CN109697679A - Intellectual property services guidance method and system - Google Patents
Intellectual property services guidance method and system Download PDFInfo
- Publication number
- CN109697679A CN109697679A CN201811609966.XA CN201811609966A CN109697679A CN 109697679 A CN109697679 A CN 109697679A CN 201811609966 A CN201811609966 A CN 201811609966A CN 109697679 A CN109697679 A CN 109697679A
- Authority
- CN
- China
- Prior art keywords
- slot position
- model
- service
- entity
- term vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 239000013598 vector Substances 0.000 claims abstract description 81
- 238000010801 machine learning Methods 0.000 claims abstract description 36
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 27
- 238000002372 labelling Methods 0.000 claims description 7
- 238000004088 simulation Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 5
- 230000008520 organization Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 2
- 238000007670 refining Methods 0.000 abstract 1
- 238000004422 calculation algorithm Methods 0.000 description 17
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000003058 natural language processing Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003032 molecular docking Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/18—Legal services
- G06Q50/184—Intellectual property management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Technology Law (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- General Physics & Mathematics (AREA)
- Operations Research (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Linguistics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Machine Translation (AREA)
Abstract
The invention patent discloses intellectual property service guidance method and system, which carries out entity type mark using term vector of the entity recognition model based on machine learning to the serializing of extraction；The term vector of serializing is used to carry out user's intent classifier as the disaggregated model based on machine learning；Corresponding content slot position model is selected according to user's intent classifier of acquisition, will be filled with the term vector that entity type marks to the content slot position model；According to the content slot position model of the term vector filled with entity type mark, corresponding intellectual property service interface is exported.The thinking that the present invention passes through the Task chat robots constructed based on machine learning, user demand is subjected to precise classification identification, and it is directed to reasonable service interface, intellectual property service category is efficiently solved to be difficult to segment, service is difficult to the problem of refining, and provide the user more accurate efficient service.
Description
Technical field
The present invention relates to natural language processing fields, and in particular to a kind of intellectual property service based on machine learning algorithm
Guidance method and system.
Background technique
In order to improve work effectiveness and intelligence, in more and more fields, begin to use artificial intelligence to the mankind from
Right language is understood and is fed back.In particular with the continuous innovation of machine learning algorithm, processing and reason for natural language
Xie Douyou significant progress the plurality of application scenes such as reads in machine translation, speech recognition, machine and achieves preferable effect
Fruit.
Intelligent chat robots are based on machine learning and construct to natural language processing algorithm.Wherein, in order to solve specific
The chat robots of business, referred to as Task talk with robot.Task humanoid robot usually provides information or clothes under given conditions
Business.Under normal conditions, be in order to meet and have the user that has a definite purpose, such as look into flow, look into telephone expenses, ticket booking, dining reservation, booking,
The Tasks scenes such as consulting.Task chat robots have been widely used for the interactive interface of a variety of smart machines, are applied to
In various interaction scenarios.
Intellectual property service field is substantially also identical as other service fields, and there is a large amount of basic interaction scenarios.
And in the routine work of intellectual property service organization, usual attendant has business scope range and knowledge expertise range, when
When business scope is complex, user demand can not be guided to correct attendant, cause biggish communication cost and manpower
Waste.
Chat robots are applied to intellectual property service guiding, there is following difficult points: (1) user does not have usually and knows
Know basis of property right knowledge, the problem of proposition usual semantic ambiguity, can not accurately correspond to corresponding entitative concept, cause input literary
There are disunities with actual demand for this；(2) different attendants possess different majors technical ability and business scope, as business is related to
Different field, may be since counterpart does not influence service quality to professional skill；(3) with the complication of Intellectual Property Rights Issues, user
There is increasingly complex careful requirement to the careful degree and professional skill of service, rough service docking causes the wave of resource
Expense and problem fail the hidden danger properly settled.
The application is intended to overcome the difficult point of intellectual property service field chat robots, proposes a kind of based on machine learning calculation
The intellectual property service guidance method and system of method.
Summary of the invention
The invention patent is designed to provide a kind of intellectual property service guiding side realized based on machine learning algorithm
Method and system carry out natural language processing by inputting text to user, are effectively finely divided service, and guide to standard
True intellectual property service interface.
To achieve the above object, the invention proposes a kind of intellectual properties to service guidance method, includes the following steps:
It obtains user and inputs information, the term vector of information extraction serializing is inputted to user；
Entity type mark is carried out using term vector of the entity recognition model based on machine learning to the serializing of extraction；
The term vector of serializing is used to carry out user's intent classifier as the disaggregated model based on machine learning；
Select corresponding content slot position model according to user's intent classifier of acquisition, by the word marked with entity type to
Amount is filled to the content slot position model；
According to the content slot position model of the term vector filled with entity type mark, corresponding intellectual property service is exported
Interface.
Wherein in a preferred embodiment, the entity recognition model based on machine learning is Bi LSTM-CRF model, tool
Body includes:
Input layer obtains the term vector of serializing；
It is LSTM layers two-way, two-way Series Modeling is carried out to the term vector for the serializing that input layer obtains, and obtain two-way sequence
Entity in column modeling marks decision probability, and the parameter of the two-way Series Modeling is obtained by the data training based on machine learning
?；
CRF layers, the joint probability of the entity mark decision probability of all term vectors is calculated, global optimum's output state is obtained
Sequence；
Output layer, global optimum's output state sequence is corresponding to the output of entity classification sample space.
Wherein in a preferred embodiment, the disaggregated model based on machine learning is CNN-LSTM model, is specifically included:
Input layer obtains the term vector of serializing；
LSTM layers, the term vector of serializing is screened by series model, and retain have semantic feature word to
The parameter of amount, the Series Modeling is obtained by the data training based on machine learning；
CNN layers, classified calculating is carried out by term vector with semantic feature of the CNN model to acquisition；
Output layer, the classification results that classified calculating is obtained are corresponding to sample space, export and divide corresponding to sample space
Category feature.
Wherein in a preferred embodiment, the content slot position model is specially following form: user's intent classifier (entity slot
Position 1=slot position information 1, entity slot position 2=slot position information 2 ... entity slot position n=slot position information n)；
Wherein, the corresponding output by the disaggregated model of user's intent classifier, entity slot position correspond to the Entity recognition
The entity type mark that model obtains, slot position information correspondent entity type mark matched term vector.
Wherein in a preferred embodiment, user's intent classifier of the content slot position model specifically include retrieval service class,
Really power service class, operation management service class, right-safeguarding service class, enterprises service class and training service class；The entity slot position is specific
Including main information, cycle information and cost information.
Wherein in a preferred embodiment, intellectual property service interface includes local service interface and external service interface, institute
It states local service interface to be provided by operator, the external service interface is provided by the third party service organization.
Wherein in a preferred embodiment, it is one of text, image and voice or a variety of that the user, which inputs information,.
It is described that data instruction is carried out by following steps based on the disaggregated model of machine learning wherein in a preferred embodiment
Practice:
Talked with by manual simulation and obtains intellectual property simulation dialogue data；
The semantic feature label that dialogue data is simulated by semantics recognition model extraction will by way of manual labeling
Corresponding service type is labeled as tag along sort；
Using the semantic feature vector with tag along sort to the training data as disaggregated model, training obtains semantic feature
The corresponding relationship of vector and tag along sort.
Wherein in a preferred embodiment, the training data of the disaggregated model is trained opening up for data by following steps
Exhibition:
User is obtained the Feedback Evaluation for the intellectual property service interface not exported, such as user feedback evaluation demand for services
It is resolved, then service guiding is carried out by contact staff；
Record carries out the classification of the service type after service guiding by contact staff, believes using service type classification as input
The tag along sort of the semantic feature vector of breath；
The semantic feature vector for indicating tag along sort is supplemented into the training data, and re -training obtains semantic spy
Levy the corresponding relationship of vector and tag along sort.
The invention also provides a kind of intellectual properties to service guidance system, including user inputs acquisition module, term vector mentions
Modulus block, entity labeling module, user's intent classifier module, content slot position model filling module and service guiding module；
Input obtains module, inputs information for obtaining user；
Term vector extraction module carries out term vector extraction for inputting text to the user of acquisition, and retains serializing
Term vector；
Entity labeling module, for using the term vector of serializing of the entity marking model based on machine learning to acquisition
Carry out entity mark；
User's intent classifier module, for using the term vector of serializing of the disaggregated model based on machine learning to acquisition
Carry out user's intent classifier；
Content slot position model fills module, for selecting content slot position model, and root according to user's intent classifier of acquisition
It fills according to the term vector that entity marks to the content slot position model；
Guiding module is serviced, for returning to intellectual property service interface according to the content slot position model filled with term vector.
The invention also provides a kind of computer equipment, including memory and processor, the memory is stored with calculating
Machine program, the processor realize intellectual property service guidance method proposed by the invention when executing the computer program.
The invention also provides a kind of computer readable storage mediums, store computer program thereon, which is characterized in that
The computer program realizes intellectual property service guidance method proposed by the invention when being executed by processor.
By adopting the above technical scheme, the invention patent has the advantages that
(1) using the entity marking model based on machine learning to input content of text understand, by term vector it
Between association, determine the semanteme of term vector.Compared to the algorithm for using Keywords matching, fuzzy matching can be effectively realized.
So that user, without accurate description intellectual property vocabulary, intention can be understood and be identified by system.It efficiently solves and knows
Know title services question and answer scene, the fuzzy problem of semantic meaning representation.
(2) user's intent classifier is carried out to input text using the disaggregated model based on machine learning, so that user inputs
It can be pin-pointed to corresponding service field, and service content is further limited by content slot position model.So that clothes
Business classification is more fine accurate, and the intellectual property service of different demands grade can be resolved, and improves and promotes service quality, and
Make the human resources of intellectual property service can be with each one does his duty, more efficiently.
(3) bidirectional relationship of sequence is taken into account model, more when solving the problems, such as entity mark by BiLSTM-CRF model
Add the application scenarios for meeting intellectual property service, it is more accurate to the mark of entity.
(4) CNN-LSTM model extracts important semantic information by LSTM unit when solving classification problem, then passes through
CNN network carries out further feature identification and classification, has preferable classifying quality for the classification of natural language.
Invention is further described in detail with reference to the accompanying drawings and embodiments；But it is of the invention to be calculated based on machine learning
The generation method of user's intent classifier of method is not limited to the embodiment.
Detailed description of the invention
Fig. 1 is the flow diagram of intellectual property service guidance method of the invention；
Fig. 2 is the flow diagram of disaggregated model of the invention；
Fig. 3 is the structure chart of BiLSTM-CRF model；
Fig. 4 is the structural schematic diagram of LSTM basic unit；
Fig. 5 is the local structural graph of CNN-LSTM model；
Fig. 6 is the whole process structure chart of CNN-LSTM model.
Specific embodiment
The invention patent is described further in the following with reference to the drawings and specific embodiments.
Shown in Figure 1, intellectual property of the invention services guidance method, successively as follows according to realization sequence.
S100 obtains user and inputs information, and the term vector of information extraction serializing is inputted to user.
It can be one of text, picture and voice or a variety of that the S101 user, which inputs information, wherein text information passes through
Word2vec algorithm is converted to term vector；Pictorial information is converted the text information in picture to by picture text conversion module
Text information, and post-processing approach identical text this information；Voice messaging is converted voice messaging to by speech text conversion module
Text information, and post processing mode identical text this information；Picture text conversion module, speech text conversion module have published
Software or open source algorithm are available, no longer do specifically describe explanation herein.
S200 carries out entity type using term vector of the entity recognition model based on machine learning to the serializing of extraction
Mark；
Shown in Figure 2, the used entity recognition model based on machine learning is BiLSTM-CRF model.LSTM
(Long Short-Term Memory) is shot and long term memory network, is a kind of time recurrent neural network (RNN)；And BiLSTM
As by the two-way network structure used of LSTM unit.
Wherein the structure of LSTM unit is shown in Figure 3, each LSTM unit has input gate, out gate and forgetting
Door, LSTM unit has the order of connection, and connects and form input gate and the corresponding sequence of out gate.Each LSTM unit all has
Door is forgotten, since LSTM structure is to form based on sequence relation training, therefore can sentence according to the sequence of reading to prediction is hereafter carried out
It is disconnected, and transmitted according to sequence direction, when information updates, outmoded information is forgotten by forgeing door, and will have
The information of value is transmitted to next LSTM unit by out gate.
By input gate, out gate and forget door, is expressed as it, ot, fiLSTM update method are as follows:
it=σ (Wixt+Uiht-1+bi)
ot=σ (Woxt+Uoht-1+bo)
ft=σ (Wfxt+Ufht-1+bf)
The state at the LSTM unit current time is ct, ht。
Two-way LSTM network is combined by two LSTM networks, wherein a LSTM network forward direction transmits information, wherein
Another LSTM network reverse transfer information.By two-way LSTM network, so that context relation is retained simultaneously.Pass through test set
Test discovery, which has preferable performance for the algorithm based on natural language processing.The present invention answers the network
For the processing of intellectual property service session data, by test, compared to the algorithm of other models, recognition accuracy has larger
It is promoted.
CRF algorithm full name is Conditional Random Field.Its form is as follows:
Wherein, it is observation variable that condition random field, which participates in the X calculated, i.e., known variables, Y are target variable, i.e., implicit to become
Amount.When being applied particularly in the part-of-speech tagging problem of natural language, modeling format be can be deformed into:
By CRF model, the joint probability that can be exported to BiLSTM network obtains the output state sequence of last global optimum
Column.
The specific structure of BiLSTM-CRF model is as follows:
S210 input layer obtains the term vector of serializing；
S220 is LSTM layers two-way, carries out two-way Series Modeling to the term vector for the serializing that input layer obtains, and obtain double
Entity into Series Modeling marks decision probability, and the parameter of the two-way Series Modeling passes through the data training based on machine learning
It obtains；
CRF layers of S230, the joint probability of the entity mark decision probability of all term vectors is calculated, it is defeated to obtain global optimum
Do well sequence；
S240 output layer, global optimum's output state sequence is corresponding to the output of entity classification sample space.
S300 uses the term vector of serializing to carry out user's intent classifier as the disaggregated model based on machine learning.
Shown in Figure 4, the disaggregated model based on machine learning is CNN-LSTM model,
S310 input layer obtains the term vector of serializing；
S320LSTM layers, the term vector of serializing is screened by series model, and retained with semantic feature
The parameter of term vector, the Series Modeling is obtained by the data training based on machine learning；
CNN layers of S330, the term vector by CNN model to acquisition with semantic feature carries out classified calculating, and CNN layers
Structure it is shown in Figure 5；
S331 reads in LSTM layers of output, and forms term vector matrix；
S332 convolutional layer selects the convolution window (n*m) an of size, and n is the number of word in window, and m is term vector dimension
Degree；Characteristic filter is carried out to convolution window by convolution kernel；
The pond S333 layer carries out the compression of data scale to the result of convolutional layer, takes maximum in fixed pane size
Value (can also be averaged), while filter the numerical fluctuations as caused by noise；
S340 output layer, i.e., the classification results obtained classified calculating by SOFTMAX function are corresponding to sample space, defeated
Characteristic of division corresponding to sample space out.
The S200 synchronous can be carried out with S300, can also be using the output result of step S200 as the input of S300 step.
The difference of two kinds of processing methods is that the synchronous scheme carried out can reach faster treatment effeciency, and the model of two steps is mutual
Independently, the problem of being less prone to over-fitting or over-sampling.And use step S200 mark after as user's intent classifier model
Input, can in advance filtration fraction speak in a low voice justice value term vector, help to improve the accuracy of sorting algorithm.
S400 selects corresponding content slot position model according to user's intent classifier of acquisition, by what is marked with entity type
Term vector is filled to the content slot position model.
During S410 such as matching content slot position model, there are slot position loss of learning, then prompt information is returned to, and will use
The supplement input information at family is combined rear return step S200 with existing input information and successively executes.
The purpose of this step is to convert normalization vector for user version by natural language understanding.And it will finally use
The input text resolution at family is the shape of act (slot1=value1, slot2=value2......slot n=value n)
Formula, i.e. intent classifier, slot position, slot position information triple form.
The content slot position model is specially following form: user's intent classifier (entity slot position 1=slot position information 1, entity
Slot position 2=slot position information 2 ... entity slot position n=slot position information n)；
Wherein, the corresponding output by the disaggregated model of user's intent classifier, entity slot position correspond to the Entity recognition
The entity type mark that model obtains, slot position information correspondent entity type mark matched term vector.
User's intent classifier of the content slot position model specifically includes retrieval service class, really power service class, operation management
Service class, right-safeguarding service class, enterprises service class and training service class.
The entity slot position specifically includes main information, cycle information and cost information.
S500 exports corresponding intellectual property according to the content slot position model of the term vector filled with entity type mark
Service interface.Intellectual property service interface includes local service interface and external service interface, and the local service interface is by transporting
It seeks quotient to provide, the external service interface is provided by the third party service organization.
Local service provides different service interfaces according to the division of labor of the intellectual property attendant of operator.Outside clothes
Business interface provides intellectual property service platform by operator, by entering for the third party service organization, provides corresponding knowledge and produces
Weigh interface.The service interface can directly be provided by attendant, can also be by automatically replying including intelligent answer system
System provides.
S600 is shown in Figure 2, described to carry out data training by following steps based on the disaggregated model of machine learning:
S610 is talked with by manual simulation obtains intellectual property simulation dialogue data；
S620 simulates the semantic feature label of dialogue data by semantics recognition model extraction, passes through the side of manual labeling
Corresponding service type is labeled by formula as tag along sort；
Using the semantic feature vector with tag along sort to the training data as disaggregated model, training obtains semantic S630
The corresponding relationship of feature vector and tag along sort.
The training data of disaggregated model described in S640 is trained the expansion of data by following steps:
S641 obtains user to the Feedback Evaluation of the intellectual property service interface exported, and evaluating service such as user feedback needs
It asks and is not resolved, then service guiding is carried out by contact staff；
S642 record carries out the classification of the service type after service guiding by contact staff, is classified using service type as defeated
Enter the tag along sort of the semantic feature vector of information；
The semantic feature vector for indicating tag along sort is supplemented into the training data by S643, and re -training obtains language
The corresponding relationship of adopted feature vector and tag along sort.
Three application examples will be enumerated below explains the course of work of the invention.
One, retrieval service class
User's input " retrieval newest patent of Internet of Things."
It inputs text to extract by term vector, generates term vector " retrieval ", " Internet of Things ", " newest ", " patent ".
By entity dimensioning algorithm, " retrieval " is noted as " behavior entity ", and " Internet of Things " is noted as " domain entities ",
" newest " is noted as " other entities ", and " patent " is noted as " domain entities ".
By sorting algorithm, inputs information and be classified to retrieval service class.The content slot position model of retrieval service class are as follows: inspection
Rope services class (searching field=slot position information 1；Searching motif=slot position information 2；Range of search=slot position information 3).
By matching, searching field slot position inserts " patent ", and searching motif slot position inserts " Internet of Things ", range of search slot position
" newest " (" newest " can be nearest 1 year or half a year according to the specific matching of setting).Match the content slot position model completed, exterior chain
To the intellectual property service interface of offer retrieval service.And by service content submission form to respective service personnel.By servicing people
Member provides retrieval service.
Two, really power services class
User sends picture or pdf document (patent accepts notice).System identification picture obtains the text letter in picture
Breath, including application number, patent name, notice issue the time.
By entity dimensioning algorithm, the patent No., patent name and the notice for obtaining mark entity issue the time.
Sorting algorithm can not obtain accurate classification of service, return to prompt information: " having been identified as patent flow file, be
The no latest tendency for needing to inquire patent? " after obtaining the further feedback of user's input text, it is identified as operation management service
Class.Corresponding content slot position model are as follows: operation management services class (service category=slot position information 1；Searching motif=slot position information
2；Service range=slot position information 3).
Matching result is that operation management services class (service category=procedure information inquiry；Searching motif=patent information；
Service range=all information).Wherein service range slot position is selected default value " all information " due to no matching.
After completing the filling of content slot position model, system is judged to that reply can be automatically provided by system queries, and system is automatic
Return to information on services.Compared to traditional mode for returning again to information by attendant's inquiry, service is more timely efficient, and
Save human resources.
Three, really power services class
User inputs voice, and " our company is wanted to apply for a new patent." system by input semantics recognition be converted into text, then
Carry out term vector extraction.Generate term vector " our company ", " thinking ", " application ", " one ", " new ", " patent ".
By entity dimensioning algorithm, " our company " is noted as " institutional bodies ", and " application " is noted as " behavior entity ",
" one ", " new " are noted as " other main bodys ", and " patent " is noted as " domain entities ".
By sorting algorithm, inputs information and be classified to true power service class.The really content slot position model of power service class are as follows: really
Power service class (service field=slot position information 1；Service theme=slot position information 2；Service range=slot position information 3).
By matching, service field slot position inserts " patent ", and service theme slot position inserts " application ", and service range retains silent
Value is recognized for sky.The content slot position model completed, the intellectual property service interface of exterior chain to the true power service of offer are provided.And it will service
Content submission form is to respective service personnel.Really power service is provided by attendant.
The foregoing is merely present pre-ferred embodiments, therefore, it cannot be limited according to technical scope of the invention, therefore Fan Yiben
Equivalent changes and modifications made by the technical spirit and description of invention, in the range of should all belonging to technical solution of the present invention.
Claims (10)
1. intellectual property services guidance method, which comprises the steps of:
It obtains user and inputs information, the term vector of information extraction serializing is inputted to user；
Entity type mark is carried out using term vector of the entity recognition model based on machine learning to the serializing of extraction；
The term vector of serializing is used to carry out user's intent classifier as the disaggregated model based on machine learning；
Corresponding content slot position model is selected according to user's intent classifier of acquisition, will be filled out with the term vector that entity type marks
It is charged to the content slot position model；
According to the content slot position model of the term vector filled with entity type mark, exports corresponding intellectual property service and connect
Mouthful.
2. intellectual property according to claim 1 services guidance method, it is characterised in that: the reality based on machine learning
Body identification model is B iLSTM-CRF model, is specifically included:
Input layer obtains the term vector of serializing；
It is LSTM layers two-way, two-way Series Modeling is carried out to the term vector for the serializing that input layer obtains, and obtain two-way sequence and build
Entity in mould marks decision probability, and the parameter of the two-way Series Modeling is obtained by the data training based on machine learning；
CRF layers, the joint probability of the entity mark decision probability of all term vectors is calculated, global optimum's output state sequence is obtained
Column；
Output layer, global optimum's output state sequence is corresponding to the output of entity classification sample space.
3. intellectual property according to claim 1 services guidance method, which is characterized in that point based on machine learning
Class model is CNN-LSTM model, is specifically included:
Input layer obtains the term vector of serializing；
LSTM layers, the term vector of serializing is screened by series model, and retains the term vector with semantic feature, institute
The parameter for stating Series Modeling is obtained by the data training based on machine learning；
CNN layers, classified calculating is carried out by term vector with semantic feature of the CNN model to acquisition；
Output layer, the classification results that classified calculating is obtained are corresponding to sample space, and it is special to export classification corresponding to sample space
Sign.
4. intellectual property according to claim 1 services guidance method, which is characterized in that the content slot position model is specific
For following form: user's intent classifier (entity slot position 1=slot position information 1, entity slot position 2=slot position information 2 ... entity slot position
N=slot position information n)；
Wherein, the corresponding output by the disaggregated model of user's intent classifier, entity slot position correspond to the entity recognition model
The entity type of acquisition marks, and slot position information correspondent entity type marks matched term vector.
5. intellectual property according to claim 1 services guidance method, it is characterised in that: the use of the content slot position model
Family intent classifier specifically includes retrieval service class, really power service class, operation management service class, right-safeguarding service class, enterprises service class
With training service class；The entity slot position specifically includes main information, cycle information and cost information.
6. intellectual property according to claim 1 services guidance method, it is characterised in that: intellectual property service interface includes
Local service interface and external service interface, the local service interface are provided by operator, and the external service interface is by
Tripartite service organization provides.
7. intellectual property according to claim 1 services guidance method, it is characterised in that: it is text that the user, which inputs information,
Originally, one of image and voice or a variety of.
8. intellectual property according to claim 1 services guidance method, which is characterized in that point based on machine learning
Class model carries out data training by following steps:
Talked with by manual simulation and obtains intellectual property simulation dialogue data；
By semantics recognition model extraction simulate dialogue data semantic feature label, by way of manual labeling, by pair
The service type answered is labeled as tag along sort；
Using the semantic feature vector with tag along sort to the training data as disaggregated model, training obtains semantic feature vector
With the corresponding relationship of tag along sort.
9. intellectual property according to claim 1 services guidance method, which is characterized in that the training number of the disaggregated model
According to the expansion for being trained data by following steps:
User is obtained not obtain the Feedback Evaluation of the intellectual property service interface exported, such as user feedback evaluation demand for services
To solution, then service guiding is carried out by contact staff；
Record carries out the classification of the service type after service guiding by contact staff, using service type classification as input information
The tag along sort of semantic feature vector；
The semantic feature vector for indicating tag along sort is supplemented into the training data, and re -training obtain semantic feature to
The corresponding relationship of amount and tag along sort.
10. intellectual property services guidance system, it is characterised in that: input data obtaining module including user, term vector extracts mould
Block, entity labeling module, user's intent classifier module, content slot position model filling module and service guiding module；
Input obtains module, inputs information for obtaining user；
Term vector extraction module carries out term vector extraction for inputting text to the user of acquisition, and retain the word of serializing to
Amount；
Entity labeling module, for using the term vector of serializing of the entity marking model based on machine learning to acquisition to carry out
Entity mark；
User's intent classifier module, for using the term vector of serializing of the disaggregated model based on machine learning to acquisition to carry out
User's intent classifier；
Content slot position model fills module, for selecting content slot position model according to user's intent classifier of acquisition, and according to tool
The term vector for having entity to mark is filled to the content slot position model；
Guiding module is serviced, for returning to intellectual property service interface according to the content slot position model filled with term vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811609966.XA CN109697679A (en) | 2018-12-27 | 2018-12-27 | Intellectual property services guidance method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811609966.XA CN109697679A (en) | 2018-12-27 | 2018-12-27 | Intellectual property services guidance method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109697679A true CN109697679A (en) | 2019-04-30 |
Family
ID=66232144
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811609966.XA Pending CN109697679A (en) | 2018-12-27 | 2018-12-27 | Intellectual property services guidance method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109697679A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110309514A (en) * | 2019-07-09 | 2019-10-08 | 北京金山数字娱乐科技有限公司 | A kind of method for recognizing semantics and device |
| CN110321564A (en) * | 2019-07-05 | 2019-10-11 | 浙江工业大学 | A kind of more wheel dialogue intension recognizing methods |
| CN110413756A (en) * | 2019-07-29 | 2019-11-05 | 北京小米智能科技有限公司 | The method, device and equipment of natural language processing |
| CN110457447A (en) * | 2019-05-15 | 2019-11-15 | 国网浙江省电力有限公司电力科学研究院 | A kind of power grid Task conversational system |
| CN112329877A (en) * | 2020-11-16 | 2021-02-05 | 山西三友和智慧信息技术股份有限公司 | Voting mechanism-based web service classification method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107562784A (en) * | 2017-07-25 | 2018-01-09 | 同济大学 | Short text classification method based on ResLCNN models |
| WO2018145098A1 (en) * | 2017-02-06 | 2018-08-09 | Thomson Reuters Global Resources Unlimited Company | Systems and methods for automatic semantic token tagging |
| CN108874774A (en) * | 2018-06-05 | 2018-11-23 | 浪潮软件股份有限公司 | A kind of service calling method and system based on intention understanding |
| CN108897857A (en) * | 2018-06-28 | 2018-11-27 | 东华大学 | The Chinese Text Topic sentence generating method of domain-oriented |
-
2018
- 2018-12-27 CN CN201811609966.XA patent/CN109697679A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018145098A1 (en) * | 2017-02-06 | 2018-08-09 | Thomson Reuters Global Resources Unlimited Company | Systems and methods for automatic semantic token tagging |
| CN107562784A (en) * | 2017-07-25 | 2018-01-09 | 同济大学 | Short text classification method based on ResLCNN models |
| CN108874774A (en) * | 2018-06-05 | 2018-11-23 | 浪潮软件股份有限公司 | A kind of service calling method and system based on intention understanding |
| CN108897857A (en) * | 2018-06-28 | 2018-11-27 | 东华大学 | The Chinese Text Topic sentence generating method of domain-oriented |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110457447A (en) * | 2019-05-15 | 2019-11-15 | 国网浙江省电力有限公司电力科学研究院 | A kind of power grid Task conversational system |
| CN110321564A (en) * | 2019-07-05 | 2019-10-11 | 浙江工业大学 | A kind of more wheel dialogue intension recognizing methods |
| CN110321564B (en) * | 2019-07-05 | 2023-07-14 | 浙江工业大学 | Multi-round dialogue intention recognition method |
| CN110309514A (en) * | 2019-07-09 | 2019-10-08 | 北京金山数字娱乐科技有限公司 | A kind of method for recognizing semantics and device |
| CN110413756A (en) * | 2019-07-29 | 2019-11-05 | 北京小米智能科技有限公司 | The method, device and equipment of natural language processing |
| CN110413756B (en) * | 2019-07-29 | 2022-02-15 | 北京小米智能科技有限公司 | Method, device and equipment for processing natural language |
| US11501078B2 (en) | 2019-07-29 | 2022-11-15 | Beijing Xiaomi Intelligent Technology Co., Ltd. | Method and device for performing reinforcement learning on natural language processing model and storage medium |
| CN112329877A (en) * | 2020-11-16 | 2021-02-05 | 山西三友和智慧信息技术股份有限公司 | Voting mechanism-based web service classification method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109697679A (en) | Intellectual property services guidance method and system | |
| CN111026842B (en) | Natural language processing method, natural language processing device and intelligent question-answering system | |
| CN110633409B (en) | Automobile news event extraction method integrating rules and deep learning | |
| CN111275401B (en) | Intelligent interview method and system based on position relation | |
| CN109753565A (en) | Intellectual Property intelligent service method and system | |
| CN110866542A (en) | Depth representation learning method based on feature controllable fusion | |
| CN111966800A (en) | Emotional dialogue generation method and device and emotional dialogue model training method and device | |
| CN115249539B (en) | Multi-mode small sample depression prediction model construction method | |
| CN109325780A (en) | A kind of exchange method of the intelligent customer service system in E-Governance Oriented field | |
| CN111339292A (en) | Training method, system, equipment and storage medium of text classification network | |
| Godse et al. | Implementation of chatbot for ITSM application Using IBM watson | |
| CN114548099A (en) | Method for jointly extracting and detecting aspect words and aspect categories based on multitask framework | |
| CN108628908A (en) | The method, apparatus and electronic equipment of sorted users challenge-response boundary | |
| CN113656564A (en) | Power grid service dialogue data emotion detection method based on graph neural network | |
| CN110287294A (en) | Intellectual property concept answers method and system automatically | |
| CN113486174A (en) | Model training, reading understanding method and device, electronic equipment and storage medium | |
| CN117056451A (en) | New energy automobile complaint text aspect-viewpoint pair extraction method based on context enhancement | |
| CN114298011B (en) | Neural network, training method, aspect emotion analysis method, device and storage medium | |
| CN110851572A (en) | Session labeling method and device, storage medium and electronic equipment | |
| CN116127954A (en) | Dictionary-based new work specialized Chinese knowledge concept extraction method | |
| CN116384372A (en) | Multi-level fusion aspect category emotion analysis method based on self-attention | |
| Frias et al. | Attention-based bilateral lstm-cnn for the sentiment analysis of code-mixed filipino-english social media texts | |
| Kreyssig | Deep learning for user simulation in a dialogue system | |
| De Vries | Reducing ambiguity during enterprise design | |
| Raut | A virtual chatbot for ITSM application |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190430 |