CN106339366A - Method and device for requirement identification based on artificial intelligence (AI) - Google Patents
Method and device for requirement identification based on artificial intelligence (AI) Download PDFInfo
- Publication number
- CN106339366A CN106339366A CN201610643685.0A CN201610643685A CN106339366A CN 106339366 A CN106339366 A CN 106339366A CN 201610643685 A CN201610643685 A CN 201610643685A CN 106339366 A CN106339366 A CN 106339366A
- Authority
- CN
- China
- Prior art keywords
- demand
- information
- syntax
- phrase
- compactness
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- 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
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 invention provides a method and device for requirement identification based on AI. The method for requirement identification based on AI comprises the following steps: obtaining requirement information; dividing the requirement information into words to form multiple phrases; obtaining syntactic tightness information among multiple phrases; carrying out the requirement identification according to a default dependency syntactic tree of the syntactic tightness information and the requirement information. By adopting the method for the requirement identification based on AI disclosed by the invention, the accuracy and identification efficiency of the requirement identification can be effectively improved.
Description
Technical field
The present invention relates to technical field of information processing, particularly to a kind of demand based on artificial intelligence know method for distinguishing and
Device.
Background technology
Artificial intelligence (artificial intelligence), english abbreviation is ai.It is research, be developed for simulation,
Extend and extend the theory of intelligence of people, new science of technology of method, technology and application system.Artificial intelligence is to calculate
One branch of machine science, it attempts to understand essence of intelligence, and produce a kind of new can be in the way of human intelligence be similar
The intelligent machine made a response, the research in this field includes robot, language identification, image recognition, natural language processing and specially
Family's system etc..
Syntactic analysis is a key issue of natural language processing field in artificial intelligence study.In order to determine sentence
Dependence between vocabulary in syntactic structure or sentence.Therefore, interdependent syntactic analysis is that the demand to user is identified
One of main method of analysis.Interdependent syntactic analysis refers to, by setting up modification level and modified relationship between word, make sentence
Constitute a syntactic structure tree.Mainly determine user input by inquiring about the plaintext collocations dictionary pre-building at present
Syntax dependence between vocabulary in content.
However, the syntactic structure being not covered with for plaintext collocations dictionary, then cannot determine dependence, because
This, in current interdependent syntactic analysis, for syntactic structure coverage rate is relatively low, generalization ability is poor, and analyze relied on bright
Cliction language dictionary of collocations scale is big, it is big to take up room, application inconvenience, and then have impact on the accuracy of demand identification, recognition efficiency.
Content of the invention
It is contemplated that at least solving above-mentioned technical problem to a certain extent.
For this reason, the first of the present invention purpose is that proposing a kind of demand based on artificial intelligence knows method for distinguishing, can
Effectively improve accuracy and the recognition efficiency of demand identification.
Second object of the present invention is to propose a kind of device of the identification of the demand based on artificial intelligence.
For reaching above-mentioned purpose, embodiment proposes a kind of identification of the demand based on artificial intelligence according to a first aspect of the present invention
Method, comprise the following steps: obtain demand information;Described demand information is carried out with participle to form multiple phrases;Obtain institute
State the syntax compactness information between multiple phrases;Default interdependent syntax according to described syntax compactness information and demand information
Tree carries out demand identification.
The demand based on artificial intelligence of the embodiment of the present invention knows method for distinguishing, can obtain the demand information of user input,
And carry out participle and form multiple phrases, obtain the syntax compactness information between multiple phrases, and according to syntax compactness information
Carry out demand identification with the default interdependent syntax tree of demand information, demand identification during combine syntax compactness information with
Interdependent syntax tree can effectively improve accuracy and the recognition efficiency of demand identification, can accurately extract, the need of identifying user
Ask, and then can more targetedly provide the user more preferable service, lift Consumer's Experience.
Second aspect present invention embodiment proposes a kind of device of the identification of the demand based on artificial intelligence, comprising: first
Acquisition module, for obtaining demand information;Word-dividing mode, for carrying out participle to form multiple phrases to described demand information;
Second acquisition module, for obtaining the syntax compactness information between the plurality of phrase;Demand identification module, for according to institute
State syntax compactness information and the default interdependent syntax tree of demand information carries out demand identification.
The device of the demand identification based on artificial intelligence of the embodiment of the present invention, can obtain the demand information of user input,
And carry out participle and form multiple phrases, obtain the syntax compactness information between multiple phrases, and according to syntax compactness information
Carry out demand identification with the default interdependent syntax tree of demand information, demand identification during combine syntax compactness information with
Interdependent syntax tree can effectively improve accuracy and the recognition efficiency of demand identification, can accurately extract, the need of identifying user
Ask, and then can more targetedly provide the user more preferable service, lift Consumer's Experience.
The additional aspect of the present invention and advantage will be set forth in part in the description, and partly will become from the following description
Obtain substantially, or recognized by the practice of the present invention.
Brief description
The above-mentioned and/or additional aspect of the present invention and advantage will become from reference to the description to embodiment for the accompanying drawings below
Substantially and easy to understand, wherein:
Fig. 1 is the flow chart that the demand based on artificial intelligence knows method for distinguishing according to one embodiment of the invention;
Fig. 2 is the schematic diagram of the interdependent syntax tree according to one embodiment of the invention;
Fig. 3 is the schematic diagram according to one embodiment of the invention based on the demand recognition methodss of artificial intelligence;
Fig. 4 is the flow chart that the demand based on artificial intelligence knows method for distinguishing according to another embodiment of the present invention;
Fig. 5 is the schematic diagram of the network model of positive example according to one embodiment of the invention and negative example Similarity Measure;
Fig. 6 is the method schematic diagram of the demand identification according to one specific embodiment of the present invention based on artificial intelligence;
Fig. 7 a is the interface schematic diagram of the depth question answering system according to one embodiment of the invention;
Fig. 7 b is the interface schematic diagram of the depth question answering system according to another embodiment of the present invention;
Fig. 8 is the structural representation of the device of the demand identification according to one embodiment of the invention based on artificial intelligence;
Fig. 9 is the structural representation of the device of the demand identification according to another embodiment of the present invention based on artificial intelligence
Figure.
Specific embodiment
Embodiments of the invention are described below in detail, the example of described embodiment is shown in the drawings, wherein from start to finish
The element that same or similar label represents same or similar element or has same or like function.Below with reference to attached
The embodiment of figure description is exemplary, is only used for explaining the present invention, and is not considered as limiting the invention.
In describing the invention it is to be understood that term " multiple " refers to two or more;Term " first ",
" second " is only used for describing purpose, and it is not intended that indicating or hint relative importance.
Below with reference to the accompanying drawings the method and apparatus describing the demand identification based on artificial intelligence according to embodiments of the present invention.
The present invention proposes a kind of demand based on artificial intelligence and knows method for distinguishing, and the demand based on artificial intelligence identifies
Method, comprises the following steps: obtains demand information;Described demand information is carried out with participle to form multiple phrases;Obtain described
Syntax compactness information between multiple phrases;Default interdependent sentence according to described syntax compactness information and described demand information
Method tree carries out demand identification.
Fig. 1 is the flow chart that the demand based on artificial intelligence knows method for distinguishing according to one embodiment of the invention.
As shown in figure 1, the demand based on artificial intelligence according to embodiments of the present invention knows method for distinguishing, comprising:
S101, obtains demand information.
Wherein, the information such as demand information can input for user under each demand scene word, voice, image.
For example, under search scene, the information for search of input in search box;Or in question and answer scene
Under, the question information of user input or answer information.
S102, carries out participle to form multiple phrases to described demand information.
In an embodiment of the present invention, if demand information is text message, directly participle can be carried out to demand information,
Form multiple phrases.If demand information is non-textual information, non-textual information then can be carried out by the such as information such as voice, image
It is identified as text message, and the text that identification is obtained carries out participle, forms multiple phrases.
For example, the demand information for user input " orders the aircraft in Shanghai of flying to a tomorrow from Beijing to me
Ticket ", can get phrase after participle: " to ", " I ", " ordering ", " one ", " tomorrow ", " from ", " Beijing ", " flying to ", " Shanghai ",
" ", " plane ticket ".
S103, obtains the syntax compactness information between the plurality of phrase.
Wherein, syntax compactness information refers to that two phrases constitute the tolerance of the probability of certain syntactic structure.
In another embodiment of the present invention, can be obtained in the plurality of phrase by query syntax compactness model and appoint
Syntax compactness information between two words of meaning.
Wherein, syntax compactness model is by depth learning technology, using neutral net, syntax compactness is learnt be
One computable model (process specifically setting up syntax compactness model can be found in Fig. 4 and embodiment illustrated in fig. 5), this model
In store the network structure of syntax compactness.Therefore, can be determined arbitrarily by the network structure in this syntax compactness model
Collocations syntax compactness information.
In one embodiment of the invention, for the sentence completing participle, can be according to the syntax compactness of foundation
Model calculates the syntax compactness information between any two.Further, also dependent in the syntax compactness information obtaining
Maximum obtained syntax compactness information is normalized, after obtaining the normalized between each word
Syntax compactness information, in order to the comparison of continued syntactical compactness information.
S104, carries out demand identification according to the default interdependent syntax tree of described syntax compactness information and demand information.
Wherein, the default interdependent syntax tree of demand information can be according to the dependency structure of sentence, modify level and modification
Relation and the tree structure set up.
In one embodiment of the invention, the interdependent syntax tree of demand information can be beforehand through existing or not
Interdependent parser may occur in which is analyzed to the demand information of user obtaining.For example, stanford- can be passed through
Parser (a kind of syntax parsing instrument released by Stanford University's natural language research group), ltp are (big by Harbin industry
Learn the language technology platform of social computing and the research and development of Research into information retrieval center) it is analyzed obtaining interdependent syntax tree.
In another embodiment of the present invention, also can be by sentence in the existing method training interdependent syntactic analysis model
Method compactness information is as features training interdependent syntactic analysis model.Specifically, for the sample requirement information sentence of participle,
The syntax compactness information of a word and other all words can be calculated, and when building interdependent syntactic analysis model, extract window
The compactness information of interior all possible syntactic structure, as feature, sets up interdependent syntactic analysis model.And then, can be previously according to
Thus obtained interdependent syntactic analysis model carries out syntactic analysis to demand information, obtains the interdependent syntax tree of demand information.
Due to, during training interdependent syntactic analysis model, increased syntax compactness feature, therefore, thus obtaining
Interdependent syntax tree more accurate.
Interdependent syntax tree includes the dependency structure of phrase and dependence in sentence.
For example, dependency structure may include but be not limited to: core (head, hed), serial verb construction (verb-verb,
Vv), relational structure (conjunctive, cnj), voice structure (mood-tense, mt), absolute construction (independent
Structure, is), verbal endocentric phrase (adverbial, adv), structure of complementation (complement, cmp), " " word structure de,
" " word structure di, " obtaining " word structure dei, " " word structure ba, " quilt " word structure bei, independent clause ic (independent
) and interdependent subordinate sentence dc (dependent clause) etc. clause.
Dependence may include but be not limited to: relation (attribute, att) in fixed, dynamic guest's relation (verb-object,
Vob), subject-predicate relation (subject-verb, sbv), quantitative relation (quantity, qun), coordination (coordinate,
Coo), apposition (appositive, app), front additional relationships (left adjunct, lad), rear additional relationships (right
Adjunct, rad), dynamic guest's relation (verb-object, vob), guest's Jie relation (preposition-object, pob) and ratio
Plan relation (similarity, sim).
For a sentence and its corresponding interdependent syntax tree, each word in sentence constitutes positive example with his qualifier,
Constitute counter-example with other words in addition to this qualifier in sentence.For example, for sentence " becoming moral model ", its correspondence
Interdependent syntax tree as shown in a in Fig. 2, wherein, between " becoming " and " model ", be guest's relation vob, " moral " and " model "
Between be surely middle relation att.In this interdependent syntax tree a, the positive and negative example of " referred to as " can be as shown in b in Fig. 2, wherein, " becoming "
Constitute positive example with " model ", " becoming " and " model " constitute negative example.
In one embodiment of the invention, step s104 may particularly include: is determined according to described syntax compactness information
Phrase as host node and the phrase as child node in the plurality of phrase;According to the described phrase as host node and work
Phrase for child node and described default interdependent syntax tree carry out demand identification.
Specifically, syntax and other phrases between can be selected from multiple phrases according to syntax compactness information tight
Degree highest phrase, and as host node, using other phrases as child node.After determining host node and child node, can enter
One step according to demand in the default syntax dependency tree of information the dependency structure between phrase and dependence determine and host node between
There is each child node of dependence, and the dependency structure between each child node and dependence.And then, can be according to master
Dependency structure between node and child node, between child node and child node and dependence identify the demand of user.
In an embodiment of the present invention, described demand identification may include demand type identification and/or demand limits identification.Its
In, demand type identification refers to the type of user's request is identified, for example, booking demand, trip requirements etc..Demand limits
Identification refers to the qualificationss of user's request are identified, for example, the qualificationss such as time, place.
For example, as shown in figure 3, the demand information for user input " orders a tomorrow from flying in Beijing to me
Sea plane ticket ", can determine that host node is " ordering " according to syntax compactness information, child node be " to ", " I ", " " one ",
" tomorrow ", " from ", " Beijing ", " flying to ", " Shanghai ", " ", " plane ticket ".And then, can get the interdependent sentence shown in 31 in Fig. 3
Method tree, and determine the dependency structure between each child node and host node and between different child node and dependence.Then,
Can identify that demand type is " ordering plane ticket " according to interdependent syntax tree 31, demand is defined to " departure place: Beijing;Reach ground: on
Sea;Time: tomorrow ".
The demand of the embodiment of the present invention knows method for distinguishing, can obtain the demand information of user input, and carry out participle and is formed
Multiple phrases, obtain the syntax compactness information between multiple phrases, and pre- according to syntax compactness information and demand information
If interdependent syntax tree carries out demand identification, combining syntax compactness information during demand identification can with interdependent syntax tree
Effectively improve accuracy and the recognition efficiency of demand identification, can accurately extract, the demand of identifying user, and then can more have pin
More preferable service is provided the user to property, lifts Consumer's Experience.
In another embodiment of the present invention, may also include the step setting up syntax compactness model.
Fig. 4 is the flow chart that the demand based on artificial intelligence knows method for distinguishing according to another embodiment of the present invention.
As shown in figure 4, setting up syntax compactness model, it may include following steps s401-s405.
S401, obtains sample requirement information.
In an embodiment of the present invention, sample requirement information can be the demand information of the history input of the user from magnanimity
The large-scale sample requirement information of middle selection, or large-scale sample requirement information is produced by machine simulation, or from history
Selected part in the demand information of input, produced part by machine simulation and with reference to obtaining large-scale sample requirement information.
S402, carries out participle to form multiple sample phrases to described sample requirement information.
S403, obtains the target sample phrase in described sample phrase, and obtains the plurality of sample phrase and described mesh
Positive example relation between this phrase of standard specimen or negative example relation.
Specifically, can successively using each sample phrase in multiple sample phrases as target sample phrase, to obtain each
Positive example relation between sample phrase and other sample phrases or negative example relation.
Wherein, for the sample phrase outside target sample phrase, if the modification phrase of target sample phrase, then with
Target sample phrase is positive example relation, is otherwise negative example relation with target sample phrase.
That is, each sample phrase constitutes positive example relation with its modification phrase, constitute negative example relation with other words.
In one embodiment of the invention, data analysis can be passed through, that is, using interdependent parser, point
Analyse large-scale sample requirement information, and generate the corresponding automatic treebank of these sample requirement information.And then can be from automatic treebank
Extract training data, for training syntax compactness information model.
Wherein, automatic treebank includes and each sample requirement information and its corresponding interdependent syntax tree.Interdependent syntax tree
Include the positive and negative example relation between each target phrase.For example, for sentence " becoming moral model ", its corresponding according to
Deposit syntax tree as shown in a in Fig. 2, wherein, between " becoming " and " model ", be guest's relation vob, between " moral " and " model "
For surely middle relation att.In this interdependent syntax tree a, the positive and negative example of " referred to as " can be as shown in b in Fig. 2, wherein, " becoming " and " pattern
Mould " constitutes positive example, and " becoming " and " model " constitute negative example.
S404, carries out deep learning according to described positive example relation or negative example relation, to set up syntax compactness model.
In one embodiment of the invention, the occurrence number of the positive and negative example of each phrase is ranked up learning, finally
The high word score of syntax compactness is made to be higher than the low word score of syntax compactness.Wherein, the similarity of two words, you can embody
Incidence relation between two words, therefore, can be used as the syntax compactness information of two words.That is, between two phrases
Syntax compactness can be weighed using the similarity of two phrases.
Specifically, first can be based on each positive example relation obtaining in initial neural network model respectively calculation procedure s403
With the similarity of phrase in negative example relation, if there is the phrase of positive example relation similarity be less than negative example relation phrase phase
Like spending, then adjust above-mentioned initial neural network model, and recalculate each positive example using the neural network model after adjustment and close
The similarity of phrase in system and negative example relation, and be compared.Repeat the above steps, until make the phase of phrase in positive example relation
Like degree higher than the similarity of phrase in negative example relation, obtain final neural network model, and as syntax compactness model.
The positive example similarity of described target sample phrase and other sample phrases can be calculated based on the network model shown in Fig. 5
Or negative example similarity.As shown in figure 5, can be according to the positive example relation between phrase and negative example relation input core word and qualifier, so
Afterwards the phrase of input is become term vector, then (network transformation) is converted by a Ge Quan UNICOM and useful information extraction is gone out
Come, finally calculate the similarity of positive example and negative example.
Wherein, the present invention does not limit to the method calculating similarity.For example, the table such as Euclidean distance, cosine value can be passed through
Show similarity.
So that similarity is as cosine similarity as a example illustrate below.When calculating the similarity of positive example, can be by calculating
The cosine of corresponding term vector a1 and b1 of two words of positive example is worth to the similarity of two words in positive example.
Further, the conversion of full UNICOM can be carried out before calculating similarity.For example, two for above-mentioned positive example
Corresponding term vector a1 and b1 of individual word, can be multiplied by a transformation matrix c respectively, complete the conversion of full UNICOM, a1 is become and turns to a2,
B1 is become and turns to b2.Then pass through to calculate the cosine value of vectorial a2 and b2, and the similarity as a1 and b1.
Thus, by embodiments of the invention, by deep learning, syntax compactness is learnt as a computable sentence
Method compactness model, model storage is network structure, and for arbitrary collocations, all cocoas pass through this syntax compactness mould
Type calculates the syntax compactness information between any two word, is not limited solely to the syntax in traditional plaintext collocations dictionary
Structure, coverage is wider.Additionally, the size of the syntax compactness model based on deep learning is solely dependent upon the size of vocabulary,
Much smaller than the scale of plaintext collocations dictionary, reduce space hold.
Embodiments of the invention can be applicable in the application scenarios such as intelligent machine question and answer, intelligent search.Fig. 6 is according to this
The method schematic diagram of the demand identification based on artificial intelligence of a bright specific embodiment.With reference to the reality to the present invention for the Fig. 6
Apply application in intelligent machine depth question answering system for the example to illustrate.
As shown in fig. 6, in the application in intelligent machine question answering system, it may include data genaration, network model are set up, spy
Levy the modeling processes such as application, and demand identification is carried out in depth question answering system according to the model set up, and according to
The application process that recognition result scans for, shows.
Wherein, data genaration, network model set up process and the process phase setting up syntax compactness model shown in Fig. 6
With.After setting up syntax compactness model, syntax compactness information can be applied in syntactic analysis as feature, training is interdependent
Syntactic analysis model.That is, can will be tight for the syntax of the deep learning of phrase in each demand information sentence in training sample
Density information is as compactness feature, and is based on compactness features training interdependent syntactic analysis model.
And then, question answering system can carry out interdependent syntactic analysis according to the problem being directed to user input with the model set up, and
Demand identification is carried out according to analysis result, and is precisely met according to recognition result.Mainly there are demand type identification, demand limit
Fixed identification and result search three steps of polymerization.Specifically, when user inputs problem (i.e. demand information) in question answering system,
Question answering system can generate the corresponding interdependent syntax tree of this problem according to the interdependent syntactic analysis model of foundation.Due to the present embodiment
In interdependent syntactic analysis model increased syntax compactness feature in the training process, therefore, the interdependent sentence of this problem of generation
Method tree also has the syntax compactness information of each phrase in problem on the basis of traditional interdependent syntax tree.And then, can basis
Dependency structure between phrase, dependence and syntax compactness information in interdependent syntax tree, carry out demand type identification and need
Ask restriction identification, and scan for being polymerized according to recognition result, Search Results are showed in question answering system user.
For example, for the demand type in such as Fig. 3, being identified according to interdependent syntax tree 31: " ordering plane ticket " and need
Ask restriction: " departure place: Beijing;Reach ground: Shanghai;Time: tomorrow ", scan for result polymerization, obtain Crestor plane ticket
Website, and booking demands ofdifferent classes condition is supplied to user's (as shown in Figure 7a) as answer.When user clicks on this answer, such as Fig. 7 b
Shown, corresponding website can be entered, and the information such as departure place, destination, time are automatically filled according to the user's request identifying, with
Search for for user and confirm.
Thus, in depth question answering system, can be according to the interdependent sentence of the addition syntax compactness information characteristics pre-building
Method model, deep identifying user demand, and provide accordingly as a result, it is possible to targetedly accurately meet user's request.
Corresponding with the embodiment of the method for the above-mentioned demand based on artificial intelligence identification, the present invention also proposes one kind and is based on people
The device of the demand identification of work intelligence.
A kind of device of the demand identification based on artificial intelligence, comprising: the first acquisition module, for obtaining demand information;
Word-dividing mode, for carrying out participle to form multiple phrases to described demand information;Second acquisition module is described many for obtaining
Syntax compactness information between individual phrase;Demand identification module, for according to described syntax compactness information and demand information
Default interdependent syntax tree carry out demand identification.
Fig. 8 is the structural representation of the device of the demand identification according to one embodiment of the invention based on artificial intelligence.
As shown in figure 8, the device of the demand identification based on artificial intelligence according to embodiments of the present invention, comprising: first obtains
Delivery block 10, word-dividing mode 20, the second acquisition module 30 and demand identification module 40.
Specifically, the first acquisition module 10 is used for obtaining demand information.
Wherein, the information such as demand information can input for user under each demand scene word, voice, image.
For example, under search scene, the information for search of input in search box;Or in question and answer scene
Under, the question information of user input or answer information.
Word-dividing mode 20 is used for carrying out participle to form multiple phrases to described demand information.
In an embodiment of the present invention, if demand information is text message, directly participle can be carried out to demand information,
Form multiple phrases.If demand information is non-textual information, non-textual information then can be carried out by the such as information such as voice, image
It is identified as text message, and the text that identification is obtained carries out participle, forms multiple phrases.
For example, the demand information for user input " orders the aircraft in Shanghai of flying to a tomorrow from Beijing to me
Ticket ", can get phrase after participle: " to ", " I ", " ordering ", " one ", " tomorrow ", " from ", " Beijing ", " flying to ", " Shanghai ",
" ", " plane ticket ".
Second acquisition module 30 is used for obtaining the syntax compactness information between the plurality of phrase.
Wherein, syntax compactness information refers to that two phrases constitute the tolerance of the probability of certain syntactic structure.
In another embodiment of the present invention, the second acquisition module 30 can be used for query syntax compactness model acquisition institute
State the syntax compactness information between any two word in multiple phrases.
Wherein, syntax compactness model is by depth learning technology, using neutral net, syntax compactness is learnt be
One computable model (process specifically setting up syntax compactness model can be found in Fig. 4 and embodiment illustrated in fig. 5), this model
In store the network structure of syntax compactness.Therefore, can be determined arbitrarily by the network structure in this syntax compactness model
Collocations syntax compactness information.
In one embodiment of the invention, for the sentence completing participle, the second acquisition module 30 can be according to built
Vertical syntax compactness model calculates the syntax compactness information between any two.Further, the second acquisition module 30
According to the maximum in the syntax compactness information obtaining, obtained syntax compactness information can be normalized, obtain
Syntax compactness information after normalized between each word, in order to the comparison of continued syntactical compactness information.
Demand identification module 40 is used for being entered according to the default interdependent syntax tree of described syntax compactness information and demand information
Row demand identifies.
Wherein, the default interdependent syntax tree of demand information can be according to the dependency structure of sentence, modify level and modification
Relation and the tree structure set up.
In one embodiment of the invention, the interdependent syntax tree of demand information can be beforehand through existing or not
Interdependent parser may occur in which is analyzed to the demand information of user obtaining.For example, stanford- can be passed through
Parser (a kind of syntax parsing instrument released by Stanford University's natural language research group), ltp are (big by Harbin industry
Learn the language technology platform of social computing and the research and development of Research into information retrieval center) it is analyzed obtaining interdependent syntax tree.
In another embodiment of the present invention, also can be by sentence in the existing method training interdependent syntactic analysis model
Method compactness information is as features training interdependent syntactic analysis model.Specifically, for the sample requirement information sentence of participle,
The syntax compactness information of a word and other all words can be calculated, and when building interdependent syntactic analysis model, extract window
The compactness information of interior all possible syntactic structure, as feature, sets up interdependent syntactic analysis model.And then, can be previously according to
Thus obtained interdependent syntactic analysis model carries out syntactic analysis to demand information, obtains the interdependent syntax tree of demand information.
Due to, during training interdependent syntactic analysis model, increased syntax compactness feature, therefore, thus obtaining
Interdependent syntax tree more accurate.
Interdependent syntax tree includes the dependency structure of phrase and dependence in sentence.
For example, dependency structure may include but be not limited to: core (head, hed), serial verb construction (verb-verb,
Vv), relational structure (conjunctive, cnj), voice structure (mood-tense, mt), absolute construction (independent
Structure, is), verbal endocentric phrase (adverbial, adv), structure of complementation (complement, cmp), " " word structure de,
" " word structure di, " obtaining " word structure dei, " " word structure ba, " quilt " word structure bei, independent clause ic (independent
) and interdependent subordinate sentence dc (dependent clause) etc. clause.
Dependence may include but be not limited to: relation (attribute, att) in fixed, dynamic guest's relation (verb-object,
Vob), subject-predicate relation (subject-verb, sbv), quantitative relation (quantity, qun), coordination (coordinate,
Coo), apposition (appositive, app), front additional relationships (left adjunct, lad), rear additional relationships (right
Adjunct, rad), dynamic guest's relation (verb-object, vob), guest's Jie relation (preposition-object, pob) and ratio
Plan relation (similarity, sim).
For a sentence and its corresponding interdependent syntax tree, each word in sentence constitutes positive example with his qualifier,
Constitute counter-example with other words in addition to this qualifier in sentence.For example, for sentence " becoming moral model ", its correspondence
Interdependent syntax tree as shown in a in Fig. 2, wherein, between " becoming " and " model ", be guest's relation vob, " moral " and " model "
Between be surely middle relation att.In this interdependent syntax tree a, the positive and negative example of " referred to as " can be as shown in b in Fig. 2, wherein, " becoming "
Constitute positive example with " model ", " becoming " and " model " constitute negative example.
In one embodiment of the invention, demand identification module 40 can be used for: true according to described syntax compactness information
Phrase as host node and the phrase as child node in fixed the plurality of phrase;According to the described phrase as host node and
The default interdependent syntax tree of the phrase as child node and described demand information carries out demand identification.
Specifically, demand identification module 40 can be selected and other phrases from multiple phrases according to syntax compactness information
Between syntax compactness highest phrase, and as host node, using other phrases as child node.Determining host node and son
After node, demand identification module 40 can the dependency structure between phrase in the default syntax dependency tree of information according to demand further
With dependence determines each child node that there is dependence and host node between, and the interdependent knot between each child node
Structure and dependence.And then, demand identification module 40 can be according between host node and child node, between child node and child node
Dependency structure and dependence identify the demand of user.
In an embodiment of the present invention, described demand identification may include demand type identification and/or demand limits identification.Its
In, demand type identification refers to the type of user's request is identified, for example, booking demand, trip requirements etc..Demand limits
Identification refers to the qualificationss of user's request are identified, for example, the qualificationss such as time, place.
For example, as shown in figure 3, the demand information for user input " orders a tomorrow from flying in Beijing to me
Sea plane ticket ", can determine that host node is " ordering " according to syntax compactness information, child node be " to ", " I ", " " one ",
" tomorrow ", " from ", " Beijing ", " flying to ", " Shanghai ", " ", " plane ticket ".And then, can get the interdependent sentence shown in 31 in Fig. 3
Method tree, and determine the dependency structure between each child node and host node and between different child node and dependence.Then,
Can identify that demand type is " ordering plane ticket " according to interdependent syntax tree 31, demand is defined to " departure place: Beijing;Reach ground: on
Sea;Time: tomorrow ".
The device of the demand identification of the embodiment of the present invention, can obtain the demand information of user input, and carry out participle and formed
Multiple phrases, obtain the syntax compactness information between multiple phrases, and pre- according to syntax compactness information and demand information
If interdependent syntax tree carries out demand identification, combining syntax compactness information during demand identification can with interdependent syntax tree
Effectively improve accuracy and the recognition efficiency of demand identification, can accurately extract, the demand of identifying user, and then can more have pin
More preferable service is provided the user to property, lifts Consumer's Experience.
Fig. 9 is the structural representation of the device of the demand identification according to another embodiment of the present invention based on artificial intelligence
Figure.
As shown in figure 9, the device of the demand identification based on artificial intelligence according to embodiments of the present invention, shown in Fig. 8
On the basis of also can further include: set up module 50.
Set up module 50 for setting up described syntax compactness model by following steps:
Obtain sample requirement information;
Described sample requirement information is carried out with participle to form multiple sample phrases;
Obtain the target sample phrase in described sample phrase, and obtain the plurality of sample phrase and described target sample
Positive example relation between phrase or negative example relation;
Deep learning is carried out according to described positive example relation or negative example relation, to set up syntax compactness model.
Wherein, specific implementation can refer to embodiment illustrated in fig. 4, will not be described here.
Syntax compactness is learnt can calculate for one by the device of the demand identification of the embodiment of the present invention by deep learning
Syntax compactness model, model storage be network structure, for arbitrary collocations, all can calculate its syntax compactness
Information, and then can be not limited solely to by the syntax compactness information between this syntax compactness model calculating any two word
Syntactic structure in traditional plaintext collocations dictionary, coverage is wider.Additionally, the syntax compactness based on deep learning
The size of model is solely dependent upon the size of vocabulary, much smaller than the scale of plaintext collocations dictionary, reduces space hold.
It should be appreciated that each several part of the present invention can be realized with hardware, software, firmware or combinations thereof.Above-mentioned
In embodiment, the software that multiple steps or method can be executed in memory and by suitable instruction execution system with storage
Or firmware is realizing.For example, if realized with hardware, and the same in another embodiment, can use well known in the art under
Any one of row technology or their combination are realizing: have the logic gates for data signal is realized with logic function
Discrete logic, there is the special IC of suitable combinational logic gate circuit, programmable gate array (pga), scene
Programmable gate array (fpga) etc..
Those skilled in the art are appreciated that to realize all or part step that above-described embodiment method carries
Suddenly the program that can be by completes come the hardware to instruct correlation, and described program can be stored in a kind of computer-readable storage medium
In matter, this program upon execution, including one or a combination set of the step of embodiment of the method.
Although embodiments of the invention have been shown and described above it is to be understood that above-described embodiment is example
Property it is impossible to be interpreted as limitation of the present invention, those of ordinary skill in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, changes, replacing and modification.
Claims (10)
1. a kind of demand based on artificial intelligence knows method for distinguishing it is characterised in that comprising the following steps:
Obtain demand information;
Described demand information is carried out with participle to form multiple phrases;
Obtain the syntax compactness information between the plurality of phrase;
Demand identification is carried out according to the default interdependent syntax tree of described syntax compactness information and described demand information.
2. as claimed in claim 1 demand is carried out according to customer problem know method for distinguishing it is characterised in that described in described basis
The default interdependent syntax tree of syntax compactness information and described demand information carries out demand identification and specifically includes:
Phrase as host node and the word as child node in the plurality of phrase are determined according to described syntax compactness information
Group;
Default interdependent syntax according to the described phrase as host node and the phrase as child node and described demand information
Tree carries out demand identification.
3. as claimed in claim 1 demand is carried out according to customer problem and know method for distinguishing it is characterised in that described demand identifies
Limit identification including demand type identification and/or demand.
4. as claimed in claim 1 demand is carried out according to customer problem know method for distinguishing it is characterised in that described in described acquisition
Syntax compactness information between multiple phrases specifically includes:
Query syntax compactness model obtains the syntax compactness information between the plurality of phrase.
5. as claimed in claim 4 demand is carried out according to customer problem and know method for distinguishing it is characterised in that described syntax is tight
Degree model is set up by following steps:
Obtain sample requirement information;
Described sample requirement information is carried out with participle to form multiple sample phrases;
Obtain the target sample phrase in described sample phrase, and obtain the plurality of sample phrase and described target sample phrase
Between positive example relation or negative example relation;
Deep learning is carried out according to described positive example relation or negative example relation, to set up described syntax compactness model.
6. a kind of device of the demand identification based on artificial intelligence is it is characterised in that include:
First acquisition module, for obtaining demand information;
Word-dividing mode, for carrying out participle to form multiple phrases to described demand information;
Second acquisition module, for obtaining the syntax compactness information between the plurality of phrase;
Demand identification module, for carrying out according to the default interdependent syntax tree of described syntax compactness information and described demand information
Demand identifies.
7. as claimed in claim 6 the device of demand identification is carried out it is characterised in that described demand identifies according to customer problem
Module is used for:
Phrase as host node and the word as child node in the plurality of phrase are determined according to described syntax compactness information
Group;
Default interdependent syntax according to the described phrase as host node and the phrase as child node and described demand information
Tree carries out demand identification.
8. as claimed in claim 6 the device of demand identification is carried out it is characterised in that described demand identifies according to customer problem
Limit identification including demand type identification and/or demand.
9. as claimed in claim 6 the device of demand identification is carried out it is characterised in that described second obtains according to customer problem
Module is used for:
Query syntax compactness model obtains the syntax compactness information between the plurality of phrase.
10. as claimed in claim 9 the device of demand identification is carried out it is characterised in that also including according to customer problem:
Set up module, described module of setting up is for setting up described syntax compactness model by following steps:
Obtain sample requirement information;
Described sample requirement information is carried out with participle to form multiple sample phrases;
Obtain the target sample phrase in described sample phrase, and obtain the plurality of sample phrase and described target sample phrase
Between positive example relation or negative example relation;
Deep learning is carried out according to described positive example relation or negative example relation, to set up described syntax compactness model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610643685.0A CN106339366B (en) | 2016-08-08 | 2016-08-08 | The method and apparatus of demand identification based on artificial intelligence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610643685.0A CN106339366B (en) | 2016-08-08 | 2016-08-08 | The method and apparatus of demand identification based on artificial intelligence |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106339366A true CN106339366A (en) | 2017-01-18 |
CN106339366B CN106339366B (en) | 2019-05-31 |
Family
ID=57824363
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610643685.0A Active CN106339366B (en) | 2016-08-08 | 2016-08-08 | The method and apparatus of demand identification based on artificial intelligence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106339366B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107168948A (en) * | 2017-04-19 | 2017-09-15 | 广州视源电子科技股份有限公司 | A kind of sentence recognition methods and system |
CN108170679A (en) * | 2017-12-28 | 2018-06-15 | 中国联合网络通信集团有限公司 | It can recognize that the semantic matching method and system of natural language description based on computer |
CN108628834A (en) * | 2018-05-14 | 2018-10-09 | 国家计算机网络与信息安全管理中心 | A kind of word lists dendrography learning method based on syntax dependence |
CN110278111A (en) * | 2019-05-29 | 2019-09-24 | 西安电子科技大学 | A kind of intention driving network generic structure and its it is intended to driving network translation method |
CN110309252A (en) * | 2018-02-28 | 2019-10-08 | 阿里巴巴集团控股有限公司 | A kind of natural language processing method and device |
DE102023208152A1 (en) | 2022-08-30 | 2024-02-29 | Robert Bosch Engineering And Business Solutions Private Limited | Method and system for processing technical requirements |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101373532A (en) * | 2008-07-10 | 2009-02-25 | 昆明理工大学 | FAQ Chinese request-answering system implementing method in tourism field |
US20120185483A1 (en) * | 2011-01-14 | 2012-07-19 | Shah Amip J | System and method for tree creation |
CN105094315A (en) * | 2015-06-25 | 2015-11-25 | 百度在线网络技术(北京)有限公司 | Method and apparatus for smart man-machine chat based on artificial intelligence |
CN105701253A (en) * | 2016-03-04 | 2016-06-22 | 南京大学 | Chinese natural language interrogative sentence semantization knowledge base automatic question-answering method |
-
2016
- 2016-08-08 CN CN201610643685.0A patent/CN106339366B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101373532A (en) * | 2008-07-10 | 2009-02-25 | 昆明理工大学 | FAQ Chinese request-answering system implementing method in tourism field |
US20120185483A1 (en) * | 2011-01-14 | 2012-07-19 | Shah Amip J | System and method for tree creation |
CN105094315A (en) * | 2015-06-25 | 2015-11-25 | 百度在线网络技术(北京)有限公司 | Method and apparatus for smart man-machine chat based on artificial intelligence |
CN105701253A (en) * | 2016-03-04 | 2016-06-22 | 南京大学 | Chinese natural language interrogative sentence semantization knowledge base automatic question-answering method |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107168948A (en) * | 2017-04-19 | 2017-09-15 | 广州视源电子科技股份有限公司 | A kind of sentence recognition methods and system |
CN108170679A (en) * | 2017-12-28 | 2018-06-15 | 中国联合网络通信集团有限公司 | It can recognize that the semantic matching method and system of natural language description based on computer |
CN108170679B (en) * | 2017-12-28 | 2021-09-03 | 中国联合网络通信集团有限公司 | Semantic matching method and system based on computer recognizable natural language description |
CN110309252A (en) * | 2018-02-28 | 2019-10-08 | 阿里巴巴集团控股有限公司 | A kind of natural language processing method and device |
CN110309252B (en) * | 2018-02-28 | 2023-11-24 | 阿里巴巴集团控股有限公司 | Natural language processing method and device |
CN108628834A (en) * | 2018-05-14 | 2018-10-09 | 国家计算机网络与信息安全管理中心 | A kind of word lists dendrography learning method based on syntax dependence |
CN108628834B (en) * | 2018-05-14 | 2022-04-15 | 国家计算机网络与信息安全管理中心 | Word expression learning method based on syntactic dependency relationship |
CN110278111A (en) * | 2019-05-29 | 2019-09-24 | 西安电子科技大学 | A kind of intention driving network generic structure and its it is intended to driving network translation method |
CN110278111B (en) * | 2019-05-29 | 2021-08-31 | 西安电子科技大学 | General architecture of intention-driven network and translation method of intention-driven network |
DE102023208152A1 (en) | 2022-08-30 | 2024-02-29 | Robert Bosch Engineering And Business Solutions Private Limited | Method and system for processing technical requirements |
Also Published As
Publication number | Publication date |
---|---|
CN106339366B (en) | 2019-05-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106339366A (en) | Method and device for requirement identification based on artificial intelligence (AI) | |
US10540965B2 (en) | Semantic re-ranking of NLU results in conversational dialogue applications | |
CN111506722A (en) | Knowledge graph question-answering method, device and equipment based on deep learning technology | |
CN110033022A (en) | Processing method, device and the storage medium of text | |
CN104750674B (en) | A kind of man-machine conversation's satisfaction degree estimation method and system | |
CN110110062A (en) | Machine intelligence answering method, device and electronic equipment | |
CN109857846B (en) | Method and device for matching user question and knowledge point | |
CN110096711A (en) | The natural language semantic matching method of the concern of the sequence overall situation and local dynamic station concern | |
CN110276023A (en) | POI changes event discovery method, apparatus, calculates equipment and medium | |
CN110222780A (en) | Object detecting method, device, equipment and storage medium | |
CN109272003A (en) | A kind of method and apparatus for eliminating unknown error in deep learning model | |
EP3726435A1 (en) | Deep neural network training method and apparatus, and computer device | |
CN112685550B (en) | Intelligent question-answering method, intelligent question-answering device, intelligent question-answering server and computer readable storage medium | |
CN106202288A (en) | The optimization method of a kind of man-machine interactive system knowledge base and system | |
CN105895089A (en) | Speech recognition method and device | |
CN112613321A (en) | Method and system for extracting entity attribute information in text | |
CN114120166B (en) | Video question-answering method and device, electronic equipment and storage medium | |
CN107967304A (en) | Session interaction processing method, device and electronic equipment | |
CN110532562A (en) | Neural network training method, Chinese idiom misuse detection method, device and electronic equipment | |
CN113420833A (en) | Visual question-answering method and device based on question semantic mapping | |
CN108932289B (en) | Question answer processing method and system based on information extraction and deep learning | |
CN114970666B (en) | Spoken language processing method and device, electronic equipment and storage medium | |
CN115617975A (en) | Intention identification method and device for few-sample and multi-turn conversations | |
CN112541557B (en) | Training method and device for generating countermeasure network and electronic equipment | |
CN110609995B (en) | Method and device for constructing Tibetan language question-answer corpus |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |