CN109960806A - A kind of natural language processing method - Google Patents
A kind of natural language processing method Download PDFInfo
- Publication number
- CN109960806A CN109960806A CN201910231479.2A CN201910231479A CN109960806A CN 109960806 A CN109960806 A CN 109960806A CN 201910231479 A CN201910231479 A CN 201910231479A CN 109960806 A CN109960806 A CN 109960806A
- Authority
- CN
- China
- Prior art keywords
- data
- sentence
- lteral
- lteral data
- information
- 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 80
- 238000003058 natural language processing Methods 0.000 title claims abstract description 26
- 230000008569 process Effects 0.000 claims abstract description 47
- 241001269238 Data Species 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims description 26
- 239000013589 supplement Substances 0.000 claims description 17
- 238000003780 insertion Methods 0.000 claims description 4
- 230000037431 insertion Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 abstract description 9
- 238000012795 verification Methods 0.000 description 10
- 230000006870 function Effects 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/253—Grammatical analysis; Style critique
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The present invention relates to a kind of natural language processing methods, which comprises semantic processes system receives phrase data, obtains sentence lteral data;Extract the fixation language information and extensive object information in sentence lteral data;When the fixation language information in current statement lteral data or the extensive object information in current statement lteral data are not empty, one or more semantic matches result datas are obtained according to the extensive object information in the fixation language information and current statement lteral data in current statement lteral data;When the fixation language information in current statement lteral data or the extensive object information in current statement lteral data are empty, semantic matches result data is obtained according to association sentence lteral data and current statement lteral data;When the number of semantic matches result data is multiple, according to the corresponding precedence information of semantic matches result data, the semantic matches result data of highest priority is determined;The highest semantic matches result data of output priority.
Description
Technical field
The present invention relates to technical field of data processing more particularly to a kind of natural language processing methods.
Background technique
Natural language processing (Natural Language Processing, NLP) is artificial intelligence and linguistics field
One of problem the most difficult in subdiscipline and artificial intelligence understands that complicated language is also the important set of artificial intelligence
It at part, and is also full of challenges to the research of natural language processing.NLP using ubiquitous, because of people's language
Most of communication, such as web search are carried out, advertisement, Email, customer service, language translation, publication journal announcement etc. is all
It can be applied among NLP.But how during Language Processing determine user point of interest (Point Of Interest,
POI) and sentence application field carries out semantic matches to sentence, and how to carry out semantic matches to semantic incomplete sentence, at
In order to which there is one of difficulty to be solved in this field.
Summary of the invention
The purpose of the present invention is in view of the drawbacks of the prior art, providing a kind of natural language processing method, by sentence
The extensive processing of clause, the fixation language and extensive object in clause can be extracted, when the fixed language of current statement or extensive object lack
When mistake, it can be associated with corresponding fixed language or extensive object from the context, and corresponding final according to priority output
Matching result determines the process of final semantic matches result fast and accurately by the method.
To achieve the above object, the present invention provides a kind of natural language processing methods, which comprises
Semantic processes system receives phrase data, obtains sentence lteral data;
The extensive processing of clause is carried out to the sentence lteral data according to syntax rule tree, extracts the sentence lteral data
In fixation language information and extensive object information;
Determine the fixation language information or extensive in the current statement lteral data in the current statement lteral data
Whether object information is empty;
When fixation language information in the current statement lteral data or extensive right in the current statement lteral data
When image information is not empty, according in the current statement lteral data fixation language information and the current statement lteral data
In extensive object information obtain one or more semantic matches result datas;
When fixation language information in the current statement lteral data or extensive right in the current statement lteral data
When image information is empty, acquisition is associated with sentence lteral data with the current statement lteral data;
Semantic matches result data is obtained according to the association sentence lteral data and the current statement lteral data;
When the number of the semantic matches result data is multiple, according to the corresponding priority of semantic matches result data
Information determines the semantic matches result data of highest priority;
Export the semantic matches result data of the highest priority.
Preferably, the phrase data includes sentence voice data and sentence lteral data;The semantic processes system connects
Phrase data is received, sentence lteral data is obtained specifically:
The speech convertor of the semantic processes system receives the phrase data, to the sentence language in the phrase data
Sound data are identified, obtain the sentence lteral data of the sentence voice data, and by the sentence of the sentence voice data
Lteral data is inserted into the end of the input rank of the semantic processes system;
The interrogator of the semantic processes system monitors the data insertion of the input rank, obtains from the input rank
The sentence lteral data for taking the input rank end obtains the current statement lteral data.
Preferably, it is described according to syntax rule tree to the sentence lteral data carry out the extensive processing of clause before, institute
State method further include:
History sentence lteral data is inputted the clause analytic modell analytical model to be trained, the clause after being trained parses mould
Type, it is extensive to sentence lteral data progress clause according to the clause analytic modell analytical model after the training to semantic processes system
Processing.
Preferably, the fixation language information according in the current statement lteral data and the current statement text number
Extensive object information in obtains one or more semantic matches result datas specifically:
The corresponding field of the sentence lteral data is determined according to the fixation sentence information in the current statement lteral data
Scape data;
Point of interest corresponding with the extensive object information in the current statement lteral data is matched in point of interest library
Data;
Semantic matches result data is obtained according to the interest point data and the contextual data.
Preferably, the association sentence lteral data includes upper sentence lteral data and next statement lteral data.
It is further preferred that when the fixation language information in the current statement lteral data is empty, it is described according to
Association sentence lteral data and the current statement lteral data obtain semantic matches result data specifically:
The fixation language information in upper sentence lteral data is obtained, and according to the fixation in the upper sentence lteral data
Extensive object information in language information and the current statement lteral data obtains semantic matches result data.
It is further preferred that when the extensive object information in the current statement lteral data is empty, it is described according to institute
It states association sentence lteral data and the current statement lteral data obtains semantic matches result data specifically:
The supplement phrase data that user inputs according to boot statement data is received, and is extracted in the supplement phrase data
Extensive object information;
According to the fixed language in the extensive object information and current statement lteral data in the supplement phrase data
Information obtains semantic matches result data.
It is further preferred that before the supplement phrase data that the reception user inputs according to boot statement data, institute
State method further include:
Semantic processes system generates and exports the boot statement.
Preferably, described according to the corresponding precedence information of semantic matches result data, determine the semanteme of highest priority
Matching result data specifically:
The multiple semantic matches result data is determined according to the extensive object information in the current statement lteral data
Corresponding sentence application field information;
Determine that each sentence application field information corresponds to the other precedence information of domain class;
The semantic matches result data of highest priority is determined according to the other precedence information of the domain class.
It is further preferred that the semantic matches result data of the output highest priority specifically:
The output highest semantic matches result data of the precedence information being encapsulated into the semantic processes system
The end of queue;
The interrogator monitors the data insertion of the output queue, and the input rank is obtained from the output queue
The highest semantic matches result data at end, and export.
Natural language processing method provided in an embodiment of the present invention can be extracted by the extensive processing of clause to sentence
Fixation language and extensive object in clause can be associated with from the context when the fixed language of current statement or extensive object missing
Corresponding final matching results are exported to corresponding fixed language or extensive object, and according to priority, are determined by the method
The process of final semantic matches result is fast and accurately.
Detailed description of the invention
Fig. 1 is the flow chart of natural language processing method provided in an embodiment of the present invention.
Specific embodiment
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
A kind of natural language processing method provided in an embodiment of the present invention, for voice system according in user's input
Hold and exports corresponding feedback content.Its method flow diagram is as shown in Figure 1, include the following steps:
Step 101, semantic processes system receives phrase data, obtains sentence lteral data;
Specifically, semantic processes system can be understood as one with input by sentence, the system for handling and exporting function.Language
Adopted processing system includes speech convertor, input rank, interrogator and processor.When the starting of semantic processes system, system is defeated
The monitor configured in page-out is activated, which can load the configuration file for voice service, domain (domain) class
And the output statement of the corresponding user profile of domain, semantic processes system specific condition, while starting voice conversion
Device, input rank, interrogator and processor.
Phrase data includes sentence voice data and sentence lteral data.That is, user can pass through voice or text
The mode of word is to system read statement data.Speech convertor receives phrase data, to the sentence voice data in phrase data
It is identified, obtains the sentence lteral data of sentence voice data, and by the sentence lteral data of sentence voice data or directly
The sentence lteral data of user's input is inserted into the end of the input rank of semantic processing system.
Interrogator can monitor always whether input rank has new message, that is, monitor whether have sentence lteral data into
Enqueue, and from input rank obtain input rank end sentence lteral data, to obtain sentence lteral data.
Step 102, the extensive processing of clause is carried out to sentence lteral data, extracts the fixation language information in sentence lteral data
With extensive object information;
Specifically, sentence lteral data is sent to processor by interrogator after interrogator gets sentence lteral data.
Processor carries out the extensive processing of clause to sentence lteral data according to syntax rule tree, extracts the fixation language in sentence lteral data
Information and extensive object information.The extensive processing of clause is understood that a variety of to be expanded to according to syntax rule tree by a sentence
The expression-form of sentence, and extract the process of key element in sentence.In this process, key element packet in sentence is extracted
Include fixed language information and extensive object information.For example, " I wants to go to " is fixed language in the sentence of one " I wants to go to cinema "
Information, is a part of clause, and " cinema " is extensive object information;For another example, in the sentence of one " I wants to buy film ticket ",
" I wants to buy " is fixed language information, and " film ticket " is extensive object information.
In a specific example, user has input the sentence voice data of " I wants to go to cinema ", language by voice
Sound converter carries out voice recognition processing to the sentence voice data, has obtained the sentence lteral data of " I wants to go to cinema ".
Then the processor in semantic processes system first recognizes " going ", then recognizes " wanting to go to " further along and then recognizes
The clause of " I wants to go to ... ", and according to the-clause of " wanting to go to "-" I wants to go to " " is gone " in preset syntax rule tree, extract language
Fixation language information " going " in sentence lteral data, and the extensive object information of word " cinema " conduct after " I wants to go to ".
It is understood that a sentence lteral data may correspond to a fixed language information and an extensive object letter
Breath, it is also possible to corresponding multiple fixed language information and multiple extensive object informations, but in a normal sentence lteral data, often
A fixed language information should at least be corresponding with an extensive object information.
In some preferred embodiments, what syntax rule tree was also possible to obtain by model training.It is further specific
, the multiple history sentence lteral datas of semantic processes system acquisition are as sample, and multiple history sentence lteral datas are instructed
Practice, obtain clause analytic modell analytical model, syntax rule tree is obtained according to clause analytic modell analytical model, thus according to syntax rule tree to sentence text
Digital data carries out the extensive processing of clause.
Step 103, determine whether fixation language information or extensive object information in current statement lteral data are empty;
Specifically, when in current statement lteral data fixation language information or extensive object information be empty when, illustrate currently
Sentence representated by sentence lteral data is a semantic incomplete sentence, and semantic processes system can not be directly according to current statement
Lteral data obtains the matching result of sentence, needs to be implemented step 104 at this time.For example, a content is the sentence of " Tianjin "
Data, fixing language information is sky, and extensive object information is " Tianjin ", another content is the phrase data of " I will book tickets ",
It is " ticket booking " that it, which fixes language information, and extensive object information is sky, and in the two sentences, speech processing system can not direct root
Its semanteme is determined according to current statement lteral data.
When in current statement lteral data fixation language information or extensive object information be not empty when, illustrate current statement
Sentence representated by lteral data is the complete sentence of semanteme, thens follow the steps 107.
Step 104, the fixation language information or extensive object information in association sentence lteral data are obtained;
Specifically, association sentence lteral data includes the supplement phrase data of upper sentence lteral data and user's input.
The supplement phrase data of user's input can be understood as the next statement of sentence lteral data relative to current statement lteral data
Lteral data.
When the extensive object information in current statement lteral data is not empty, but the fixation language in current statement lteral data
When information is empty, processor needs to obtain the fixation language information in upper sentence lteral data.Obtain upper sentence lteral data
In the mode of fixation language information may include following two.
In the first way, processor obtains upper sentence lteral data from input rank, then to a upper sentence
Lteral data carries out the extensive processing of clause, extracts the fixation language information in upper sentence lteral data.In the process, due to defeated
The fixation language information in history phrase data is not saved in enqueue, so if needing to obtain in upper sentence lteral data
Fixed language information, then need after getting upper sentence lteral data, then parse to upper sentence lteral data, extracts
Fixation language information in upper sentence lteral data.
In the second way, the fixation language that processor can obtain upper sentence lteral data by cache module is believed
Breath.That is, if necessary to obtain the fixation language information in upper sentence lteral data, then it can be directly from cache module
The fixation language information in upper sentence lteral data is obtained, without parsing to upper sentence lteral data.With
Family can according to need the above two mode of selection.
When the fixation language information in current statement lteral data is not empty, but the extensive object information in current statement data
When for sky, processor generates and exports boot statement data, extensive right in input current statement data to prompt user to supplement
Image information.Then processor receives the supplement phrase data that inputs according to boot statement data of user and extracts supplement phrase data
In extensive object information.Extracting method is referred to above-mentioned steps 102.
Preferably, boot statement data are processors according to the fixation language information in current statement lteral data, and are combined
Default that clause is putd question to generate, in a specific example, phrase data is " I will book tickets ", and fixing language information is " to order
Ticket ", extensive object information are sky, then semantic processes system according to be sky fixation language information for " ticket booking ", in conjunction with " ' may I ask
Need '+verb+' what '+noun " default enquirement clause, generate " may I ask and what ticket needed to book " boot statement data, and
Output.The way of output may include voice prompting and/or text prompt.
Step 105, new fixation language information or new extensive object information are verified;
Specifically, here, new fixation language information can be understood as the fixation got from upper sentence lteral data
Language information.New extensive object information can be understood as the extensive object letter got from supplement phrase data.
Since new fixation language information or the new extensive object information of extensive object information are not necessarily applied to current language
In sentence lteral data, the scene of current statement might not be with the scene in a upper sentence or the scene in supplement sentence in other words
It is identical, if not carrying out scene verification to new fixation language information or new extensive object information is likely to mistake occur.Therefore,
Processor was needed the extensive object information generation in the fixation language information or supplement phrase data in upper sentence lteral data
Enter current statement lteral data and carries out scene verification.
Further specifically, in the scene verification for fixed language information, processor needs to obtain upper sentence text
Contextual data corresponding to fixation language information in data substitutes into contextual data corresponding to the fixation language information and its current
Sentence lteral data carries out semantic rules matching with the extensive object information of current statement lteral data, if semantic rules
With success, then illustrate scene verification pass through, on the contrary it is then illustrate scene verify do not pass through.
In a specific example, current statement lteral data is " Tianjin ", and fixing language information is sky, extensive right
Image information is " Tianjin ".A upper sentence lteral data is " my Pekinese Xiang Cha weather ", and fixing language information is " Cha Tianqi ",
The corresponding contextual data of fixation language information " Cha Tianqi " is " weather ".By the fixation language information in upper sentence lteral data
" Cha Tianqi " and the contextual data " weather " corresponding to it substitute into current statement lteral data in extensive object information " Tianjin ",
Semantic rules determine that " Cha Tianqi " matches with " Tianjin ", it is determined that verification passes through.
In the scene verification for extensive object information, processor, which needs to supplement the extensive object in phrase data, to be believed
Breath substitutes into current statement lteral data, semantic rules matching is carried out with the fixation language information of current statement lteral data, if language
On the contrary the success of adopted rule match then illustrates that scene verification passes through, then illustrate that scene verifies and do not pass through.
Step 106, when passed the verification, new sentence lteral data is generated;
Specifically, if by extensive right in new fixation language information substitution current statement data, with current statement data
Image information carries out substituting into current statement data by verification, or by new extensive object information after semantic rules matching, and works as
Fixation language information in preceding phrase data carries out after semantic rules matching through verification, then processor by new fixation language information and
Extensive object information in current statement data combines, or by consolidating in new extensive object information and current statement data
Attribute information combines, and generates new sentence lteral data.
In a specific example, current statement data are " Tianjin ", and fixing language information is sky, extensive object letter
Breath is " Tianjin ", then it is " my Pekinese Xiang Cha day that semantic processes system, which obtains the upper sentence lteral data of the phrase data,
Gas ", fixing language information is " Cha Tianqi ", by extensive object information " Tianjin " He Shangyi sentence text in current statement data
The middle fixed language information " Cha Tianqi " of data combines, and obtains the new sentence lteral data of " Cha Tianjin weather ".
Step 107, according in current statement lteral data fixation language information and extensive object information determine current statement
The corresponding contextual data of lteral data and interest point data;
Specifically, processor determines the corresponding contextual data of sentence lteral data according to fixed sentence information first.Each
Fixed sentence information all can be mapped to a contextual data.Here, each contextual data can be understood as an independent use
Family behavior scene.
In a specific example, current statement lteral data is " I wants to buy film ticket ", then semantic processes system is true
The fixation sentence information of the fixed sentence lteral data is " I wants to buy ", then determines that the fixation sentence information " I wants to buy " is corresponding
Contextual data be user want to buy the scene of some article.For another example, current statement lteral data is " I wants to go to cinema ", then
Semantic processes system determines that the fixation sentence information of the sentence lteral data is " I wants to go to ", then determines the fixation sentence information
Contextual data corresponding to " I wants to go to " is the scene that user wants to go to somewhere.
Meanwhile processor matches interest point data corresponding with extensive object information in point of interest library.Point of interest library
It can be understood as user setting, matching relationship in current semantics processing system, for storing point of interest Yu extensive object
Database.Interest point data in point of interest library can be what user was set as needed.If in point of interest library
When with less than interest point data corresponding with currently extensive object information, then semantic processes system connects outer according to external interface
Database is connect, and in external database matching interest point data corresponding with extensive object information.
In some preferred embodiments, interest point data corresponding with extensive object information is matched in point of interest library
When, it needs acquisition to obtain position data from data source and carries out interest point data as matching condition, and according to matching condition
Matching.That is, needing to obtain user in the corresponding interest point data of the extensive object information of semantic processes system matches
Current location information, and using current location information as matching condition, screen the interest point data being consistent with matching condition.
It is understood that if the extensive object information extracted be it is multiple, need to match in point of interest library every
The corresponding interest point data of one extensive object information, obtains the matching result of each extensive object.
Step 108, one or more semantemes are obtained according to the contextual data of current statement lteral data and interest point data
Matching result data,
Specifically, processor brings the interest point data of current statement lteral data in the data set of contextual data into, obtain
To one or more semantic matches results.It is understood that if being obtained after carrying out clause extensive processing to phrase data
Fixed language information and extensive object information be it is multiple, then according to fixed language information and the obtained semantic matches of extensive object information
Result data is also multiple.Semantic matches result can be the matching result in semantic processes system itself, be also possible to semanteme
Processing system opens the matching result shown after other applications.
Step 109, when the number of semantic matches result data is multiple, the semantic matches result of highest priority is determined
Data;
Specifically, processor is first according to each semantic matches result data when semantic matches result data is multiple
Corresponding extensive object information determines sentence application field information corresponding with the extensive object information.Then, processor
It determines that each sentence application field information corresponds to the other precedence information of domain class, and determines priority from multiple fields classification
The highest field classification of information.The field classification of each highest level corresponds to a precedence information.Precedence information can be with
It is that business personnel is set as needed, is also possible to semantic processes system according to history sentence application field information and user's root
It is obtained according to the training of behavior command performed by sentence application field information.Finally, processor is according to the other priority of domain class
Information determines the semantic matches result data of highest priority.
Sentence application field information is understood that be independent application field, such as " life periphery " field, " joke event
Thing " field, " listening song " field, " health " field, " booking " field etc..Relative to the corresponding scene number of each fixed language information
According to each extensive object information also corresponds to a sentence application field information.For example, one " I wants to go to cinema " sentence number
According to, wherein fixed language information " I wants to go to " corresponding user wants to go to the contextual data in somewhere, extensive object information " film
The sentence application field information of institute " one corresponding " life periphery ".
In a specific example, " purchase film ticket " is that two different sentence application fields are believed with " going to the cinema "
Breath, corresponding field classification information are respectively " 2A " and " 2B ", and the field classification information of " 2A " represents " the A class second level ",
The field classification information of " 2B " represents " the B class second level ".Then semantic processes system queries are right to field classification information " 2A " institute
The highest level field classification information " 1A " answered, " 1A " representative " shopping " field, highest corresponding to field classification information " 2B "
Rank field classification information " 1B ", " 1B " representative " life periphery " field.Then semantic processes system determines that current area classification is believed
Breath is with Wei not " 1A " and " 1B ".The priority in " if shopping " field is higher than the priority in " life periphery " field, that is,
The priority of " 1A " is higher than " 1B ", then processor determines that the highest sentence application field information of precedence information is " 1A ",
It exactly " does shopping " field, through using semantic matches result data corresponding to sentence application field information " 1A " as highest priority
Semantic matches result data.
Step 110, the highest semantic matches result data of output priority.
Specifically, the processor in semantic processes system will be excellent when output priority highest semantic matches result data
The first highest semantic matches result data of grade is encapsulated into the end of the output queue in semantic processes system.Interrogator monitors output
The data of queue are inserted into, and the semantic matches result data at input rank end are obtained from output queue, that is, priority is most
High semantic matches result data, and export.
In some preferred embodiments, processor will record poll in the semanteme of match statement lteral data first
Device obtains the first time of the sentence lteral data at input rank end from input rank, and monitors system time, according to
One time and system time obtain the semantic processes time, and the semantic processes time can be understood as processor according to the language being calculated
The time of adopted processing system processing current statement lteral data.When processor monitors that the semantic processes time is greater than preset time
When, processor output feedback sentence.What preset time can be that user is set as needed allows to wait semantic matches result
Maximum time.When processor monitors that the semantic processes time is greater than preset time, the current language of declarative semantics processing system processing
The time of sentence lteral data has been more than the maximum time for allowing to wait semantic matches result, then for example " I needs for processor output
The feedback sentence of more information ", to overtime to user feedback current semantics matching process.
Natural language processing method provided in an embodiment of the present invention can be extracted by the extensive processing of clause to sentence
Fixation language and extensive object in clause can be associated with from the context when the fixed language of current statement or extensive object missing
Corresponding final matching results are exported to corresponding fixed language or extensive object, and according to priority, are determined by the method
The process of final semantic matches result is fast and accurately.
Professional should further appreciate that, described in conjunction with the examples disclosed in the embodiments of the present disclosure
Unit and algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, hard in order to clearly demonstrate
The interchangeability of part and software generally describes each exemplary composition and step according to function in the above description.
These functions are implemented in hardware or software actually, the specific application and design constraint depending on technical solution.
Professional technician can use different methods to achieve the described function each specific application, but this realization
It should not be considered as beyond the scope of the present invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can be executed with hardware, user terminal
Software module or the combination of the two implement.Software module can be placed in random access memory (RAM), memory, read-only storage
Device (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology neck
In any other form of storage medium well known in domain.
Above-described specific embodiment has carried out further the purpose of the present invention, technical scheme and beneficial effects
It is described in detail, it should be understood that being not intended to limit the present invention the foregoing is merely a specific embodiment of the invention
Protection scope, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of natural language processing method, which is characterized in that the described method includes:
Semantic processes system receives phrase data, obtains sentence lteral data;
The extensive processing of clause is carried out to the sentence lteral data according to syntax rule tree, is extracted in the sentence lteral data
Fixed language information and extensive object information;
Determine the fixation language information in the current statement lteral data or the extensive object in the current statement lteral data
Whether information is empty;
When the fixation language information in the current statement lteral data or the extensive object in the current statement lteral data are believed
When breath is not empty, according in the fixation language information and the current statement lteral data in the current statement lteral data
Extensive object information obtains one or more semantic matches result datas;
When the fixation language information in the current statement lteral data or the extensive object in the current statement lteral data are believed
When breath is empty, acquisition is associated with sentence lteral data with the current statement lteral data;
Semantic matches result data is obtained according to the association sentence lteral data and the current statement lteral data;
When the number of the semantic matches result data is multiple, believed according to the corresponding priority of semantic matches result data
Breath, determines the semantic matches result data of highest priority;
Export the semantic matches result data of the highest priority.
2. natural language processing method according to claim 1, which is characterized in that the phrase data includes sentence voice
Data and sentence lteral data;The semantic processes system receives phrase data, obtains sentence lteral data specifically:
The speech convertor of the semantic processes system receives the phrase data, to the sentence voice number in the phrase data
According to being identified, the sentence lteral data of the sentence voice data is obtained, and by the sentence text of the sentence voice data
Data are inserted into the end of the input rank of the semantic processes system;
The interrogator of the semantic processes system monitors the data insertion of the input rank, and institute is obtained from the input rank
The sentence lteral data for stating input rank end obtains the current statement lteral data.
3. natural language processing method according to claim 1, which is characterized in that it is described according to syntax rule tree to institute
Before predicate sentence lteral data carries out the extensive processing of clause, the method also includes:
History sentence lteral data is inputted the clause analytic modell analytical model to be trained, the clause analytic modell analytical model after being trained,
The extensive place of clause is carried out to the sentence lteral data according to the clause analytic modell analytical model after the training to semantic processes system
Reason.
4. natural language processing method according to claim 1, which is characterized in that described according to the current statement text
The extensive object information in fixation language information and the current statement lteral data in data obtains one or more semantic
With result data specifically:
The corresponding scene number of the sentence lteral data is determined according to the fixation sentence information in the current statement lteral data
According to;
Interest point data corresponding with the extensive object information in the current statement lteral data is matched in point of interest library;
Semantic matches result data is obtained according to the interest point data and the contextual data.
5. natural language processing method according to claim 1, which is characterized in that the association sentence lteral data includes
Upper sentence lteral data and next statement lteral data.
6. natural language processing method according to claim 5, which is characterized in that when in the current statement lteral data
Fixation language information when being empty, it is described that semanteme is obtained according to the association sentence lteral data and the current statement lteral data
Matching result data specifically:
The fixation language information in upper sentence lteral data is obtained, and is believed according to the fixation language in the upper sentence lteral data
Extensive object information in breath and the current statement lteral data obtains semantic matches result data.
7. natural language processing method according to claim 5, which is characterized in that when in the current statement lteral data
Extensive object information when being empty, it is described that language is obtained according to the association sentence lteral data and the current statement lteral data
Adopted matching result data specifically:
The supplement phrase data that user inputs according to boot statement data is received, and is extracted extensive in the supplement phrase data
Object information;
According to the fixed language information in the extensive object information and current statement lteral data in the supplement phrase data
Obtain semantic matches result data.
8. natural language processing method according to claim 7, which is characterized in that in the reception user according to leading question
Before the supplement phrase data of sentence data input, the method also includes:
Semantic processes system generates and exports the boot statement.
9. natural language processing method according to claim 1, which is characterized in that described according to semantic matches result data
Corresponding precedence information determines the semantic matches result data of highest priority specifically:
Determine that the multiple semantic matches result data institute is right according to the extensive object information in the current statement lteral data
The sentence application field information answered;
Determine that each sentence application field information corresponds to the other precedence information of domain class;
The semantic matches result data of highest priority is determined according to the other precedence information of the domain class.
10. natural language processing method according to claim 2, which is characterized in that the output highest priority
Semantic matches result data specifically:
The output queue highest semantic matches result data of the precedence information being encapsulated into the semantic processes system
End;
The interrogator monitors the data insertion of the output queue, and the input rank end is obtained from the output queue
Highest semantic matches result data, and export.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910231479.2A CN109960806A (en) | 2019-03-26 | 2019-03-26 | A kind of natural language processing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910231479.2A CN109960806A (en) | 2019-03-26 | 2019-03-26 | A kind of natural language processing method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109960806A true CN109960806A (en) | 2019-07-02 |
Family
ID=67025059
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910231479.2A Pending CN109960806A (en) | 2019-03-26 | 2019-03-26 | A kind of natural language processing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109960806A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111177338A (en) * | 2019-12-03 | 2020-05-19 | 北京博瑞彤芸科技股份有限公司 | Context-based multi-turn dialogue method |
CN112241630A (en) * | 2020-10-30 | 2021-01-19 | 深圳供电局有限公司 | Method and system for analyzing transformer variable-research standard vocabulary entry based on natural language processing |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104516949A (en) * | 2014-12-10 | 2015-04-15 | 北京百度网讯科技有限公司 | Webpage data processing method and apparatus, query processing method and question-answering system |
CN105468583A (en) * | 2015-12-09 | 2016-04-06 | 百度在线网络技术(北京)有限公司 | Entity relationship obtaining method and device |
CN106202270A (en) * | 2016-06-28 | 2016-12-07 | 广州幽联信息技术有限公司 | Interactive method based on natural language and device |
CN106326307A (en) * | 2015-06-30 | 2017-01-11 | 芋头科技(杭州)有限公司 | Language interaction method |
-
2019
- 2019-03-26 CN CN201910231479.2A patent/CN109960806A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104516949A (en) * | 2014-12-10 | 2015-04-15 | 北京百度网讯科技有限公司 | Webpage data processing method and apparatus, query processing method and question-answering system |
CN106326307A (en) * | 2015-06-30 | 2017-01-11 | 芋头科技(杭州)有限公司 | Language interaction method |
CN105468583A (en) * | 2015-12-09 | 2016-04-06 | 百度在线网络技术(北京)有限公司 | Entity relationship obtaining method and device |
CN106202270A (en) * | 2016-06-28 | 2016-12-07 | 广州幽联信息技术有限公司 | Interactive method based on natural language and device |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111177338A (en) * | 2019-12-03 | 2020-05-19 | 北京博瑞彤芸科技股份有限公司 | Context-based multi-turn dialogue method |
CN111177338B (en) * | 2019-12-03 | 2023-07-21 | 北京博瑞彤芸科技股份有限公司 | Context-based multi-round dialogue method |
CN112241630A (en) * | 2020-10-30 | 2021-01-19 | 深圳供电局有限公司 | Method and system for analyzing transformer variable-research standard vocabulary entry based on natural language processing |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11250842B2 (en) | Multi-dimensional parsing method and system for natural language processing | |
KR102577584B1 (en) | Method and apparatus for performing machine translation | |
US20180061408A1 (en) | Using paraphrase in accepting utterances in an automated assistant | |
US8548805B2 (en) | System and method of semi-supervised learning for spoken language understanding using semantic role labeling | |
JP5099953B2 (en) | Generation of unified task-dependent language model by information retrieval method | |
US20050154580A1 (en) | Automated grammar generator (AGG) | |
EP1647971A2 (en) | Apparatus and method for spoken language understanding by using semantic role labeling | |
EP0836144A2 (en) | Method and system for identifying and resolving commonly confused words in a natural language parser | |
US20020087310A1 (en) | Computer-implemented intelligent dialogue control method and system | |
JP2001519043A (en) | Text normalization using context-free grammar | |
JP2000353161A (en) | Method and device for controlling style in generation of natural language | |
JP2002334071A (en) | Text compression intelligent as language | |
KR102041621B1 (en) | System for providing artificial intelligence based dialogue type corpus analyze service, and building method therefor | |
WO2023040493A1 (en) | Event detection | |
CN110008471A (en) | A kind of intelligent semantic matching process based on phonetic conversion | |
CN109960806A (en) | A kind of natural language processing method | |
CN109992651B (en) | Automatic identification and extraction method for problem target features | |
CN109960807A (en) | A kind of intelligent semantic matching process based on context relation | |
CN109960805A (en) | A kind of intelligent semantic matching process divided based on semantic domain | |
CN115132182B (en) | Data identification method, device, equipment and readable storage medium | |
CN109977405A (en) | A kind of intelligent semantic matching process | |
CN115620726A (en) | Voice text generation method, and training method and device of voice text generation model | |
US11361761B2 (en) | Pattern-based statement attribution | |
JP3691773B2 (en) | Sentence analysis method and sentence analysis apparatus capable of using the method | |
JP2018181259A (en) | Dialogue rule collation device, dialogue device, dialogue rule collation method, dialogue method, dialogue rule collation program, and dialogue program |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190702 |
|
RJ01 | Rejection of invention patent application after publication |