CN110263346A - Lexical analysis method, electronic equipment and storage medium based on small-sample learning - Google Patents

Lexical analysis method, electronic equipment and storage medium based on small-sample learning Download PDF

Info

Publication number
CN110263346A
CN110263346A CN201910569780.4A CN201910569780A CN110263346A CN 110263346 A CN110263346 A CN 110263346A CN 201910569780 A CN201910569780 A CN 201910569780A CN 110263346 A CN110263346 A CN 110263346A
Authority
CN
China
Prior art keywords
sample
user
dialog
information
relation
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
Application number
CN201910569780.4A
Other languages
Chinese (zh)
Other versions
CN110263346B (en
Inventor
龚泽熙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhuo Erzhi Lian Wuhan Research Institute Co Ltd
Original Assignee
Zhuo Erzhi Lian Wuhan Research Institute Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhuo Erzhi Lian Wuhan Research Institute Co Ltd filed Critical Zhuo Erzhi Lian Wuhan Research Institute Co Ltd
Priority to CN201910569780.4A priority Critical patent/CN110263346B/en
Publication of CN110263346A publication Critical patent/CN110263346A/en
Application granted granted Critical
Publication of CN110263346B publication Critical patent/CN110263346B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35Clustering; Classification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis

Abstract

The present invention relates to a kind of lexical analysis method, electronic equipment and storage medium based on small-sample learning.The described method includes: obtaining dialog information;Dialog model is established, and the dialog information is analyzed to obtain intent information according to the dialog model;Judge in the intent information whether to include predetermined keyword;And the problem of user is proposed is determined from the intent information when in the intent information including the predetermined keyword, response message corresponding with described problem is searched out from Q & A database for described problem, and voice plays the response message.In the present invention, the dialog information is analyzed to obtain intent information by establishing dialog model, and according to the dialog model, improves the accuracy of the intention analysis to interlocutor.

Description

Lexical analysis method, electronic equipment and storage medium based on small-sample learning
Technical field
The present invention relates to artificial intelligence fields, and in particular to a kind of lexical analysis method based on small-sample learning, electronics Equipment and storage medium.
Background technique
In the processing technique of existing intention of the dialogue to analyze interlocutor based on small-sample learning applicant, only lead to It crosses the sample vector adduction to dialogue or averages, due to natural language as the term of speaker is accustomed to difference, so The same dialogue for being intended to classification occurs greatly to change because of the difference of speaker, finally results in the classification of sample vector Judgement accuracy it is not high.
Summary of the invention
In view of the foregoing, it is necessary to propose it is a kind of by the lexical analysis method of small-sample learning, electronic equipment and based on Calculation machine readable storage medium storing program for executing is to improve the accuracy that the intention of interlocutor is analyzed.
The first aspect of the application provides a kind of lexical analysis method based on small-sample learning, which comprises
Obtain dialog information;
Based on dialog model described in Encoder-Induction-Relation three-level framework establishment and utilize small sample Learning method is trained the dialog model, and is analyzed according to the dialog model the dialog information and be intended to Information;
Judge in the intent information whether to include predetermined keyword;And
Determine what user was proposed from the intent information when in the intent information including the predetermined keyword Problem searches out response message corresponding with described problem for described problem, wherein the basis from Q & A database The dialog model is analyzed to obtain intent information to the dialog information
Encoder module in the Encoder-Induction-Relation three-level frame is believed according to the dialogue of acquisition Breath building sentence term vector matrix, and by the encoded semanteme for obtaining Sentence-level of the term vector matrix, and according to the sentence The semanteme of grade obtains the sample vector in the supported collection of the dialog model;
Induction module in the Encoder-Induction-Relation three-level frame construct the sample to The mapping process for measuring class vector obtains the class vector of classification;And
Relation module in the Encoder-Induction-Relation three-level frame constructs each sample Interactive relation between this vector and the class vector pair, and using full articulamentum to the sample vector and the class vector pair Between interactive relation give a mark with analyze obtain the intent information.
Preferably, the Encoder module in the Encoder-Induction-Relation three-level frame uses two-way Length memory network models, and the Induction module in the Encoder-Induction-Relation three-level frame uses Dynamic routing algorithm models, and the Relation module in the Encoder-Induction-Relation three-level frame uses mind Through tensor network modelling.
Preferably, described that the interactive relation between the sample vector and the class vector pair is carried out using full articulamentum It gives a mark to analyze to obtain the intent information and includes:
Marking between the sample vector to match and the class vector pair is intended to 1;And
Marking between the sample vector not matched that and the class vector pair is intended to 0.
Preferably, the method also includes:
By set interface, the predetermined keyword is set.
Preferably, described the predetermined keyword is arranged by set interface to include:
The predetermined keyword that user edits is received when detecting that user presses the Edit button of the set interface and is incited somebody to action The predetermined keyword is stored;And
The predetermined keyword for being chosen user when detecting that user presses the delete button of the set interface carries out It deletes.
Preferably, the method also includes:
It is scored according to the problem of intent information and proposed user user, and exports appraisal result.
Preferably, described to be scored according to the problem of intent information and proposed user user and export scoring Result includes:
The default point scoring for being included is analyzed in the problem of from the intent information and proposed user;And
According to formulaThe appraisal result is calculated, wherein Pi is the default point scoring, and wi is pair The weight for the default point scoring answered, N are the quantity of the default point scoring.
Preferably, the acquisition dialog information includes:
Receive user by press or touch terminal device on entity button or display on a user interface virtually press Instruction is talked in the starting that button generates, and is talked with instruction according to the starting and obtained the dialog information.
The second aspect of the application provides a kind of electronic equipment, and the electronic equipment includes processor, and the processor is used The lexical analysis method based on small-sample learning is realized when executing the computer program stored in memory.
The third aspect of the application provides a kind of computer readable storage medium, is stored thereon with computer program, described The lexical analysis method based on small-sample learning is realized when computer program is executed by processor.
In the present invention, the dialog information analyze by establishing dialog model, and according to the dialog model To intent information, the accuracy of the intention analysis to interlocutor is improved.When including described in the intent information in the present invention The problem of user is proposed is determined when predetermined keyword from the intent information, is searched from Q & A database for described problem Rope goes out response message corresponding with described problem and voice plays the response message, so as to what is proposed for user Problem searches out response message corresponding with described problem from Q & A database and voice plays the response message, thus So that company more preferably understands the demand of job hunter and judges whether job hunter meets job position request.
Detailed description of the invention
Fig. 1 is the flow chart of the lexical analysis method based on small-sample learning in an embodiment of the present invention.
Fig. 2 is the schematic diagram of set interface in an embodiment of the present invention.
Fig. 3 is the structure chart of the lexical analysis device based on small-sample learning in an embodiment of the present invention.
Fig. 4 is the schematic diagram of electronic equipment in an embodiment of the present invention.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real Applying example, the present invention will be described in detail.It should be noted that in the absence of conflict, embodiments herein and embodiment In feature can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, described embodiment is only It is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill Personnel's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Unless otherwise defined, all technical and scientific terms used herein and belong to technical field of the invention The normally understood meaning of technical staff is identical.Term as used herein in the specification of the present invention is intended merely to description tool The purpose of the embodiment of body, it is not intended that in the limitation present invention.
Preferably, the present invention is based on the lexical analysis methods of small-sample learning to apply in one or more electronic equipment In.The electronic equipment is that one kind can be automatic to carry out at numerical value calculating and/or information according to the instruction for being previously set or storing The equipment of reason, hardware include but is not limited to microprocessor, specific integrated circuit (Application Specific Integrated Circuit, ASIC), programmable gate array (Field-Programmable Gate Array, FPGA), number Word processing device (Digital Signal Processor, DSP), embedded device etc..
The electronic equipment can be the calculating such as desktop PC, laptop, tablet computer and cloud server Equipment.The equipment can carry out man-machine friendship by modes such as keyboard, mouse, remote controler, touch tablet or voice-operated devices with user Mutually.
Embodiment 1
Fig. 1 is the flow chart of the lexical analysis method in an embodiment of the present invention based on small-sample learning.According to difference Demand, the sequence of step can change in the flow chart, and certain steps can be omitted.
As shown in fig.1, the lexical analysis method based on small-sample learning specifically includes the following steps:
Step S11 obtains dialog information.
In present embodiment, user is received by pressing or touching the entity button on terminal device or be shown in user circle Instruction is talked in the starting that virtual push button on face generates, and is talked with instruction according to the starting and obtained the dialog information.Specifically , pass through the entity button for pressing or touching on terminal device or the virtual push button shown on a user interface when receiving user After the starting dialogue instruction of generation, the dialog information is obtained by a microphone apparatus according to starting dialogue instruction.
Step S12 establishes dialog model, and is analyzed according to the dialog model the dialog information and be intended to Information.
In present embodiment, the dialog model of establishing includes: based on Encoder-Induction-Relation three-level Dialog model described in framework establishment is simultaneously trained the dialog model using small-sample learning method.In present embodiment, Encoder module in the Encoder-Induction-Relation three-level frame is built using two-way length memory network Mould.In other embodiments, convolutional neural networks or Transformer structural modeling also can be used in the Encoder module. In present embodiment, the Induction module in the Encoder-Induction-Relation three-level frame uses dynamic Routing algorithm modeling.Relation in present embodiment, in the Encoder-Induction-Relation three-level frame Module uses neural tensor network modelling.In present embodiment, the dialog model is being carried out using small-sample learning method C classification is randomly selected in training set when training, K sample of each classification, C × K data are configured to a meta- in total Task, the support collection as the dialog model input, then extract prediction of a collection of sample as dialog model from this C classification Object.In the training process, every wheel can all sample to obtain different meta-task, that is, contain different category combinations, this Mechanism makes the general character part of dialog model association difference meta-task, for example, how to extract important feature and comparative sample phase Like etc., to ignore field relevant portion in meta-task.In present embodiment, using small-sample learning method to described right Words model also passes through the expression that the classification is calculated in a small amount of sample in classification when being trained, then use metric method again Final classification results are calculated.
In present embodiment, the terminal device is also stored with dialogue and is intended to relation table, the dialogue and intention relationship The corresponding relationship of multiple dialog informations Yu multiple intent informations is stored in table.The method also includes: the dialogue letter that will acquire It ceases and is matched with the dialogue with the dialog information being intended in relation table with determining corresponding with the dialog information of the acquisition Intent information, and using the determining intent information as the intent information of the dialog information of the acquisition.Present embodiment In, the method also includes: the dialog information determined is supplied to user and is confirmed, and is receiving user's input really The intent information is exported after recognizing errorless instruction.In present embodiment, the building is also used to analysis module 402 in determination When dialogue is with being intended to that the intent information to match with the dialog information of the acquisition is not present in relation table, according to the dialogue mould Type analyzes the dialog information of the acquisition to obtain intent information, and the dialog information of the acquisition and analysis are obtained Intent information is associated, and is stored to the dialogue and learnt with intention relation table with the intent information to user.
It is described that the dialog information is analyzed according to the dialog model to obtain intent information packet in present embodiment It includes:
A) the Encoder module constructs sentence term vector matrix according to the dialog information of acquisition, and by the term vector The encoded semanteme for obtaining Sentence-level of matrix, and obtained in the supported collection of the dialog model according to the semanteme of the Sentence-level Sample vector;
B) the Induction module construct the sample vector to class vector mapping process, obtain the class of classification to Amount;And
C) the Relation module constructs the interactive relation between each sample vector and class vector pair, and using connecting entirely Layer is connect to give a mark to analyze and obtain the intent information to the interactive relation between the sample vector and class vector pair.
In present embodiment, the Induction module constructs the sample vector into the mapping process of class vector, will Sample vector in supported collection is considered as input capsule, and output capsule is considered as the semanteme of each class after one layer of dynamic routing converts Character representation.Specifically, doing a matrix conversion to the sample in all supported collections first to turn the semantic space of sample grade The semantic space of classification grade is changed to, the irrelevant information in semantic space is then filtered by way of dynamic routing.This embodiment party In formula, modeling sample vector effectively can filter and classify to the mapping process of class vector by way of the dynamic routing Unrelated interference information, obtains category feature.
In present embodiment, the interactive relation between the sample vector and class vector pair is beaten using full articulamentum Point with analysis, to obtain the intent information include: that the Relation module is trained using least square loss;It will match Sample vector and class vector pair between marking be intended to 1;And it will be between the sample vector that do not matched that and class vector pair Marking is intended to 0.
Whether step S13 judges in the intent information to include predetermined keyword.
In present embodiment, the predetermined keyword, which can according to need, to be preset.It, can be in present embodiment By a set interface 20, the predetermined keyword is set.Referring to FIG. 2, showing set interface in an embodiment of the present invention 20 schematic diagram.The set interface 20 includes, but are not limited to Edit button 21 and delete button 22.It is described to pass through a setting circle Face 20 be arranged the predetermined keyword include: when detecting that user presses the Edit button 21 receive user edit preset Keyword simultaneously stores the predetermined keyword;And it will be selected by user when detecting that user presses the delete button 22 In predetermined keyword deleted.In present embodiment, the predetermined keyword can be the application in manpower resources domain The related contents such as the requirement and wages treatment in post.
Step S14 determines user when in the intent information including the predetermined keyword from the intent information The problem of proposed, searches out response message corresponding with described problem for described problem from Q & A database.
In present embodiment, the Q & A database is stored in electronic equipment or cloud server.The question and answer data Stored in library it is problematic with response message mapping table, it is described to be searched out from Q & A database for described problem and institute Stating the corresponding response message of problem includes: storing the problematic mapping table with response message from described according to described problem The response message that middle determination and described problem match, and the response message is played by speech player.Present embodiment In, when predetermined keywords such as the requirements and wages treatment in the intent information including application post, from the intent information Middle determining user proposes aiming at the problem that applying for the requirement and wages treatment in post, for described problem from Q & A database It searches out response message corresponding with described problem and voice plays the response message, asked so that company more preferably understands The demand of duty person and judge whether job hunter meets job position request.
In present embodiment, the method also includes steps: according to the problem of the intent information and proposed user pairs User scores, and exports appraisal result.
It is described to be scored according to the problem of intent information and proposed user user and defeated in present embodiment Appraisal result includes: out
The default point scoring for being included is analyzed in the problem of from the intent information and proposed user;And
According to formulaThe appraisal result is calculated, wherein Pi is default point scoring, and wi is corresponding pre- If the weight of point scoring, N is the quantity of the default point scoring.
In present embodiment, the default point scoring and weight corresponding with the default point scoring can according to need by User sets.
In present embodiment, the method is after step s 14 further include: voice plays the response message.
In the present invention, the dialog information analyze by establishing dialog model, and according to the dialog model To intent information, the accuracy of the intention analysis to interlocutor is improved.When including described in the intent information in the present invention The problem of user is proposed is determined when predetermined keyword from the intent information, is searched from Q & A database for described problem Rope goes out response message corresponding with described problem and voice plays the response message, so as to what is proposed for user Problem searches out response message corresponding with described problem from Q & A database and voice plays the response message, thus So that company more preferably understands the demand of job hunter and judges whether job hunter meets job position request.
Embodiment 2
Fig. 3 is the structure chart of the lexical analysis device 40 based on small-sample learning in an embodiment of the present invention.
In some embodiments, believe that the lexical analysis device 40 based on small-sample learning is run in electronic equipment.Institute Stating the lexical analysis device 40 based on small-sample learning may include multiple functional modules as composed by program code segments.It is described The program code of each program segment in lexical analysis device 40 based on small-sample learning can store in memory, and by Performed by least one processor.
In the present embodiment, the function of the lexical analysis device 40 according to performed by it based on small-sample learning can be with It is divided into multiple functional modules.As shown in fig.3, the lexical analysis device 40 based on small-sample learning may include obtaining Modulus block 401, building and analysis module 402, judgment module 403, responder module 404 and grading module 405.The present invention is so-called Module, which refers to, a kind of performed by least one processor and can complete the series of computation machine program of fixed function Section, storage is in memory.It is described in some embodiments, about each module function will in subsequent embodiment in detail It states.
The acquisition module 401 obtains dialog information.
In present embodiment, the acquisition module 401 receives user and is pressed by pressing or touching the entity on terminal device Instruction is talked in the starting that the virtual push button of button or display on a user interface generates, and is talked with instruction according to the starting and obtained institute State dialog information.Specifically, when receiving user by pressing or touching the entity button on terminal device or be shown in user After the starting dialogue instruction that virtual push button on interface generates, obtained according to starting dialogue instruction by a microphone apparatus The dialog information.
The building establishes dialog model with analysis module 402, and according to the dialog model to the dialog information into Row analysis obtains intent information.
In present embodiment, it includes: based on Encoder- that the building, which establishes dialog model with analysis module 402, Dialog model described in Induction-Relation three-level framework establishment simultaneously utilizes small-sample learning method to the dialog model It is trained.In present embodiment, the Encoder module in the Encoder-Induction-Relation three-level frame makes It is modeled with two-way length memory network.In other embodiments, the Encoder module also can be used convolutional neural networks or Transformer structural modeling.In present embodiment, in the Encoder-Induction-Relation three-level frame Induction module is modeled using dynamic routing algorithm.In present embodiment, the Encoder-Induction-Relation Relation module in three-level frame uses neural tensor network modelling.In present embodiment, the building and analysis module 402 randomly select C classification when being trained using small-sample learning method to the dialog model in training set, each K sample of classification, C × K data are configured to a meta-task in total, and the support collection as the dialog model inputs, then Prediction object of a collection of sample as dialog model is extracted from this C classification.In the training process, every wheel can all sample to obtain not Same meta-task, that is, contain different category combinations, this mechanism makes dialog model association difference meta-task's General character part, for example, important feature how is extracted and comparative sample is similar etc., to ignore field dependent part in meta-task Point.Also pass through a small amount of sample in classification in present embodiment, when being trained using small-sample learning method to the dialog model Originally the expression of the classification is calculated, final classification results are then calculated with metric method again.
In present embodiment, the terminal device is also stored with dialogue and is intended to relation table, the dialogue and intention relationship The corresponding relationship of multiple dialog informations Yu multiple intent informations is stored in table.It is described building with analysis module 402 be also used to by The dialog information of acquisition is matched with the dialogue with the dialog information being intended in relation table with determining pair with the acquisition The corresponding intent information of information is talked about, and is believed the determining intent information as the intention of the dialog information of the acquisition Breath.In present embodiment, the dialog information that the building is also used to determine with analysis module 402 is supplied to user and carries out really Recognize, and exports the intent information after the errorless instruction of the confirmation for receiving user's input.In present embodiment, the building It is also used to match in determining dialogue and intention relation table there is no the dialog information with the acquisition with analysis module 402 When intent information, analyzed to obtain intent information according to dialog information of the dialog model to the acquisition, and will be described The intent information that the dialog information of acquisition is obtained with analysis is associated, and store to it is described dialogue be intended to relation table with to The intent information at family is learnt.
In present embodiment, the building carries out the dialog information according to the dialog model with analysis module 402 Analysis obtains intent information and includes:
A) the Encoder module constructs sentence term vector matrix according to the dialog information of acquisition, and by the term vector The encoded semanteme for obtaining Sentence-level of matrix, and obtained in the supported collection of the dialog model according to the semanteme of the Sentence-level Sample vector;
B) the Induction module construct the sample vector to class vector mapping process, obtain the class of classification to Amount;And
C) the Relation module constructs the interactive relation between each sample vector and class vector pair, and using connecting entirely Layer is connect to give a mark to analyze and obtain the intent information to the interactive relation between the sample vector and class vector pair.
In present embodiment, the Induction module constructs the sample vector into the mapping process of class vector, will Sample vector in supported collection is considered as input capsule, and output capsule is considered as the semanteme of each class after one layer of dynamic routing converts Character representation.Specifically, doing a matrix conversion to the sample in all supported collections first to turn the semantic space of sample grade The semantic space of classification grade is changed to, the irrelevant information in semantic space is then filtered by way of dynamic routing.This embodiment party In formula, modeling sample vector effectively can filter and classify to the mapping process of class vector by way of the dynamic routing Unrelated interference information, obtains category feature.
In present embodiment, it is described using full articulamentum to the interactive relation between the sample vector and class vector pair into Row marking includes: that the Relation module is trained using least square loss to analyze to obtain the intent information;By phase Marking between matched sample vector and class vector pair is intended to 1;And by the sample vector not matched that and class vector to it Between marking be intended to 0.
Whether the judgment module 403 judges in the intent information to include predetermined keyword.
In present embodiment, the predetermined keyword, which can according to need, to be preset.It is described in present embodiment The predetermined keyword can be arranged by a set interface 20 in judgment module 403.Referring to FIG. 2, it is real to show the present invention one Apply the schematic diagram of set interface 20 in mode.The set interface 20 includes, but are not limited to Edit button 21 and delete button 22. It is described by a set interface 20 be arranged the predetermined keyword include: when detect user press the Edit button 21 when connect It receives the predetermined keyword that user edits and stores the predetermined keyword;And when detect user pressing it is described delete by The predetermined keyword for being chosen user when button 22 is deleted.In present embodiment, the predetermined keyword can be manpower The related contents such as the requirement and wages treatment in the application post in resources domain.
The responder module 404 is used for when in the intent information including the predetermined keyword from the intent information The problem of middle determining user is proposed, response corresponding with described problem is searched out for described problem from Q & A database Information, and voice plays the response message.
In present embodiment, the Q & A database is stored in electronic equipment or cloud server.The question and answer data Stored in library it is problematic with response message mapping table, it is described to be searched out from Q & A database for described problem and institute Stating the corresponding response message of problem includes: storing the problematic mapping table with response message from described according to described problem The response message that middle determination and described problem match, and the response message is played by speech player.Present embodiment In, when predetermined keywords such as the requirements and wages treatment in the intent information including application post, from the intent information Middle determining user proposes aiming at the problem that applying for the requirement and wages treatment in post, for described problem from Q & A database It searches out response message corresponding with described problem and voice plays the response message, asked so that company more preferably understands The demand of duty person and judge whether job hunter meets job position request.
In present embodiment, institute's scoring module 405 according to the problem of the intent information and proposed user to user It scores, and exports appraisal result.
In present embodiment, institute's scoring module 405 according to the problem of the intent information and proposed user to user It is scored and exports appraisal result and include:
The default point scoring for being included is analyzed in the problem of from the intent information and proposed user;And
According to formulaThe appraisal result is calculated, wherein Pi is default point scoring, and wi is corresponding pre- If the weight of point scoring, N is the quantity of the default point scoring.
In present embodiment, the default point scoring and weight corresponding with the default point scoring can according to need by User sets.
In the present invention, the dialog information analyze by establishing dialog model, and according to the dialog model To intent information, the accuracy of the intention analysis to interlocutor is improved.When including described in the intent information in the present invention The problem of user is proposed is determined when predetermined keyword from the intent information, is searched from Q & A database for described problem Rope goes out response message corresponding with described problem and voice plays the response message, so as to what is proposed for user Problem searches out response message corresponding with described problem from Q & A database and voice plays the response message, thus So that company more preferably understands the demand of job hunter and judges whether job hunter meets job position request.
Embodiment 3
Fig. 4 is the schematic diagram of electronic equipment 6 in an embodiment of the present invention.
The electronic equipment 6 includes memory 61, processor 62 and is stored in the memory 61 and can be described The computer program 63 run on processor 62.The processor 62 is realized above-mentioned based on small when executing the computer program 63 Step in the lexical analysis embodiment of the method for sample learning, such as step S11~S14 shown in FIG. 1.Alternatively, the processing Device 62 realized when executing the computer program 63 each module in the above-mentioned lexical analysis Installation practice based on small-sample learning/ The function of unit, such as the module 401~405 in Fig. 3.
Illustratively, the computer program 63 can be divided into one or more module/units, it is one or Multiple module/units are stored in the memory 61, and are executed by the processor 62, to complete the present invention.Described one A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, and described instruction section is used In implementation procedure of the description computer program 63 in the electronic equipment 6.For example, the computer program 63 can be by It is divided into acquisition module 401 in Fig. 3, building and analysis module 402, judgment module 403, responder module 404 and grading module 405, each module concrete function is referring to embodiment 2.
In present embodiment, it is whole that the electronic equipment 6 can be desktop PC, notebook, palm PC and cloud End device etc. calculates equipment.It will be understood by those skilled in the art that the schematic diagram is only the example of electronic equipment 6, not Restriction to electronic equipment 6 is constituted, may include perhaps combining certain components or not than illustrating more or fewer components Same component, such as the electronic equipment 6 can also include input-output equipment, network access equipment, bus etc..
Alleged processor 62 can be central processing module (Central Processing Unit, CPU), can also be Other general processors, digital signal processor (Digital Signal Processor, DSP), specific integrated circuit (Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field- Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic, Discrete hardware components etc..General processor can be microprocessor or the processor 62 is also possible to any conventional processing Device etc., the processor 62 are the control centres of the electronic equipment 6, utilize various interfaces and the entire electronic equipment of connection 6 various pieces.
The memory 61 can be used for storing the computer program 63 and/or module/unit, and the processor 62 passes through Operation executes the computer program and/or module/unit being stored in the memory 61, and calls and be stored in memory Data in 61 realize the various functions of the electronic equipment 6.The memory 61 can mainly include storing program area and storage Data field, wherein storing program area can application program needed for storage program area, at least one function (for example sound plays Function, image player function etc.) etc.;Storage data area, which can be stored, uses created data (such as sound according to electronic equipment 6 Frequency evidence, phone directory etc.) etc..In addition, memory 61 may include high-speed random access memory, it can also include non-volatile Memory, such as hard disk, memory, plug-in type hard disk, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash card (Flash Card), at least one disk memory, flush memory device or other Volatile solid-state part.
If the integrated module/unit of the electronic equipment 6 is realized in the form of software function module and as independent Product when selling or using, can store in a computer readable storage medium.Based on this understanding, the present invention is real All or part of the process in existing above-described embodiment method, can also instruct relevant hardware come complete by computer program At the computer program can be stored in a computer readable storage medium, and the computer program is held by processor When row, it can be achieved that the step of above-mentioned each embodiment of the method.Wherein, the computer program includes computer program code, institute Stating computer program code can be source code form, object identification code form, executable file or certain intermediate forms etc..It is described Computer-readable medium may include: any entity or device, recording medium, U that can carry the computer program code Disk, mobile hard disk, magnetic disk, CD, computer storage, read-only memory (ROM, Read-Only Memory), arbitrary access Memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..It needs It is bright, the content that the computer-readable medium includes can according in jurisdiction make laws and patent practice requirement into Row increase and decrease appropriate, such as do not include electric load according to legislation and patent practice, computer-readable medium in certain jurisdictions Wave signal and telecommunication signal.
In several embodiments provided by the present invention, it should be understood that disclosed electronic equipment and method, Ke Yitong Other modes are crossed to realize.For example, electronic equipment embodiment described above is only schematical, for example, the module Division, only a kind of logical function partition, there may be another division manner in actual implementation.
It, can also be in addition, each functional module in each embodiment of the present invention can integrate in same treatment module It is that modules physically exist alone, can also be integrated in equal modules with two or more modules.Above-mentioned integrated mould Block both can take the form of hardware realization, can also realize in the form of hardware adds software function module.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie In the case where without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims Variation is included in the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.This Outside, it is clear that one word of " comprising " is not excluded for other modules or step, and odd number is not excluded for plural number.It is stated in electronic equipment claim Multiple modules or electronic equipment can also be implemented through software or hardware by the same module or electronic equipment.The first, the Second-class word is used to indicate names, and is not indicated any particular order.
Finally it should be noted that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although reference Preferred embodiment describes the invention in detail, those skilled in the art should understand that, it can be to of the invention Technical solution is modified or equivalent replacement, without departing from the spirit and scope of the technical solution of the present invention.

Claims (10)

1. a kind of lexical analysis method based on small-sample learning, which is characterized in that the described method includes:
Obtain dialog information;
Based on dialog model described in Encoder-Induction-Relation three-level framework establishment and utilize small-sample learning side Method is trained the dialog model, and is analyzed according to the dialog model the dialog information to obtain intention letter Breath;
Judge in the intent information whether to include predetermined keyword;And
The problem of user is proposed is determined from the intent information when in the intent information including the predetermined keyword, Response message corresponding with described problem is searched out from Q & A database for described problem, wherein described according to Dialog model is analyzed to obtain intent information to the dialog information
Encoder module in the Encoder-Induction-Relation three-level frame is according to the dialog information structure of acquisition Sentence term vector matrix is built, and by the encoded semanteme for obtaining Sentence-level of the term vector matrix, and according to the Sentence-level Semanteme obtains the sample vector in the supported collection of the dialog model;
Induction module in the Encoder-Induction-Relation three-level frame constructs the sample vector and arrives The mapping process of class vector obtains the class vector of classification;And
Relation module in the Encoder-Induction-Relation three-level frame construct each sample to Interactive relation between amount and the class vector pair, and using full articulamentum between the sample vector and the class vector pair Interactive relation give a mark with analyze obtain the intent information.
2. the lexical analysis method based on small-sample learning as described in claim 1, which is characterized in that the Encoder- Encoder module in Induction-Relation three-level frame is modeled using two-way length memory network, described Induction module in Encoder-Induction-Relation three-level frame is modeled using dynamic routing algorithm, described Relation module in Encoder-Induction-Relation three-level frame uses neural tensor network modelling.
3. the lexical analysis method based on small-sample learning as described in claim 1, which is characterized in that described to use full connection Layer gives a mark to analyze and obtain the intent information packet to the interactive relation between the sample vector and the class vector pair It includes:
Marking between the sample vector to match and the class vector pair is intended to 1;And
Marking between the sample vector not matched that and the class vector pair is intended to 0.
4. the lexical analysis method based on small-sample learning as described in claim 1, which is characterized in that the method is also wrapped It includes:
By set interface, the predetermined keyword is set.
5. the lexical analysis method based on small-sample learning as claimed in claim 4, which is characterized in that described by the way that boundary is arranged The predetermined keyword is arranged in face
Received when detecting that user presses the Edit button of the set interface predetermined keyword edit of user and will described in Predetermined keyword is stored;And
The predetermined keyword for being chosen user when detecting that user presses the delete button of the set interface is deleted.
6. the lexical analysis method based on small-sample learning as described in claim 1, which is characterized in that the method is also wrapped It includes:
It is scored according to the problem of intent information and proposed user user, and exports appraisal result.
7. the lexical analysis method based on small-sample learning as claimed in claim 6, which is characterized in that described according to the meaning Figure information and the problem of proposed user, which score to user and export appraisal result, includes:
The default point scoring for being included is analyzed in the problem of from the intent information and proposed user;And
According to formulaThe appraisal result is calculated, wherein Pi is the default point scoring, and wi is corresponding The weight of default point scoring, N are the quantity of the default point scoring.
8. the lexical analysis method based on small-sample learning as described in claim 1, which is characterized in that the acquisition dialogue letter Breath includes:
It is raw by pressing or touching the virtual push button of entity button or display on a user interface on terminal device to receive user At starting talk with instruction, and according to the starting dialogue instruction acquisition dialog information.
9. a kind of electronic equipment, it is characterised in that: the electronic equipment includes processor, and the processor is for executing memory The lexical analysis side based on small-sample learning as described in any one of claim 1-8 is realized when the computer program of middle storage Method.
10. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer program The lexical analysis method as described in any one of claim 1-8 based on small-sample learning is realized when being executed by processor.
CN201910569780.4A 2019-06-27 2019-06-27 Semantic analysis method based on small sample learning, electronic equipment and storage medium Active CN110263346B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910569780.4A CN110263346B (en) 2019-06-27 2019-06-27 Semantic analysis method based on small sample learning, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910569780.4A CN110263346B (en) 2019-06-27 2019-06-27 Semantic analysis method based on small sample learning, electronic equipment and storage medium

Publications (2)

Publication Number Publication Date
CN110263346A true CN110263346A (en) 2019-09-20
CN110263346B CN110263346B (en) 2023-01-24

Family

ID=67922531

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910569780.4A Active CN110263346B (en) 2019-06-27 2019-06-27 Semantic analysis method based on small sample learning, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN110263346B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114036264A (en) * 2021-11-19 2022-02-11 四川大学 E-mail author identity attribution identification method based on small sample learning

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108363690A (en) * 2018-02-08 2018-08-03 北京十三科技有限公司 Dialog semantics Intention Anticipation method based on neural network and learning training method
CN108427722A (en) * 2018-02-09 2018-08-21 卫盈联信息技术(深圳)有限公司 intelligent interactive method, electronic device and storage medium
CN109271505A (en) * 2018-11-12 2019-01-25 深圳智能思创科技有限公司 A kind of question answering system implementation method based on problem answers pair
CN109446306A (en) * 2018-10-16 2019-03-08 浪潮软件股份有限公司 A kind of intelligent answer method of more wheels dialogue of task based access control driving

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108363690A (en) * 2018-02-08 2018-08-03 北京十三科技有限公司 Dialog semantics Intention Anticipation method based on neural network and learning training method
CN108427722A (en) * 2018-02-09 2018-08-21 卫盈联信息技术(深圳)有限公司 intelligent interactive method, electronic device and storage medium
CN109446306A (en) * 2018-10-16 2019-03-08 浪潮软件股份有限公司 A kind of intelligent answer method of more wheels dialogue of task based access control driving
CN109271505A (en) * 2018-11-12 2019-01-25 深圳智能思创科技有限公司 A kind of question answering system implementation method based on problem answers pair

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114036264A (en) * 2021-11-19 2022-02-11 四川大学 E-mail author identity attribution identification method based on small sample learning

Also Published As

Publication number Publication date
CN110263346B (en) 2023-01-24

Similar Documents

Publication Publication Date Title
WO2022095380A1 (en) Ai-based virtual interaction model generation method and apparatus, computer device and storage medium
CN108428446A (en) Audio recognition method and device
CN107657017A (en) Method and apparatus for providing voice service
CN107591155A (en) Audio recognition method and device, terminal and computer-readable recording medium
CN110491383A (en) A kind of voice interactive method, device, system, storage medium and processor
CN110427610A (en) Text analyzing method, apparatus, computer installation and computer storage medium
CN105723360A (en) Improving natural language interactions using emotional modulation
CN112468659B (en) Quality evaluation method, device, equipment and storage medium applied to telephone customer service
CN107506166A (en) Information cuing method and device, computer installation and readable storage medium storing program for executing
WO2022178969A1 (en) Voice conversation data processing method and apparatus, and computer device and storage medium
Krijnders et al. Sound event recognition through expectancy-based evaluation ofsignal-driven hypotheses
CN111179935B (en) Voice quality inspection method and device
CN111694940A (en) User report generation method and terminal equipment
CN111933127A (en) Intention recognition method and intention recognition system with self-learning capability
CN113420556B (en) Emotion recognition method, device, equipment and storage medium based on multi-mode signals
CN111182162A (en) Telephone quality inspection method, device, equipment and storage medium based on artificial intelligence
CN109918627A (en) Document creation method, device, electronic equipment and storage medium
CN110110049A (en) Service consultation method, apparatus, system, service robot and storage medium
CN111081280A (en) Text-independent speech emotion recognition method and device and emotion recognition algorithm model generation method
CN111694937A (en) Interviewing method and device based on artificial intelligence, computer equipment and storage medium
CN106649253A (en) Auxiliary control method and system based on post verification
CN112632244A (en) Man-machine conversation optimization method and device, computer equipment and storage medium
CN110851650B (en) Comment output method and device and computer storage medium
CN114065720A (en) Conference summary generation method and device, storage medium and electronic equipment
CN111798118B (en) Enterprise operation risk monitoring method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant