is passed through
The Fitting Calculation and iterative calculation are converted into process template, and make to have reserved in to-fill procedure template wait fill out by training design
Slot being filled, for filling the entity in natural language sentence into process template, forming the corresponding program word of natural language sentence
Speech.Above-mentioned entity type is the corresponding type of word of trunk word for indicating natural language sentence, such as including filename,
Pathname, absolute path name, time etc., above-mentioned entity include filename, pathname, and absolute path name, time etc. are right respectively
The assignment answered.The process of said extracted entity type, for the semantic process for determining trunk word of identification, that is, the word extracted retains
The original semantic expression of former natural language sentence.Above-mentioned natural language template is the corresponding entity class of above-mentioned trunk word
Type, the sequence formed according to the arrangement order in former natural language sentence.Then by by above-mentioned sequence inputting to training
In good neural network, by the fitting and iterative calculation in coding and decoding, natural language template is made to be gradually converted into program
The sequence of template, and reserved slot to be filled.The present embodiment is realized by the way that natural language sentence is converted directly into program language
Programming personnel can be allowed to express expected operation using natural language, programmer is facilitated to examine and use.
Further, it is described by all entities respectively correspond filling it is described to be filled into described program template
Slot generates the step S5 of the corresponding program language of the natural language sentence, comprising:
S51: according to the determination of preset matching calculation formula and the matched first instance of the first slot, wherein the first instance
It is contained in all entities, first slot is contained in all slots to be filled.
S52: if so, filling the first instance into first slot.
S53: filling out for all slots to be filled in described program template is completed according to the filling process of first slot
It fills, generates described program language.
The present embodiment is by carrying out matching pairing for entity and slot, and realization accurately fills entity in slot, and solving will
Correctly input character string is aligned with the parameter value of output order.The preset matching calculation formula of the present embodiment is to pass through local
It improves k- nearest neighbor algorithm with function to complete, since the Euclidean distance etc. in existing k- nearest neighbor algorithm is not suitable for the application
In slot and entity matching, slot and entity in the application are without specific corresponding vector, and the present embodiment is by will be apart from nearest
Entity and slot composition data pair, be spliced into a long vector, by calculate two long vectors between similarity, and then
Matching degree between slot and entity to be matched.The accurate alignment for realizing slot and entity through the above way, is improved from nature
Accurate correspondence of the language statement to program language.
Further, the step S51 according to preset matching calculation formula determination and the matched first instance of the first slot,
Include:
S511: pass through calculation formulaCalculate separately all realities
Matching fractional value between body and first slot, wherein d(i,j),(c,d)=cos ((hi,h`j),(hc,h`d)),hiIndicate the corresponding vector of all entities, hj` indicates that first slot is corresponding
Vector, (ec,sd) it is the second instance and the second slot for having marked matching status, hcIndicate the second instance ecIt is corresponding to
Amount, hd` indicates the second slot sdCorresponding vector.
S512: when determining the matching fractional value maximum, designated entities corresponding with first slot.
S513: the designated entities are determined as the first instance.
In the present embodiment, k- nearest neighbor algorithm is improved by local matching function, according to k- nearest neighbor algorithm by calculating
Probability of occurrence in adjacent area is modified to judge the principle of similitude by selecting apart from entity to be matched and slot number according to right
The entity and slot number of nearest marked matched data are according to matching score that is right, and calculating above-mentioned two pairs of entities and slot number evidence pair
Value to overcome the problems, such as not being directly realized by accurate map vector between existing entity and slot, and is calculated by improved k- neighbour
Method reduces calculation amount and improves the matching accuracy of entity and slot.Citing ground, above-mentioned second instance ecWith the second slot sd, point
Not in the inner corresponding hidden layer variable h of encoding model (Encoder model) and decoded model (Decoder model)c,hd
` forms long vector (h by splicing variablec,hd`), reference entity and slot number are constituted according to right, and identifies matching status,
Matching identification is 1, is 0 than matching identification.By fixed slot to be filled, i.e. the first slot, all entities, group are matched respectively
At multiple entities for including the first slot and slot number according to right, same hiIndicate that all entities are corresponding in Encoder model respectively
Hidden layer variable, hj` indicates corresponding hidden layer variable in the first slot Decoder model, by being spliced to form long vector
(hi, hj`), the entity by screening apart from the first slot to be analyzed and slot number are according to right, the nearest mark matching status of distance value
Second instance and the second slot, and calculate with marked the second instance of matching status and the matching fractional value of the second slot, match
Fractional value is bigger, then the entity and slot number for showing the first slot to be analyzed according to right, with the second instance for having marked matching status and
Second slot is more similar.Such as marked the second instance of matching status and the second slot is labeled as matching, then with it is similar to point
The entity and slot number of first slot of analysis realize the accurate correspondence of entity and slot according to being also estimated as matching.
Further, the neural network model includes coding-decoded model, described to pass through the natural language template
The neural network model of pre-training is converted into the step S4 of program language template, comprising:
S41: it by the hidden layer of the decoding encoding model of natural language template input pre-training, is fitted to obtain
Fitting state natural language template simultaneously exports, wherein and the hidden layer of the decoding encoding model includes the form of the first specified quantity,
The quantity of entity type for including in first specified quantity and the natural language template is identical.
S42: the hidden layer of the decoded model of the fitting state natural language template input pre-training is inputted by iteration
List and the output for being converted to ordering described program language template are calculated, wherein described program language template is according to default lattice
Formula retains the slot to be filled.
The present embodiment is realized by the hidden layer of Encoder and is fitted to the dimensionality reduction of natural language template, and treatment process is reduced
In complexity, and improve the correct recognition rata to the extraction feature of natural language template.Encoder includes in the present embodiment
The corresponding 7 fitting forms of 4 layers of hidden layer are followed successively by H1 to H7, set 20 for the step-length of every layer of LSTM；In MSCOCO data set
On, hidden layer output is set as 1000, and on Flickr30K data set, the output of LSTM hidden layer is 512, hidden on Flickr8K
Layer output is set as 256, to improve dimensionality reduction effect.The hidden layer of the Decoder of the present embodiment is by receiving Encoder output
The last one hidden layer is fitted form as input, by successive ignition input and output, obtains ordering described program language
The list of template, and slot to be filled is remained according to preset mode.The hidden layer of above-mentioned Encoder and hiding for Decoder
Layer is connected, and trains optimum parameter by marked training sample, the program for exporting Encoder-Decoder model
Template infinite approach preset mode.
Further, the hidden layer of the decoded model by the fitting state natural language template input pre-training, leads to
Cross the step S42 that iteration input calculates the list for being converted to ordering described program language template and exports, comprising:
S421: using the fitting state natural language template as the first input data, it is input to the hidden layer of decoded model
First gradient layer, determine the corresponding first instance type of the highest vector of weight coefficient and export.
S422: defeated using the first instance type and the calculating data of the first gradient layer as the second input data
Enter the second gradient layer to the hidden layer of decoded model, determines time high corresponding second instance type of vector of weight coefficient and defeated
Out, wherein the first gradient layer and the second gradient layer are the iteration layer of adjacent connection in the hidden layer of the decoded model.
S423: according to the iterative process for inputting data into second input data by described first, successively iteration is inputted
To the hidden layer of the decoded model, until all trunk words of the natural language sentence are determined, wherein each trunk
Word is determined according to the sequence of corresponding weight coefficient respectively, and according to the sequence of weight coefficient, forms the described of sequence
The list of program language template.
The present embodiment passes through multiple gradient layers interconnected, is iterated input and output, such as including H1` to H5`, will
The last one hidden layer of Encoder output is fitted form, i.e., fitting state natural language template is as the first input data, input
To the H1` of the hidden layer of Decoder, the primitive most important first instance type Y1` of justice to natural language sentence is exported, then
By Y1` and the process data for extracting Y1` while the input for being used as H2` are calculated, exports the primitive justice time to natural language sentence
Then the process data that Y2` and calculating are extracted to Y2` is used as the input of H3` by important second instance type Y2` simultaneously,
Successively iteration carries out calculating extraction, until output includes the column of the described program language template of all entity types of preset quantity
Table.Citing ground, user's input " journal file before searching 15 days ", corresponding natural language template are as follows: find all [file
Name] files older than [Timespan], by Encoder fitting output after, by Decoder hidden layer repeatedly
It is calculated for input and output, the list of the described program language template of obtained sequence are as follows: Y1:find [path]-name
[regex]-mtime[+Timespan]；Y2:find[path]-type f-name[regex]-mtime[+Timespan]；
Y3:find[path]-type f-perm[permission]；Y4:find[path]-name[regex]-mtime[+
Timespan]xargs ls；Y5:....
Further, described to distinguish all entity types for including in the natural language sentence, and extract all described
The step S2 of the corresponding entity of entity type, comprising:
S21: the regular expression being pre-configured according to the constituent element of described program template is called, wherein the regular expressions
The type of formula includes at least two kinds.
S22: judged in the natural language sentence according to the first regular expression, if include and the first canonical table
The given entity type being adapted to up to formula, wherein first regular expression is contained in all regular expressions.
S23: if so, extracting the corresponding entity of the given entity type by first regular expression.
The present embodiment identifies different entity types by matching different regular expressions, and extracts corresponding real
Body, method is easy, and calculation amount is small and identifies that the accuracy for extracting entity is high.By by natural language sentence be sequentially inputted to it is each just
Then in expression formula, to judge in natural language sentence whether to include entity type with current regular expression matching, and if it exists,
Then extract the entity.Multiple regular expressions of the present embodiment arrangement can previously according to the important procedure of each entity type,
Successively descending is arranged, and is arranged so that the entity type of identification is formed according to important order.
Further, the entity type includes mode class entity and quantity class entity, described according to all entities
The step S3 of type matching natural language template, comprising:
S31: obtaining the corresponding entity of the mode class entity and physical quantities respectively and the quantity class entity is corresponding
Entity and physical quantities.
S32: corresponding according to the corresponding entity of the mode class entity and physical quantities and the quantity class entity
Entity and physical quantities, arrange the corresponding entity of the mode class entity and and the quantity class entity according to preset order
Corresponding entity forms the natural language template.
The entity type of the present embodiment includes multiple mode class entities and multiple quantity class entities, such as including 5 modes
Class entity, comprising: " File:file name；Directory:directory name；Path:absolute path；Date/
Time:date and time expression；Regex:other pattern arguments " and 3 quantity classes are real
Body, including " Number:number；Size:file size；Timespan:time duration ", to cover certainly as far as possible comprehensively
All trunk words of right language statement.By determining that different entities type and variant entity type correspond to natural language
The importance ranking of sentence forms natural language template, while determining and extracting the corresponding entity of variant entity type,
Assignment as slot to be filled.For example, user's input " journal file before searching 15 days, i.e. find all log files
Older than 15days ", trunk word include " file name:log；Timespan:15days ", corresponding natural language
Template are as follows: find all [file name] files older than [Timespan].
Referring to Fig. 2, the natural language of one embodiment of the application is converted into the device of program language, comprising:
Module 1 is obtained, for obtaining the natural language sentence of user's input.
Discriminating module 2 for distinguishing all entity types for including in the natural language sentence, and is extracted all described
The corresponding entity of entity type.
Matching module 3, for matching natural language template according to all entity types.
Conversion module 4, for the natural language template to be converted into program word by the neural network model of pre-training
Say template, wherein described program language template carries slot to be filled.
Module 5 is filled, it is described to be filled into described program template for all entities to be respectively corresponded filling
Slot generates the corresponding program language of the natural language sentence.
The present embodiment is given birth to by extracting the entity type in natural language sentence, and according to the entity type of extraction
At natural language template, then natural language template is input in trained neural network, natural language template is passed through
The Fitting Calculation and iterative calculation are converted into process template, and make to have reserved in to-fill procedure template wait fill out by training design
Slot being filled, for filling the entity in natural language sentence into process template, forming the corresponding program word of natural language sentence
Speech.Above-mentioned entity type is the corresponding type of word of trunk word for indicating natural language sentence, such as including filename,
Pathname, absolute path name, time etc., above-mentioned entity include filename, pathname, and absolute path name, time etc. are right respectively
The assignment answered.The process of said extracted entity type, for the semantic process for determining trunk word of identification, that is, the word extracted retains
The original semantic expression of former natural language sentence.Above-mentioned natural language template is the corresponding entity class of above-mentioned trunk word
Type, the sequence formed according to the arrangement order in former natural language sentence.Then by by above-mentioned sequence inputting to training
In good neural network, by the fitting and iterative calculation in coding and decoding, natural language template is made to be gradually converted into program
The sequence of template, and reserved slot to be filled.The present embodiment is realized by the way that natural language sentence is converted directly into program language
Programming personnel can be allowed to express expected operation using natural language, programmer is facilitated to examine and use.
Further, the filling module 5, comprising:
Determination unit, for according to the determination of preset matching calculation formula and the matched first instance of the first slot, wherein described
First instance is contained in all entities, and first slot is contained in all slots to be filled.
Fills unit, for if so, filling the first instance into first slot.
Generation unit is completed all described wait fill out in described program template for the filling process according to first slot
The filling of slot is filled, described program language is generated.
The present embodiment is by carrying out matching pairing for entity and slot, and realization accurately fills entity in slot, and solving will
Correctly input character string is aligned with the parameter value of output order.The preset matching calculation formula of the present embodiment is to pass through local
It improves k- nearest neighbor algorithm with function to complete, since the Euclidean distance etc. in existing k- nearest neighbor algorithm is not suitable for the application
In slot and entity matching, slot and entity in the application are without specific corresponding vector, and the present embodiment is by will be apart from nearest
Entity and slot composition data pair, be spliced into a long vector, by calculate two long vectors between similarity, and then
Matching degree between slot and entity to be matched.The accurate alignment for realizing slot and entity through the above way, is improved from nature
Accurate correspondence of the language statement to program language.
Further, the determination unit, comprising:
Computation subunit, for passing through calculation formulaIt calculates separately
Matching fractional value between all entities and first slot, wherein d(i,j),(c,d)=cos ((hi,h`j),(hc,h`d)),hiIndicate the corresponding vector of all entities, hj` indicates that first slot is corresponding
Vector, (ec,sd) it is the second instance and the second slot for having marked matching status, hcIndicate the second instance ecIt is corresponding to
Amount, hd` indicates the second slot sdCorresponding vector.
First determines subelement, when for determining the matching fractional value maximum, specified reality corresponding with first slot
Body.
Second determines subelement, for the designated entities to be determined as the first instance.
In the present embodiment, k- nearest neighbor algorithm is improved by local matching function, according to k- nearest neighbor algorithm by calculating
Probability of occurrence in adjacent area is modified to judge the principle of similitude by selecting apart from entity to be matched and slot number according to right
The entity and slot number of nearest marked matched data are according to matching score that is right, and calculating above-mentioned two pairs of entities and slot number evidence pair
Value to overcome the problems, such as not being directly realized by accurate map vector between existing entity and slot, and is calculated by improved k- neighbour
Method reduces calculation amount and improves the matching accuracy of entity and slot.Citing ground, above-mentioned second instance ecWith the second slot sd, point
A corresponding hidden layer variable h not in Encoder model and Decoder modelc,hd`, by the way that variable is spliced, formed it is long to
Measure (hc,hd`), reference entity and slot number are constituted according to right, and identifies matching status, matching identification 1 is than matching identification
0.By fixed slot to be filled, i.e. the first slot, match all entities respectively, form it is multiple include the first slot entity and
Slot number is according to right, same hiIndicate all entities corresponding hidden layer variable, h in Encoder model respectivelyj` indicates described first
Corresponding hidden layer variable in slot Decoder model, by being spliced to form long vector (hi, hj`), by screening apart to be analyzed
The first slot entity and slot number according to right, the distance value nearest second instance for having marked matching status and the second slot, and calculate
With the matching fractional value of the second instance for having marked matching status and the second slot, matching fractional value is bigger, then shows to be analyzed
The entity of first slot and slot number are more similar to the second instance for having marked matching status and the second slot according to right.For example it marks
Second instance and the second slot with state are labeled as matching, then with it the entity of similar first slot to be analyzed and slot number according to right
Also it is estimated as matching, realizes the accurate correspondence of entity and slot.
Further, the neural network model includes coding-decoded model, conversion module 4, comprising:
Fitting unit, for the natural language template to be inputted in the hidden layer for decoding encoding model of pre-training, into
Row fitting obtains fitting state natural language template and exports, wherein the hidden layer of the decoding encoding model includes first specified
The quantity of entity type for including in the form of quantity, first specified quantity and the natural language template is identical.
Input unit, for leading to the hidden layer of the decoded model of the fitting state natural language template input pre-training
It crosses iteration input and calculates list and the output for being converted to ordering described program language template, wherein described program language template
Retain the slot to be filled according to preset format.
The present embodiment is realized by the hidden layer of Encoder and is fitted to the dimensionality reduction of natural language template, and treatment process is reduced
In complexity, and improve the correct recognition rata to the extraction feature of natural language template.Encoder includes in the present embodiment
The corresponding 7 fitting forms of 4 layers of hidden layer are followed successively by H1 to H7, set 20 for the step-length of every layer of LSTM；In MSCOCO data set
On, hidden layer output is set as 1000, and on Flickr30K data set, the output of LSTM hidden layer is 512, hidden on Flickr8K
Layer output is set as 256, to improve dimensionality reduction effect.The hidden layer of the Decoder of the present embodiment is by receiving Encoder output
The last one hidden layer is fitted form as input, by successive ignition input and output, obtains ordering described program language
The list of template, and slot to be filled is remained according to preset mode.The hidden layer of above-mentioned Encoder and hiding for Decoder
Layer is connected, and trains optimum parameter by marked training sample, the program for exporting Encoder-Decoder model
Template infinite approach preset mode.
Further, the input unit, comprising:
First input subelement, for being input to solution using the fitting state natural language template as the first input data
The first gradient layer of the hidden layer of code model, determines the corresponding first instance type of the highest vector of weight coefficient and exports.
Second input subelement, for using the calculating data of the first instance type and the first gradient layer as the
Two input datas are input to the second gradient layer of the hidden layer of decoded model, determine time high vector of weight coefficient corresponding the
Two entity types simultaneously export, wherein the first gradient layer and the second gradient layer are adjacent company in the hidden layer of the decoded model
The iteration layer connect.
Third inputs subelement, for according to the iteration mistake for inputting data into second input data by described first
Journey, successively iteration is input to the hidden layer of the decoded model, until determine all trunk words of the natural language sentence,
Wherein each trunk word is determined according to the sequence of corresponding weight coefficient respectively, and according to the sequence of weight coefficient,
Form the list of the described program language template of sequence.
The present embodiment passes through multiple gradient layers interconnected, is iterated input and output, such as including H1` to H5`, will
The last one hidden layer of Encoder output is fitted form, i.e., fitting state natural language template is as the first input data, input
To the H1` of the hidden layer of Decoder, the primitive most important first instance type Y1` of justice to natural language sentence is exported, then
By Y1` and the process data for extracting Y1` while the input for being used as H2` are calculated, exports the primitive justice time to natural language sentence
Then the process data that Y2` and calculating are extracted to Y2` is used as the input of H3` by important second instance type Y2` simultaneously,
Successively iteration carries out calculating extraction, until output includes the column of the described program language template of all entity types of preset quantity
Table.Citing ground, user's input " journal file before searching 15 days ", corresponding natural language template are as follows: find all [file
Name] files older than [Timespan], by Encoder fitting output after, by Decoder hidden layer repeatedly
It is calculated for input and output, the list of the described program language template of obtained sequence are as follows: Y1:find [path]-name
[regex]-mtime[+Timespan]；Y2:find[path]-type f-name[regex]-mtime[+Timespan]；
Y3:find[path]-type f-perm[permission]；Y4:find[path]-name[regex]-mtime[+
Timespan]xargs ls；Y5:....
Further, the discriminating module 2, comprising:
Call unit, for calling the regular expression being pre-configured according to the constituent element of described program template, wherein institute
The type for stating regular expression includes at least two kinds.
Judging unit, for judge in the natural language sentence according to the first regular expression, if include with it is described
The given entity type of first regular expression adaptation, wherein first regular expression is contained in all canonical tables
Up in formula.
Extraction unit is used for if so, corresponding by first regular expression extraction given entity type
Entity.
The present embodiment identifies different entity types by matching different regular expressions, and extracts corresponding real
Body, method is easy, and calculation amount is small and identifies that the accuracy for extracting entity is high.By by natural language sentence be sequentially inputted to it is each just
Then in expression formula, to judge in natural language sentence whether to include entity type with current regular expression matching, and if it exists,
Then extract the entity.Multiple regular expressions of the present embodiment arrangement can previously according to the important procedure of each entity type,
Successively descending is arranged, and is arranged so that the entity type of identification is formed according to important order.
Further, the matching module 3, comprising:
Acquiring unit, for obtaining the corresponding entity of the mode class entity and physical quantities and the quantity respectively
The corresponding entity of class entity and physical quantities.
Unit is formed, for according to the corresponding entity of the mode class entity and physical quantities and the quantity
The corresponding entity of class entity and physical quantities arrange the corresponding entity of the mode class entity and and described according to preset order
The corresponding entity of quantity class entity forms the natural language template.
The entity type of the present embodiment includes multiple mode class entities and multiple quantity class entities, such as including 5 modes
Class entity, comprising: " File:file name；Directory:directory name；Path:absolute path；Date/
Time:date and time expression；Regex:other pattern arguments " and 3 quantity classes are real
Body, including " Number:number；Size:file size；Timespan:time duration ", to cover certainly as far as possible comprehensively
All trunk words of right language statement.By determining that different entities type and variant entity type correspond to natural language
The importance ranking of sentence forms natural language template, while determining and extracting the corresponding entity of variant entity type,
Assignment as slot to be filled.For example, user's input " journal file before searching 15 days, i.e. find all log files
Older than 15days ", trunk word include " file name:log；Timespan:15days ", corresponding natural language
Template are as follows: find all [file name] files older than [Timespan].
Referring to Fig. 3, a kind of computer equipment is also provided in the embodiment of the present application, which can be server,
Its internal structure can be as shown in Figure 3.The computer equipment includes processor, the memory, network connected by system bus
Interface and database.Wherein, the processor of the Computer Design is for providing calculating and control ability.The computer equipment is deposited
Reservoir includes non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program
And database.The internal memory provides environment for the operation of operating system and computer program in non-volatile memory medium.It should
All data that the database of computer equipment is needed for the matched process of storage address.The network interface of the computer equipment
For being communicated with external end by network connection.To realize that natural language is converted into when the computer program is executed by processor
The method of program language.
Above-mentioned processor executes the method that above-mentioned natural language is converted into program language, comprising: obtains oneself of user's input
Right language statement；All entity types for including in the natural language sentence are distinguished, and extract all entity types point
Not corresponding entity；Natural language template is matched according to all entity types；By the natural language template by instructing in advance
Experienced neural network model is converted into program language template, wherein described program language template carries slot to be filled；By all institutes
It states entity and respectively corresponds the slot to be filled filled into described program template, generate the corresponding journey of the natural language sentence
Sequence language.
Above-mentioned computer equipment, by extracting the entity type in natural language sentence, and according to the reality of extraction
Body type generates natural language template, then natural language template is input in trained neural network, by natural language
Template is converted into process template by the Fitting Calculation and iterative calculation, and is made in to-fill procedure template in advance by training design
Slot to be filled having been stayed, for filling the entity in natural language sentence into process template, it is corresponding to form natural language sentence
Program language.By the way that natural language sentence is converted directly into program language, realization can allow programming personnel to use nature language
Speech facilitates programmer to examine and use to express expected operation.K- nearest neighbor algorithm is improved by local matching function come complete
At, solve the problems, such as that the Euclidean distance etc. in existing k- nearest neighbor algorithm is not suitable for the matching of slot and entity in the application,
By being spliced into a long vector for the data pair apart from nearest entity and slot composition, by calculating between two long vectors
Similarity, and then obtain the matching degree between slot and entity to be matched.The accurate of slot and entity is realized through the above way
Alignment improves the accurate correspondence from natural language sentence to program language.It is identified by matching different regular expressions
Different entity types, and corresponding entity is extracted, method is easy, and calculation amount is small and identifies that the accuracy for extracting entity is high.It is logical
Cross and natural language sentence be sequentially inputted in each regular expression, with judge in natural language sentence whether include with currently just
The then matched entity type of expression formula, and if it exists, then extract the entity.
In one embodiment, all entities are respectively corresponded filling into described program template by above-mentioned processor
The slot to be filled, the step of generating the natural language sentence corresponding program language, comprising: calculated according to preset matching public
Formula determination and the matched first instance of the first slot, wherein the first instance is contained in all entities, first slot
It is contained in all slots to be filled；If so, filling the first instance into first slot；According to described first
The filling process of slot completes the filling of all slots to be filled in described program template, generates described program language.
In one embodiment, above-mentioned processor is determining real with the first slot matched first according to preset matching calculation formula
The step of body, comprising: pass through calculation formulaIt calculates separately all described
Matching fractional value between entity and first slot, wherein d(i,j),(c,d)=cos ((hi,h`j),(hc,h`d)),hiIndicate the corresponding vector of all entities, hj` indicates that first slot is corresponding
Vector, (ec,sd) it is the second instance and the second slot for having marked matching status, hcIndicate the second instance ecIt is corresponding to
Amount, hd` indicates the second slot sdCorresponding vector；When determining the matching fractional value maximum, finger corresponding with first slot
Determine entity；The designated entities are determined as the first instance.
In one embodiment, the neural network model includes coding-decoded model, and above-mentioned processor is by the nature
The step of language template is converted into program language template by the neural network model of pre-training, comprising: by the natural language
Template inputs in the hidden layer of decoding encoding model of pre-training, is fitted to obtain fitting state natural language template and export,
Wherein, it is described decoding encoding model hidden layer include the first specified quantity form, first specified quantity and it is described from
The quantity for the entity type for including in right language template is identical；By the decoding of the fitting state natural language template input pre-training
The hidden layer of model calculates list and the output for being converted to ordering described program language template by iteration input, wherein
Described program language template retains the slot to be filled according to preset format.
In one embodiment, the fitting state natural language template is inputted the decoded model of pre-training by above-mentioned processor
Hidden layer, by iteration input calculate be converted to list and the output of ordering described program language template the step of, packet
It includes: using the fitting state natural language template as the first input data, being input to the first gradient of the hidden layer of decoded model
Layer, determines the corresponding first instance type of the highest vector of weight coefficient and exports；By the first instance type and described the
The calculating data of one gradient layer are input to the second gradient layer of the hidden layer of decoded model as the second input data, determine power
The weight high corresponding second instance type of vector of coefficient time simultaneously exports, wherein the first gradient layer and the second gradient layer are described
The iteration layer of adjacent connection in the hidden layer of decoded model；Second input data is input data into according to by described first
Iterative process, successively iteration is input to the hidden layer of the decoded model, until determining all masters of the natural language sentence
Dry word, wherein each trunk word is determined according to the sequence of corresponding weight coefficient respectively, and according to weight coefficient
Sequence, form the list of the described program language template of sequence.
In one embodiment, above-mentioned processor distinguishes all entity types for including in the natural language sentence, and
The step of extracting all entity types corresponding entity, comprising: call the constituent element according to described program template
The regular expression of pre-configuration, wherein the type of the regular expression includes at least two kinds；Sentenced according to the first regular expression
Break in the natural language sentence, if include the given entity type being adapted to first regular expression, wherein described
First regular expression is contained in all regular expressions；If so, extracting institute by first regular expression
State the corresponding entity of given entity type.
In one embodiment, the entity type includes mode class entity and quantity class entity, according to all realities
The step of body type matching natural language template, comprising: the corresponding entity of the mode class entity and physical quantities are obtained respectively,
And the corresponding entity of the quantity class entity and physical quantities；According to the corresponding entity of the mode class entity and entity
Quantity and the corresponding entity of the quantity class entity and physical quantities arrange the mode class entity pair according to preset order
The entity answered with and the corresponding entity of the quantity class entity, form the natural language template.
It will be understood by those skilled in the art that structure shown in Fig. 3, only part relevant to application scheme is tied
The block diagram of structure does not constitute the restriction for the computer equipment being applied thereon to application scheme.
One embodiment of the application also provides a kind of computer readable storage medium, is stored thereon with computer program, calculates
The method that natural language is converted into program language is realized when machine program is executed by processor, comprising: obtain the nature of user's input
Language statement；All entity types for including in the natural language sentence are distinguished, and extract all entity type difference
Corresponding entity；Natural language template is matched according to all entity types；The natural language template is passed through into pre-training
Neural network model be converted into program language template, wherein described program language template carries slot to be filled；It will be all described
Entity respectively corresponds the slot to be filled filled into described program template, generates the corresponding program of the natural language sentence
Language.
Above-mentioned computer readable storage medium, by the way that the entity type in natural language sentence is extracted, and according to
The entity type of extraction generates natural language template, and then natural language template is input in trained neural network, will
Natural language template is converted into process template by the Fitting Calculation and iterative calculation, and makes to-fill procedure by training design
Slot to be filled has been reserved in template, for filling the entity in natural language sentence into process template, forms natural language
The corresponding program language of sentence.By the way that natural language sentence is converted directly into program language, realization can allow programming personnel to make
Expected operation is expressed with natural language, and programmer is facilitated to examine and use.K- neighbour is improved by local matching function
Algorithm is completed, and solves the matching that the Euclidean distance etc. in existing k- nearest neighbor algorithm is not suitable for slot and entity in the application
The problem of, by the data pair that will be formed apart from nearest entity and slot, be spliced into a long vector, by calculate it is two long to
Similarity between amount, and then obtain the matching degree between slot and entity to be matched.Slot and entity are realized through the above way
Accurate alignment, improve accurate correspondence from natural language sentence to program language.By matching different regular expressions
It identifies different entity types, and extracts corresponding entity, method is easy, and calculation amount is small and the accuracy of entity is extracted in identification
It is high.By the way that natural language sentence is sequentially inputted in each regular expression, with judge in natural language sentence whether include with
The entity type of current regular expression matching, and if it exists, then extract the entity.
In one embodiment, all entities are respectively corresponded filling into described program template by above-mentioned processor
The slot to be filled, the step of generating the natural language sentence corresponding program language, comprising: calculated according to preset matching public
Formula determination and the matched first instance of the first slot, wherein the first instance is contained in all entities, first slot
It is contained in all slots to be filled；If so, filling the first instance into first slot；According to described first
The filling process of slot completes the filling of all slots to be filled in described program template, generates described program language.
In one embodiment, above-mentioned processor is determining real with the first slot matched first according to preset matching calculation formula
The step of body, comprising: pass through calculation formulaIt calculates separately all described
Matching fractional value between entity and first slot, wherein d(i,j),(c,d)=cos ((hi,h`j),(hc,h`d)),hiIndicate the corresponding vector of all entities, hj` indicates that first slot is corresponding
Vector, (ec,sd) it is the second instance and the second slot for having marked matching status, hcIndicate the second instance ecIt is corresponding to
Amount, hd` indicates the second slot sdCorresponding vector；When determining the matching fractional value maximum, finger corresponding with first slot
Determine entity；The designated entities are determined as the first instance.
In one embodiment, the neural network model includes coding-decoded model, and above-mentioned processor is by the nature
The step of language template is converted into program language template by the neural network model of pre-training, comprising: by the natural language
Template inputs in the hidden layer of decoding encoding model of pre-training, is fitted to obtain fitting state natural language template and export,
Wherein, it is described decoding encoding model hidden layer include the first specified quantity form, first specified quantity and it is described from
The quantity for the entity type for including in right language template is identical；By the decoding of the fitting state natural language template input pre-training
The hidden layer of model calculates list and the output for being converted to ordering described program language template by iteration input, wherein
Described program language template retains the slot to be filled according to preset format.
In one embodiment, the fitting state natural language template is inputted the decoded model of pre-training by above-mentioned processor
Hidden layer, by iteration input calculate be converted to list and the output of ordering described program language template the step of, packet
It includes: using the fitting state natural language template as the first input data, being input to the first gradient of the hidden layer of decoded model
Layer, determines the corresponding first instance type of the highest vector of weight coefficient and exports；By the first instance type and described the
The calculating data of one gradient layer are input to the second gradient layer of the hidden layer of decoded model as the second input data, determine power
The weight high corresponding second instance type of vector of coefficient time simultaneously exports, wherein the first gradient layer and the second gradient layer are described
The iteration layer of adjacent connection in the hidden layer of decoded model；Second input data is input data into according to by described first
Iterative process, successively iteration is input to the hidden layer of the decoded model, until determining all masters of the natural language sentence
Dry word, wherein each trunk word is determined according to the sequence of corresponding weight coefficient respectively, and according to weight coefficient
Sequence, form the list of the described program language template of sequence.
In one embodiment, above-mentioned processor distinguishes all entity types for including in the natural language sentence, and
The step of extracting all entity types corresponding entity, comprising: call the constituent element according to described program template
The regular expression of pre-configuration, wherein the type of the regular expression includes at least two kinds；Sentenced according to the first regular expression
Break in the natural language sentence, if include the given entity type being adapted to first regular expression, wherein described
First regular expression is contained in all regular expressions；If so, extracting institute by first regular expression
State the corresponding entity of given entity type.
In one embodiment, the entity type includes mode class entity and quantity class entity, according to all realities
The step of body type matching natural language template, comprising: the corresponding entity of the mode class entity and physical quantities are obtained respectively,
And the corresponding entity of the quantity class entity and physical quantities；According to the corresponding entity of the mode class entity and entity
Quantity and the corresponding entity of the quantity class entity and physical quantities arrange the mode class entity pair according to preset order
The entity answered with and the corresponding entity of the quantity class entity, form the natural language template.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, above-mentioned computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
Any reference used in provided herein and embodiment to memory, storage, database or other media,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double speed are according to rate SDRAM (SSRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, device, article or the method that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, device, article or method institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, device of element, article or method.
The foregoing is merely preferred embodiment of the present application, are not intended to limit the scope of the patents of the application, all utilizations
Equivalent structure or equivalent flow shift made by present specification and accompanying drawing content is applied directly or indirectly in other correlations
Technical field, similarly include in the scope of patent protection of the application.
Claims (10)
1. a kind of method that natural language is converted into program language characterized by comprising
Obtain the natural language sentence of user's input；
All entity types for including in the natural language sentence are distinguished, and it is corresponding to extract all entity types
Entity；
Natural language template is matched according to all entity types；
The natural language template is converted into program language template by the neural network model of pre-training, wherein the journey
Sequence language template carries slot to be filled；
All entities are respectively corresponded into the slot to be filled filled into described program template, generate the natural language
The corresponding program language of sentence.
2. the method that natural language according to claim 1 is converted into program language, which is characterized in that described by all institutes
It states entity and respectively corresponds the slot to be filled filled into described program template, generate the corresponding journey of the natural language sentence
The step of sequence language, comprising:
According to the determination of preset matching calculation formula and the matched first instance of the first slot, wherein the first instance is contained in institute
Have in the entity, first slot is contained in all slots to be filled；
If so, filling the first instance into first slot；
The filling of all slots to be filled in described program template is completed according to the filling process of first slot, generates institute
State program language.
3. the method that natural language according to claim 2 is converted into program language, which is characterized in that the basis is default
Matching primitives formula determines the step of first instance matched with the first slot, comprising:
Pass through calculation formulaCalculate separately all entities with it is described
Matching fractional value between first slot, wherein d(i,j),(c,d)=cos ((hi,hj`),(hc,hd`)),hiIndicate the corresponding vector of all entities, hj` indicates that first slot is corresponding
Vector, (ec,sd) it is the second instance and the second slot for having marked matching status, hcIndicate the second instance ecIt is corresponding to
Amount, hd` indicates the second slot sdCorresponding vector；
When determining the matching fractional value maximum, designated entities corresponding with first slot；
The designated entities are determined as the first instance.
4. the method that natural language according to claim 1 is converted into program language, which is characterized in that the neural network
Model includes coding-decoded model, described that the natural language template is converted into journey by the neural network model of pre-training
The step of sequence language template, comprising:
The natural language template is inputted in the hidden layer of decoding encoding model of pre-training, is fitted to obtain fitting state certainly
Right language template simultaneously exports, wherein form of the hidden layer comprising the first specified quantity of the decoding encoding model, described first
The quantity of entity type for including in specified quantity and the natural language template is identical；
By the hidden layer of the decoded model of the fitting state natural language template input pre-training, is inputted by iteration and calculate conversion
List and output for ordering described program language template, wherein described program language template retains institute according to preset format
State slot to be filled.
5. the method that natural language according to claim 4 is converted into program language, which is characterized in that described described to intend
The hidden layer for closing the decoded model of state natural language template input pre-training, is calculated by iteration input and is converted to ordering institute
The step of stating the list of program language template and output, comprising:
Using the fitting state natural language template as the first input data, it is input to the first gradient of the hidden layer of decoded model
Layer, determines the corresponding first instance type of the highest vector of weight coefficient and exports；
Using the first instance type and the calculating data of the first gradient layer as the second input data, it is input to decoding mould
Second gradient layer of the hidden layer of type determines the high corresponding second instance type of vector of weight coefficient time and exports, wherein institute
It states first gradient layer and the second gradient layer is the iteration layer of adjacent connection in the hidden layer of the decoded model；
According to the iterative process for inputting data into second input data by described first, successively iteration is input to the decoding
The hidden layer of model, until all trunk words of the natural language sentence are determined, wherein each trunk word distinguishes root
It is determined according to the sequence of corresponding weight coefficient, and according to the sequence of weight coefficient, forms the described program language mould of sequence
The list of plate.
6. the method that natural language according to claim 1 is converted into program language, which is characterized in that described in the differentiation
All entity types for including in natural language sentence, and the step of extracting all entity types corresponding entity,
Include:
The regular expression being pre-configured according to the constituent element of described program template is called, wherein the type of the regular expression
Including at least two kinds；
Judged in the natural language sentence according to the first regular expression, if comprising being adapted to first regular expression
Given entity type, wherein first regular expression is contained in all regular expressions；
If so, extracting the corresponding entity of the given entity type by first regular expression.
7. the method that natural language according to claim 1 is converted into program language, which is characterized in that the entity type
Including mode class entity and quantity class entity, described the step of matching natural language template according to all entity types, packet
It includes:
Obtain respectively the corresponding entity of the mode class entity and physical quantities and the corresponding entity of the quantity class entity and
Physical quantities；
According to the corresponding entity of the mode class entity and physical quantities and the corresponding entity of the quantity class entity and
Physical quantities, according to preset order arrange the corresponding entity of the mode class entity with and the corresponding reality of the quantity class entity
Body forms the natural language template.
8. the device that a kind of natural language is converted into program language characterized by comprising
Module is obtained, for obtaining the natural language sentence of user's input；
Discriminating module for distinguishing all entity types for including in the natural language sentence, and extracts all entities
The corresponding entity of type；
Matching module, for matching natural language template according to all entity types；
Conversion module, for the natural language template to be converted into program language mould by the neural network model of pre-training
Plate, wherein described program language template carries slot to be filled；
Module is filled, it is raw for all entities to be respectively corresponded the slot to be filled filled into described program template
At the corresponding program language of the natural language sentence.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In the step of processor realizes any one of claims 1 to 7 the method when executing the computer program.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method described in any one of claims 1 to 7 is realized when being executed by processor.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910600648.5A CN110442332A (en) | 2019-07-04 | 2019-07-04 | Natural language is converted into the method, apparatus and computer equipment of program language |
| PCT/CN2019/122480 WO2021000512A1 (en) | 2019-07-04 | 2019-12-02 | Method and apparatus for converting natural language into programing language, and computer device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910600648.5A CN110442332A (en) | 2019-07-04 | 2019-07-04 | Natural language is converted into the method, apparatus and computer equipment of program language |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110442332A true CN110442332A (en) | 2019-11-12 |
Family
ID=68428517
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910600648.5A Pending CN110442332A (en) | 2019-07-04 | 2019-07-04 | Natural language is converted into the method, apparatus and computer equipment of program language |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110442332A (en) |
| WO (1) | WO2021000512A1 (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111008213A (en) * | 2019-12-23 | 2020-04-14 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating language conversion model |
| CN111191439A (en) * | 2019-12-16 | 2020-05-22 | 浙江大搜车软件技术有限公司 | Natural sentence generation method and device, computer equipment and storage medium |
| WO2021000512A1 (en) * | 2019-07-04 | 2021-01-07 | 深圳壹账通智能科技有限公司 | Method and apparatus for converting natural language into programing language, and computer device |
| CN112306497A (en) * | 2020-11-03 | 2021-02-02 | 高炼 | Method and system for converting natural language into program code |
| CN113934684A (en) * | 2021-12-16 | 2022-01-14 | 四川省交通勘察设计研究院有限公司 | Data format automatic conversion method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101930428A (en) * | 2009-06-18 | 2010-12-29 | 万继华 | System and method for ensuring computer to understand natural languages |
| CN109144493A (en) * | 2018-07-18 | 2019-01-04 | 平安科技（深圳）有限公司 | Generation method, electronic device and the readable storage medium storing program for executing of business rule |
| CN109739483A (en) * | 2018-12-28 | 2019-05-10 | 北京百度网讯科技有限公司 | Method and apparatus for generated statement |
| CN109918680A (en) * | 2019-03-28 | 2019-06-21 | 腾讯科技（上海）有限公司 | Entity recognition method, device and computer equipment |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120101803A1 (en) * | 2007-11-14 | 2012-04-26 | Ivaylo Popov | Formalization of a natural language |
| CN103617159B (en) * | 2012-12-07 | 2016-10-12 | 万继华 | Natural language translation is become the method for computer language, semantic analyzer and interactive system |
| CN107291701B (en) * | 2016-04-01 | 2020-12-01 | 阿里巴巴集团控股有限公司 | Machine language generation method and device |
| CN110442332A (en) * | 2019-07-04 | 2019-11-12 | 深圳壹账通智能科技有限公司 | Natural language is converted into the method, apparatus and computer equipment of program language |
-
2019
- 2019-07-04 CN CN201910600648.5A patent/CN110442332A/en active Pending
- 2019-12-02 WO PCT/CN2019/122480 patent/WO2021000512A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101930428A (en) * | 2009-06-18 | 2010-12-29 | 万继华 | System and method for ensuring computer to understand natural languages |
| CN109144493A (en) * | 2018-07-18 | 2019-01-04 | 平安科技（深圳）有限公司 | Generation method, electronic device and the readable storage medium storing program for executing of business rule |
| CN109739483A (en) * | 2018-12-28 | 2019-05-10 | 北京百度网讯科技有限公司 | Method and apparatus for generated statement |
| CN109918680A (en) * | 2019-03-28 | 2019-06-21 | 腾讯科技（上海）有限公司 | Entity recognition method, device and computer equipment |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021000512A1 (en) * | 2019-07-04 | 2021-01-07 | 深圳壹账通智能科技有限公司 | Method and apparatus for converting natural language into programing language, and computer device |
| CN111191439A (en) * | 2019-12-16 | 2020-05-22 | 浙江大搜车软件技术有限公司 | Natural sentence generation method and device, computer equipment and storage medium |
| CN111008213A (en) * | 2019-12-23 | 2020-04-14 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating language conversion model |
| CN112306497A (en) * | 2020-11-03 | 2021-02-02 | 高炼 | Method and system for converting natural language into program code |
| CN112306497B (en) * | 2020-11-03 | 2024-04-26 | 高炼 | Method and system for converting natural language into program code |
| CN113934684A (en) * | 2021-12-16 | 2022-01-14 | 四川省交通勘察设计研究院有限公司 | Data format automatic conversion method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021000512A1 (en) | 2021-01-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110442332A (en) | Natural language is converted into the method, apparatus and computer equipment of program language | |
| CN111506722B (en) | Knowledge graph question-answering method, device and equipment based on deep learning technology | |
| CN110147445A (en) | Intension recognizing method, device, equipment and storage medium based on text classification | |
| CN112256825A (en) | Medical field multi-turn dialogue intelligent question-answering method and device and computer equipment | |
| KR102452123B1 (en) | Apparatus for Building Big-data on unstructured Cyber Threat Information, Method for Building and Analyzing Cyber Threat Information | |
| CN109829692A (en) | Contract trial method, apparatus, equipment and storage medium based on artificial intelligence | |
| CA3135717A1 (en) | System and method for transferable natural language interface | |
| CN112016279A (en) | Electronic medical record structuring method and device, computer equipment and storage medium | |
| CN112052684A (en) | Named entity identification method, device, equipment and storage medium for power metering | |
| CN109992664A (en) | Mark classification method, device, computer equipment and the storage medium of central issue | |
| CN111710383A (en) | Medical record quality control method and device, computer equipment and storage medium | |
| CN108664595A (en) | Domain knowledge base construction method, device, computer equipment and storage medium | |
| CN113032568A (en) | Query intention identification method based on bert + bilstm + crf and combined sentence pattern analysis | |
| CN112766319A (en) | Dialogue intention recognition model training method and device, computer equipment and medium | |
| JP7579022B1 (en) | Method and system for intelligent analysis of bills based on semantic graph model | |
| CN113204968B (en) | Concept identification method, device, equipment and storage medium of medical entity | |
| CN112036151A (en) | Method and device for constructing gene disease relation knowledge base and computer equipment | |
| CN116340467B (en) | Text processing method, text processing device, electronic equipment and computer readable storage medium | |
| CN113868422A (en) | Multi-label inspection work order problem traceability identification method and device | |
| CN114004231A (en) | Chinese special word extraction method, system, electronic equipment and storage medium | |
| CN109885830A (en) | Sentence interpretation method, device, computer equipment | |
| CN116341519A (en) | Event causal relation extraction method, device and storage medium based on background knowledge | |
| CN111680132A (en) | Noise filtering and automatic classifying method for internet text information | |
| CN111782821B (en) | Medical hotspot prediction method and device based on FM model and computer equipment | |
| CN113051920A (en) | Named entity recognition method and device, computer equipment and storage medium |
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 |