CN112256824B - Service guiding method, system and storage medium for robot service - Google Patents
Service guiding method, system and storage medium for robot service Download PDFInfo
- Publication number
- CN112256824B CN112256824B CN202010981046.1A CN202010981046A CN112256824B CN 112256824 B CN112256824 B CN 112256824B CN 202010981046 A CN202010981046 A CN 202010981046A CN 112256824 B CN112256824 B CN 112256824B
- Authority
- CN
- China
- Prior art keywords
- intention
- probability
- service
- alternative
- intentions
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012795 verification Methods 0.000 claims abstract description 69
- RWSOTUBLDIXVET-UHFFFAOYSA-N Dihydrogen sulfide Chemical compound S RWSOTUBLDIXVET-UHFFFAOYSA-N 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000003993 interaction Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000005611 electricity Effects 0.000 description 17
- 238000012549 training Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/31—Indexing; Data structures therefor; Storage structures
- G06F16/316—Indexing structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3343—Query execution using phonetics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Acoustics & Sound (AREA)
- Probability & Statistics with Applications (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Manipulator (AREA)
Abstract
According to the business guiding method for robot service, the candidate intentions corresponding to the user expression information and the probability corresponding to each candidate intention are obtained through the intention recognition model; and extracting keywords from the user expression information, and acquiring all intention labels corresponding to the keywords through a keyword index model to serve as verification intention. According to the method, the intention recognition model trained by big data is used for directly carrying out intention recognition on the expression information, then the acquired alternative intention is further verified and probability adjusted by combining keywords, finally whether the alternative intention is correct or not is confirmed with a client in sequence according to probability, so that the target intention required by the client is obtained, and efficient and accurate intelligent service guidance is realized.
Description
Technical Field
The present invention relates to the field of intelligent robots, and in particular, to a method, a system, and a storage medium for guiding a service of a robot.
Background
The dialogue system is a man-machine dialogue system based on natural language, and people can use the natural language to perform multiple dialogue interactions with a computer to complete certain tasks, such as inquiring information, acquiring help and the like.
Generally, the dialog system framework is divided into a Task oriented system (Task-oriented Dialogue System) and a Non-Task oriented system (Non-Task-Oriented Dialogue System) according to the Task type of the dialog system. The non-task oriented system is characterized in that: non-object-oriented, generally expected are semantic relevance and asymptotic, low accuracy requirements, and open-ended, and no feedback from the user is required in terms of dialog quality. The non-task oriented system is implemented by a generation method or a search-based manner.
The task oriented system is represented by task dialogues and questions and answers that meet user-specific target needs and require user feedback during training (require user feedback on dialogue quality). The question-answering system focuses on asking a question for an answer, i.e. giving an accurate answer directly from the user's question. The question-answering system is closer to an information retrieval process, which may involve simple context handling. The most fundamental difference between the question-answering system and the task-oriented system is whether a representation of the user's target state needs to be maintained and whether a decision process is required to complete the task. Most current business goal oriented dialog systems remain very field-oriented and rely heavily on artificial architectural features.
The task type multi-wheel dialogue robot has only ten years of development history, and can better solve the multi-wheel task with high certainty at present. However, the current task robot can work normally in an ideal scene, so that the user cannot accurately express the task robot into an act-slot-value triplet under most situations, and the subsequent flow constructed on the basis becomes fragile. Many scholars have proposed various end-to-end research schemes in an attempt to improve the robustness of use of task robots. However, these schemes basically need to train with massive historical dialogue data, and the effect is not verified in a real complex scene.
How to improve the accuracy, intelligence and high efficiency of the service robot is essentially faced with the problem of realizing accurate recognition of the user intention by the robot.
Disclosure of Invention
In order to solve the defect that the robot in the prior art cannot accurately identify the user intention, the invention provides a service guiding method, a system and a storage medium for robot service.
The invention adopts the following technical scheme:
a business guiding method for a robot service, comprising the steps of:
s1, acquiring an intention recognition model and a keyword index model; the intention recognition model is used for screening a preset number of alternative intentions from a plurality of preset intention labels according to the user expression information and the probability corresponding to each alternative intention, wherein the preset number is greater than or equal to 1; the keyword index model is used for detecting each intention label corresponding to the acquired keywords; the intention labels are categorized into non-business classes and business classes;
s2, acquiring user expression information, and acquiring alternative intentions corresponding to the user expression information and probabilities corresponding to the alternative intentions through an intention recognition model;
s3, extracting keywords from the user expression information, and acquiring all intention labels corresponding to the keywords through a keyword index model to serve as verification intention;
s4, judging whether a coincidence item exists in the alternative intention and the verification intention; if yes, adjusting the probability of each alternative intention according to a preset probability adjustment model; if not, directly executing the step S5;
s5, sorting the candidate intentions according to the probability, performing man-machine interaction with the client, and sequentially verifying the candidate intentions until the candidate intentions confirmed by the user are obtained as target intentions.
Preferably, the step S5 specifically includes the following steps:
s51, sequencing each alternative intention according to the sequence from high probability to low probability;
s52, taking the first alternative intention as a current verification tag;
s53, judging whether the current verification tag is a non-business class or a business class;
s54, when the current verification tag is of a non-business type, responding to the client according to a preset call, and returning to the step S2;
s55, when the current verification tag is a service class, generating service verification information for inquiring whether a client needs to transact a corresponding service or not and sending the service verification information to the client;
s56, judging whether the current verification tag passes verification according to the reply content of the client to the service verification information; if yes, executing step S6;
s57, if not, judging whether the alternative intention is completely verified; if yes, sending preset service query information to the user, and returning to the step S2; if not, the next alternative intention is taken as the current verification tag, and then the step S53 is returned.
Preferably, the method further comprises step S0: establishing a mapping relation between the intention labels of the non-business classes and preset call words;
the step S54 specifically includes: and when the current verification tag is a non-business class, calling the caller mapped by the current verification tag to respond to the client.
Preferably, in step S55, the manner of generating the service verification information is as follows: and filling the service name corresponding to the current verification tag into a preset structural statement to generate service verification information.
Preferably, in step S4, the candidate intention with the coincident verification intention is taken as the coincident intention; the probability adjustment model is as follows:
when the number of the overlapped intentions is 1 and the probability of the overlapped intentions is the maximum value in the probabilities corresponding to the alternative intentions, setting the probability of the overlapped intentions to 1, and resetting the probabilities of the rest alternative intentions;
when the number of overlapped intents is 1 and the probability of the overlapped intention is not the maximum value in the probabilities corresponding to the alternative intents, or when the number of overlapped intents is greater than 1; the probability of the overlapped intention is improved, and the probability of the rest alternative intention is adaptively adjusted;
the principle of adaptive adjustment of the probability of the remaining alternative intents is: the sum of the lifted probabilities corresponding to the registered intent and the adjusted probabilities corresponding to the remaining candidate intents is equal to 1.
Preferably, in step S4, after the probability of each candidate intention is adjusted according to the preset probability adjustment model, the candidate intention with the corresponding probability of 0 is deleted.
Preferably, the predetermined number is at least 3.
Preferably, the method further comprises step S6: generating a service execution command according to the target intention, acquiring a feedback result of the execution system on the service execution command and outputting the feedback result to a client;
in the step S2, the user expression information is text information, and is obtained by text conversion of user voices; in step S6, the feedback result is output in a voice manner.
The service guiding system for the robot service comprises a storage unit and a processor, wherein the storage unit is used for storing a computer program, and the processor is used for realizing the service guiding method for the robot service when the computer program is run.
A storage medium for a robot service carrying a computer program which, when executed, implements the service guidance method for a robot service.
The invention has the advantages that:
(1) The intention recognition model trained by big data is used for directly carrying out intention recognition on the expression information, then the acquired alternative intention is further verified and probability adjusted by combining keywords, finally, whether the alternative intention is correct or not is confirmed with a client in sequence according to the probability, so that the target intention required by the client is obtained, and efficient and accurate intelligent service guidance is realized.
(2) The method ensures the accuracy of the finally obtained target intention by confirming the alternative intention through man-machine interaction with the client, and is convenient for providing accurate service on the basis of correctly understanding the client requirement.
(3) And by combining intention recognition and keyword index, efficient screening and efficient recognition of client intention are ensured, and service efficiency is improved.
(4) Through mutual independence of the client intention recognition and execution system, mutual interference of the client intention recognition and service execution is avoided, and high efficiency and accuracy of the client intention recognition and execution system work are further ensured.
(5) The invention provides a guiding method for improving the accuracy, the intelligence and the high efficiency of robot service in a task type dialogue scene.
Drawings
FIG. 1 is a flow chart of a business guidance method for robotic services;
fig. 2 is a flowchart of a method for obtaining a target intention.
Detailed Description
Referring to fig. 1, a business guiding method for a robot service according to the present embodiment includes the following steps.
S1, acquiring an intention recognition model and a keyword index model. The intention recognition model is used for screening a preset number of alternative intentions from a plurality of preset intention labels according to the user expression information and the probability corresponding to each alternative intention, wherein the preset number is greater than or equal to 1. And the keyword index model is used for detecting each intention label corresponding to the acquired keywords. The intent labels are categorized into non-business classes and business classes.
Specifically, in this embodiment, the intention recognition model, the keyword index model, and the intention labels of the service classes are all set according to the service domain.
In the present embodiment, the intention recognition model is obtained by machine learning. Specifically, firstly, collecting user expression information samples aiming at the service field, and labeling a preset number of alternative intentions and the probability of each alternative intention aiming at each sample; then dividing the user expression information sample into a training sample, a test sample and a verification sample, selecting a neural network model to learn the training sample, obtaining an initial intention recognition model, and correcting and training the initial intention recognition model through the verification sample until the obtained result of testing the intention recognition model through the test sample meets the requirement.
Specifically, the sum of probabilities of the respective candidate intents output by the intention recognition model is equal to 1.
In this embodiment, the keyword index model is configured to obtain each intention label corresponding to a keyword according to a preset mapping relationship between a keyword and the intention label. Specifically, in this embodiment, one keyword may correspond to multiple intention labels, and one intention label may also correspond to multiple keywords.
S2, acquiring user expression information, and acquiring alternative intentions corresponding to the user expression information and probabilities corresponding to the alternative intentions through an intention recognition model. The sum of the probabilities of the alternative intents is equal to 1.
In particular, the user expression information may be obtained by text conversion of the user's speech.
And S3, extracting keywords from the user expression information, and acquiring all intention labels corresponding to the keywords through a keyword index model to serve as verification intention.
S4, judging whether a coincidence item exists in the alternative intention and the verification intention; if yes, adjusting the probability of each alternative intention according to a preset probability adjustment model; if not, step S5 is directly performed.
Specifically, in the present embodiment, the candidate intention having the coincident verification intention is taken as the coincident intention; the probability adjustment model is as follows:
when the number of the overlapped intentions is 1 and the probability of the overlapped intentions is the maximum value in the probabilities corresponding to the alternative intentions, setting the probability of the overlapped intentions to 1, and resetting the probabilities of the rest alternative intentions;
when the number of overlapped intents is 1 and the probability of the overlapped intention is not the maximum value in the probabilities corresponding to the alternative intents, or when the number of overlapped intents is greater than 1; the probability of the overlapped intention is improved, and the probability of the rest alternative intention is adaptively adjusted;
the principle of adaptive adjustment of the probability of the remaining alternative intents is: the sum of the lifted probabilities corresponding to the registered intent and the adjusted probabilities corresponding to the remaining candidate intents is equal to 1.
That is, after the probability adjustment, the sum of the probability of the overlapped intention and the probability of the remaining candidate intention is always equal to 1.
Assuming that the number of candidate probabilities is 3 in an embodiment, in this embodiment, the probability of the overlapped intention is Pa1, and the probabilities of the remaining two candidate intentions are Pa2 and Pa3, respectively, and Pa1+pa2+pa3 = 1; assuming that the first probability increment is 0.4, the probability of the overlapped intention and the probability of the remaining two alternative intentions are respectively recorded as: pa1', pa2', and Pa3', then
Assuming that in an embodiment, the number of false device selection probabilities is 3, wherein two alternative intentions are coincident intentions, the probabilities thereof are respectively marked as Pb1 and Pb2, the probability of the remaining one alternative intention is Pb3, and the second probability increment is set to be 0.2; the probability of the overlapped intention and the probability of the remaining two alternative intentions are respectively recorded as: pb1', pb2' and Pb3', then
S5, sorting the candidate intentions according to the probability, performing man-machine interaction with the client, and sequentially verifying the candidate intentions until the candidate intentions confirmed by the user are obtained as target intentions.
In this way, in this embodiment, first, the intention recognition model trained by big data is used to directly recognize the intention of the expression information, then the keyword is combined to further verify and probability adjust the obtained candidate intention, and finally, according to the probability, the user is sequentially confirmed whether the candidate intention is correct or not, so as to obtain the target intention required by the user, and accurate interpretation and extraction of the user intention are realized.
In this way, in the embodiment, the accuracy of the finally obtained target intention is ensured by the confirmation of the alternative intention through the man-machine interaction with the client, so that accurate service can be conveniently provided on the basis of correctly understanding the client requirement. Meanwhile, by combining intention recognition and keyword index, efficient screening and efficient recognition of client intention are guaranteed, and service efficiency is improved.
In the intention recognition model in this embodiment, the number of intention labels output by the intention recognition model is at least 3, that is, the preset number is at least 3. Therefore, the situation of missing the user demands is avoided by outputting a plurality of intention labels, repeated inquiry is reduced, and efficient extraction of the user demands is ensured.
In the specific implementation, in step S4, after the probability of each candidate intention is adjusted according to the preset probability adjustment model, the candidate intention with the corresponding probability of 0 is deleted. Therefore, verification of meaningless intention labels can be avoided, redundant work is reduced, and service efficiency is further improved.
S6, generating a service execution command according to the target intention, acquiring a feedback result of the execution system on the service execution command, and outputting the feedback result to the client. In this way, in this embodiment, through mutual independence of the client intention recognition and execution system, mutual interference between client intention recognition and service execution is avoided, and further high efficiency and accuracy of the client intention recognition and execution system work are ensured.
Specifically, in step S6, the feedback result is output in a voice manner. Therefore, the user expression information is obtained through text conversion of the client voice in the step S2, natural language answering between the robot and the user is achieved, the use of the user is further facilitated, and the working efficiency is improved.
Referring to fig. 2, in the present embodiment, the acquisition of the target intention in step S5 specifically includes the following steps.
S51, sequencing the candidate intents according to the sequence from high probability to low probability.
S52, taking the first alternative intention as a current verification tag.
S53, judging whether the current verification tag is a non-service type or a service type.
And S54, when the current verification tag is of a non-business type, responding to the client according to a preset call, and returning to the step S2.
In the implementation, a mapping relationship between the intent label of the non-service class and the preset caller can be established in advance, so that in the step S54, the mapped caller is directly called to respond to the client according to the current verification label.
Assuming that the current verification tag is an intention tag A1 used for representing that user expression information is ' hello ', ' hi ' and the like in a non-business class, and in a preset mapping relation, a call sign mapped by the intention tag A1 is ' hello ', ' in; then in this step S4, the robot replies to the client "your good", while the execution step returns to S2.
In particular implementations, the intent label may be set to a label for representing certain information, e.g., an nth intent label of a non-business class may be represented as An, and An nth intent label of a business class may be represented as Bn; the intent labels for expressing that the user expresses information as "hello", "hi" and the like in the non-business class can be set as "hello" directly, and the intent labels of the business class can be set as the business names directly.
And S55, when the current verification tag is a service class, generating service verification information for inquiring whether the client needs to transact the corresponding service or not and sending the service verification information to the client.
Specifically, in this embodiment, the service verification information may be generated by filling the service name corresponding to the current verification tag into a preset structural statement. For example, the preset structural statement is "do you transact business? "is the service name corresponding to the current verification tag C, the service verification information is" is you to transact the C service? ".
S56, judging whether the current verification tag passes verification according to the reply content of the client to the service verification information; if yes, step S6 is performed.
Specifically, in this step, whether the semantics of the reply is affirmative or negative is determined directly according to the reply content of the client, and if the semantics indicate affirmative, the current verification tag is determined to pass the verification.
S57, if not, judging whether the alternative intention is completely verified; if yes, sending preset service query information to the user, and returning to the step S2; if not, the next alternative intention is taken as the current verification tag, and then the step S53 is returned.
Specifically, in this step, whether the current verification tag is at the last in the sorted candidate intention sorting queue is determined, and if yes, it is determined that all the candidate intents have been verified; otherwise, the alternative intention is not verified.
In this embodiment, the service inquiry information may be specifically set as "please ask you what service to do", "please ask me what can help you," please ask what you need to be ", and so on.
The present embodiment also provides a service guiding system for a robot service, including a storage unit and a processor, where the storage unit is configured to store a computer program, and the processor is configured to implement the service guiding method for a robot service provided in the present embodiment when the computer program is executed.
The present embodiment also proposes a storage medium for robot service, the storage medium carrying a computer program which, when executed, implements the service guidance method for robot service provided in the present embodiment.
The invention is further explained below with reference to specific examples.
Example 1
In this embodiment, as an example of a robot applied to an electric power business hall,
in this embodiment, the acquisition method of the intent recognition model includes the following steps:
s11, under an actual scene, collecting a certain number of customer questions and converting the customer questions into user expression information, adding 3 labels to each collected customer question, wherein the labels are intent labels and corresponding probabilities of the customer questions, and taking the certain number of expression information with the labels as sample data of a machine learning model. Specifically, each piece of sample data may be denoted as { Q (A1, pa 1); (A2, pa 2); (A3, pa 3) }, wherein Q represents user expression information after user problem conversion, A1, A2, A3 represent labeled intention labels, pa1, pa2, and Pa3 represent probabilities corresponding to A1, A2, A3, respectively, and Pa1+pa2+pa 3=1.
S12, dividing the sample data into a training set, a testing set and a verification set according to the proportion.
And S13, training the machine learning model based on the training set.
S14, verifying and testing the trained machine learning model based on the test set and the verification set, enabling the trained machine learning model to reach a preset effect through adjusting parameters, and then using the model as an intention recognition model.
Specifically, the service class intention label in this embodiment includes: distributed power supply, commercial power supply change resident power supply, resident power supply change commercial power supply, purchase of electric automobile charging card, resident electric charge payment, commercial electric charge payment, online business handling, resident application ammeter, resident replacement time sharing, resident ammeter capacity increasing, resident application household separating, resident sales household, resident passing household, resident verification ammeter, commercial application ammeter, commercial ammeter capacity increasing, commercial power meter volume reducing, commercial address transferring, commercial sales household, commercial passing household renaming, commercial replacement time sharing, high-low voltage definition, high-voltage new power supply, high-voltage capacity increasing, high-voltage volume reducing, high-voltage passing household renaming, high-voltage sales household, resident time-of-use electricity price, step electricity price and electricity price standard, power supply report guide, functions of national network security electric power public number, platform functions of electricity e-treasure APP, other time-of-use electricity price, resident power service, distributed photovoltaic power generation.
The non-business class labels include: you are good.
In particular, customer questions may be expressed in natural language. The user expression information is text information after the text conversion of the client problem.
In this embodiment, a mapping relationship between the keyword and the intention labels is also established, so as to obtain each intention label corresponding to the keyword. Specifically, in this embodiment, one keyword may correspond to multiple intention labels, and one intention label may also correspond to multiple keywords.
The method in this embodiment is described below in connection with a specific scenario.
The first step: obtaining user expression information 'hello', calling a corresponding intention label-associated call-back client 'hello', and marking the end of the round of dialogue.
And a second step of: obtaining user expression information of 'frequent trips of me' and inputting 'frequent trips of me' into an intention recognition model to obtain: labels (Bi, 0.99998), (Bj, 0.99999 e-05), (Bk, 0.99999 e-05); in brackets, the left side is the intention label, and the right side is the probability corresponding to the intention label. Specifically, in this embodiment, bi, bj, bk respectively represent: the residential electricity meter increases capacity, handles business on line and changes residential electricity into business electricity, and Bi, bj and Bk are all taken as alternative intents.
And a third step of: extracting a keyword ' trip ' from user expression information ' I ' family frequent trip ', and acquiring an intention label corresponding to the keyword as ' resident ammeter capacity increase ' as a verification intention through a keyword index model based on the mapping relation between the keyword and the intention label.
Fourth step: and judging that the verification intention coincides with one of the candidate intents with the highest probability, and therefore, adjusting the probabilities corresponding to the three candidate intents, namely the residential electricity meter capacity increasing, the online business handling and the residential electricity utilization to business electricity utilization, to be 1, 0 and 0 respectively.
Fifth step: asking the customer if you need to transact the residential electricity meter capacity-increasing service.
Sixth step: and obtaining the reply of the client to the question of the previous step, and judging whether the reply is affirmative or negative.
Specifically, in this embodiment, the reply of the user may be determined according to the semantic recognition, if the user reply semantic is determined to be "yes", the execution system is notified to transact the residential electric meter capacity-increasing service for the client, and after the transacting is completed, the transacting end information is fed back to the user.
If the user reply semantic is judged to be 'not', inquiring 'please ask you what to transact' to the client, and judging the alternative intention and the verification intention from the new according to the client reply.
Specifically, in the fourth step of the embodiment, since the last two alternatives are "online transacting business" and "residential electricity-to-business electricity" adjusted probability is 0. Therefore, in the sixth step, even if the user denies to transact the "resident electric meter capacity increasing" service, the user is not asked whether to transact the "online transacting service" and the "resident electric power to business electric power.
In the sixth step, when the client denies that the resident ammeter capacity-increasing service is to be processed, the robot inquires whether the client needs to process the online service, if the client is affirmative, the executing system is informed to process the online service for the client, and after the processing is finished, the processing finishing information is fed back to the user; if the client is negative, the robot inquires whether the user needs to transact domestic electricity to change business electricity service, if the user is positive, the executing system is informed to transact domestic electricity to change business electricity service for the client, and after transacting is finished, the transacting finishing information is fed back to the user; if the user is negative, the client is queried for 'please ask you for what business to transact', and according to the client reply, the alternative intention and the verification intention are newly judged.
The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention, and any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the present invention.
Claims (10)
1. A business guiding method for a robot service, comprising the steps of:
s1, acquiring an intention recognition model and a keyword index model; the intention recognition model is used for screening a preset number of alternative intentions from a plurality of preset intention labels according to the user expression information and the probability corresponding to each alternative intention, wherein the preset number is greater than or equal to 1; the keyword index model is used for detecting each intention label corresponding to the acquired keywords; the intention labels are categorized into non-business classes and business classes;
s2, acquiring user expression information, and acquiring alternative intentions corresponding to the user expression information and probabilities corresponding to the alternative intentions through the intention recognition model;
s3, extracting keywords from the user expression information, and acquiring all intention labels corresponding to the keywords through the keyword index model to serve as verification intention;
s4, judging whether a coincidence item exists in the alternative intention and the verification intention; if yes, adjusting the probability of each alternative intention according to a preset probability adjustment model; if not, directly executing the step S5;
s5, sorting the candidate intentions according to the probability, performing man-machine interaction with the client, and sequentially verifying the candidate intentions until the candidate intentions confirmed by the user are obtained as target intentions.
2. The traffic guidance method for robot service according to claim 1, wherein the step S5 specifically comprises the steps of:
s51, sequencing each alternative intention according to the sequence from high probability to low probability;
s52, taking the first alternative intention as a current verification tag;
s53, judging whether the current verification tag is a non-business class or a business class;
s54, when the current verification tag is of a non-business type, responding to the client according to a preset call, and returning to the step S2;
s55, when the current verification tag is a service class, generating service verification information for inquiring whether a client needs to transact a corresponding service or not and sending the service verification information to the client;
s56, judging whether the current verification tag passes verification according to the reply content of the client to the service verification information; if yes, executing step S6;
s57, if not, judging whether the alternative intention is completely verified; if yes, sending preset service query information to the user, and returning to the step S2; if not, the next alternative intention is taken as the current verification tag, and then the step S53 is returned.
3. The traffic guidance method for a robot service according to claim 2, further comprising the step of S0: establishing a mapping relation between the intention labels of the non-business classes and preset call words;
the step S54 specifically includes: and when the current verification tag is a non-business class, calling the caller mapped by the current verification tag to respond to the client.
4. The traffic guidance method for robot service according to claim 3, wherein in step S55, the traffic verification information is generated by: and filling the service name corresponding to the current verification tag into a preset structural statement to generate service verification information.
5. The traffic guidance method for a robot service according to claim 1, wherein in step S4, an alternative intention in which there is a coincident verification intention is taken as a coincident intention; the probability adjustment model is as follows:
when the number of the overlapped intentions is 1 and the probability of the overlapped intentions is the maximum value in the probabilities corresponding to the alternative intentions, setting the probability of the overlapped intentions to 1, and resetting the probabilities of the rest alternative intentions;
when the number of overlapped intents is 1 and the probability of the overlapped intention is not the maximum value in the probabilities corresponding to the alternative intents, or when the number of overlapped intents is greater than 1; the probability of the overlapped intention is improved, and the probability of the rest alternative intention is adaptively adjusted;
the principle of adaptive adjustment of the probability of the remaining alternative intents is: the sum of the lifted probabilities corresponding to the registered intent and the adjusted probabilities corresponding to the remaining candidate intents is equal to 1.
6. The traffic guidance method for a robot service according to claim 5, wherein in step S4, after adjusting the probability of each candidate intention according to a preset probability adjustment model, the candidate intention having the corresponding probability of 0 is deleted.
7. The traffic guidance method for robot service according to claim 1, wherein the preset number is at least 3.
8. The traffic guidance method for a robot service according to claim 1, further comprising the step of S6: generating a service execution command according to the target intention, acquiring a feedback result of the execution system on the service execution command and outputting the feedback result to a client;
in the step S2, the user expression information is text information, and is obtained by text conversion of user voices; in step S6, the feedback result is output in a voice manner.
9. A traffic guidance system for a robot service, characterized by comprising a storage unit for storing a computer program and a processor for implementing the traffic guidance method for a robot service according to any of claims 1 to 8 when the computer program is run.
10. A storage medium for a robot service, characterized in that a computer program is carried, which computer program, when being executed, implements the traffic guidance method for a robot service according to any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010981046.1A CN112256824B (en) | 2020-09-17 | 2020-09-17 | Service guiding method, system and storage medium for robot service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010981046.1A CN112256824B (en) | 2020-09-17 | 2020-09-17 | Service guiding method, system and storage medium for robot service |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112256824A CN112256824A (en) | 2021-01-22 |
| CN112256824B true CN112256824B (en) | 2023-10-20 |
Family
ID=74231824
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010981046.1A Active CN112256824B (en) | 2020-09-17 | 2020-09-17 | Service guiding method, system and storage medium for robot service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112256824B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108073600A (en) * | 2016-11-11 | 2018-05-25 | 阿里巴巴集团控股有限公司 | A kind of intelligent answer exchange method, device and electronic equipment |
| WO2020006835A1 (en) * | 2018-07-03 | 2020-01-09 | 平安科技（深圳）有限公司 | Customer service method, apparatus, and device for engaging in multiple rounds of question and answer, and storage medium |
| CN111368085A (en) * | 2020-03-05 | 2020-07-03 | 北京明略软件***有限公司 | Recognition method and device of conversation intention, electronic equipment and storage medium |
| KR20200092455A (en) * | 2019-01-04 | 2020-08-04 | 주식회사 케이티 | Server, method and computer program for predicting intention of user |
-
2020
- 2020-09-17 CN CN202010981046.1A patent/CN112256824B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108073600A (en) * | 2016-11-11 | 2018-05-25 | 阿里巴巴集团控股有限公司 | A kind of intelligent answer exchange method, device and electronic equipment |
| WO2020006835A1 (en) * | 2018-07-03 | 2020-01-09 | 平安科技（深圳）有限公司 | Customer service method, apparatus, and device for engaging in multiple rounds of question and answer, and storage medium |
| KR20200092455A (en) * | 2019-01-04 | 2020-08-04 | 주식회사 케이티 | Server, method and computer program for predicting intention of user |
| CN111368085A (en) * | 2020-03-05 | 2020-07-03 | 北京明略软件***有限公司 | Recognition method and device of conversation intention, electronic equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 融合敏感词规则和字符级RCNN模型的用户意图识别;王冲;张虎;王鑫;刘全明;;计算机应用与软件(03);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112256824A (en) | 2021-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108073600B (en) | Intelligent question-answer interaction method and device and electronic equipment | |
| CN111212190B (en) | Conversation management method, device and system based on conversation strategy management | |
| CN110727779A (en) | Question-answering method and system based on multi-model fusion | |
| US7412383B1 (en) | Reducing time for annotating speech data to develop a dialog application | |
| CN110019742B (en) | Method and device for processing information | |
| CN104462600A (en) | Method and device for achieving automatic classification of calling reasons | |
| CA2531455A1 (en) | Improving error prediction in spoken dialog systems | |
| CN112732871B (en) | Multi-label classification method for acquiring client intention labels through robot induction | |
| CN111858854B (en) | Question-answer matching method and relevant device based on historical dialogue information | |
| CN111191000A (en) | Dialog management method, device and system of intelligent voice robot | |
| CN114297987B (en) | Document information extraction method and system based on text classification and reading understanding | |
| CN110717027B (en) | Multi-round intelligent question-answering method, system, controller and medium | |
| CN113064980A (en) | Intelligent question and answer method and device, computer equipment and storage medium | |
| CN112527986A (en) | Multi-round dialog text generation method, device, equipment and storage medium | |
| CN113297365A (en) | User intention determination method, device, equipment and storage medium | |
| CN113591463B (en) | Intention recognition method, device, electronic equipment and storage medium | |
| CN112069833B (en) | Log analysis method, log analysis device and electronic equipment | |
| CN110674276A (en) | Robot self-learning method, robot terminal, device and readable storage medium | |
| CN112256824B (en) | Service guiding method, system and storage medium for robot service | |
| CN112860873A (en) | Intelligent response method, device and storage medium | |
| CN113722452B (en) | Semantic-based rapid knowledge hit method and device in question-answering system | |
| CN113033160B (en) | Method and device for classifying intention of dialogue and method for generating intention classification model | |
| CN114116975A (en) | Multi-intention identification method and system | |
| CN113868396A (en) | Task intelligent dialogue construction method and system based on knowledge graph | |
| CN111949776A (en) | Method and device for evaluating user tag and electronic equipment |
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 |