CN111429157A - Method, device and equipment for evaluating and processing complaint work order and storage medium - Google Patents
Method, device and equipment for evaluating and processing complaint work order and storage medium Download PDFInfo
- Publication number
- CN111429157A CN111429157A CN202010227729.8A CN202010227729A CN111429157A CN 111429157 A CN111429157 A CN 111429157A CN 202010227729 A CN202010227729 A CN 202010227729A CN 111429157 A CN111429157 A CN 111429157A
- Authority
- CN
- China
- Prior art keywords
- complaint
- user
- work order
- candidate
- preset
- 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
- 238000012545 processing Methods 0.000 title claims abstract description 124
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000003860 storage Methods 0.000 title claims abstract description 21
- 238000011156 evaluation Methods 0.000 claims abstract description 56
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 230000002194 synthesizing effect Effects 0.000 claims abstract description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 31
- 238000013145 classification model Methods 0.000 claims description 15
- 230000015654 memory Effects 0.000 claims description 15
- 230000000977 initiatory effect Effects 0.000 claims description 14
- 238000012216 screening Methods 0.000 claims description 14
- 238000005516 engineering process Methods 0.000 claims description 13
- 238000004140 cleaning Methods 0.000 claims description 12
- 230000002457 bidirectional effect Effects 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 6
- 238000003672 processing method Methods 0.000 claims description 6
- 238000012163 sequencing technique Methods 0.000 claims description 5
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 238000003786 synthesis reaction Methods 0.000 claims description 2
- 238000004590 computer program Methods 0.000 claims 1
- 230000008569 process Effects 0.000 description 11
- 238000007726 management method Methods 0.000 description 9
- 238000012549 training Methods 0.000 description 8
- 239000013598 vector Substances 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000003909 pattern recognition Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 3
- 239000002131 composite material Substances 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000009223 counseling Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
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
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
- G06Q30/016—After-sales
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
Abstract
The invention relates to the technical field of logistics transportation, and discloses a method, a device, equipment and a storage medium for evaluating and processing a complaint work order, which are used for improving the evaluation and processing efficiency of the complaint work order and improving the finishing rate of the complaint work order. The method comprises the following steps: when a service system receives a complaint work order initiated by a user, reading order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order; synthesizing the processing result into initial voice based on the deep neural network; sending the initial voice to a target terminal; acquiring reply voice fed back by a target terminal within a preset duration range to obtain a reply text; performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user; and calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through the robot system.
Description
Technical Field
The invention relates to the technical field of logistics, in particular to a method, a device, equipment and a storage medium for evaluating and processing a complaint work order.
Background
At present, online shopping is part of life of most people, and articles purchased on the internet are transported in the form of express mails, but the express mails may be lost in the transportation process.
In the existing scheme, a customer sending an express sends a complaint through a channel provided by an express enterprise after the express has a problem, and after the express enterprise processes and finishes a work order in time, because of limited manpower, the customer needs to send a short message to obtain the evaluation of the customer on the complaint processing result, so that the customer cannot receive the complaint, and the complaint order cannot be finished in time due to the fact that the complaint cannot be finished in time because the short message cannot be replied.
Disclosure of Invention
The invention provides an evaluation processing method, device, equipment and storage medium for a complaint work order, which are used for initiating a call to a user through synthesized intelligent voice, quickly acquiring the evaluation of the processing result of the user on the initiated complaint work order, improving the evaluation processing efficiency of the complaint work order and improving the finishing rate of the complaint work order.
The invention provides a method for evaluating and processing a complaint work order, which comprises the following steps: when a service system receives a complaint work order initiated by a user, reading order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order; synthesizing the processing result into initial voice based on a deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems; initiating a voice conversation to a target terminal of the complaint user according to the contact way of the complaint user, and sending the initial voice to the target terminal; acquiring reply voice fed back by the target terminal within a preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text; performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user; and calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through a robot system.
Optionally, in a first implementation manner of the first aspect of the present invention, the performing intent recognition on the reply text by using a natural language understanding engine to obtain a real intent of the complaint user includes: performing intention recognition on the reply text through a natural language understanding engine to obtain a plurality of first candidate intentions; performing semantic recognition on the reply text based on a semantic slot dialogue model to obtain a plurality of second candidate intents; and screening the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the real intentions of the complaint users.
Optionally, in a second implementation manner of the first aspect of the present invention, the performing intent recognition on the reply text by using a natural language understanding engine to obtain a plurality of first candidate intents includes: inputting the reply text into a natural language understanding engine, and analyzing the reply text to obtain a plurality of candidate words; extracting high-frequency phrases from the candidate words to obtain a plurality of high-frequency phrases, wherein the high-frequency phrases are the candidate words with the occurrence frequency exceeding a threshold value; judging whether the high-frequency phrases are matched with a preset phrase library or not; if the high-frequency phrases are matched with a preset phrase library, directly matching a plurality of preset intentions, and determining the preset intentions as a plurality of first candidate intentions; and if the high-frequency phrases are not matched with a preset phrase library, respectively calling a plurality of preset identification models to identify the reply text to obtain a plurality of first candidate intents.
Optionally, in a third implementation manner of the first aspect of the present invention, the performing semantic recognition on the reply text based on a semantic slot dialogue model to obtain a plurality of second candidate intents includes: identifying the field of the reply text to obtain a target field; performing intention recognition on the reply text through a preset recognition algorithm to obtain a plurality of intentions to be selected; filling slots in the reply text through a semantic slot dialogue model to obtain a plurality of slot attributes; and screening the plurality of intentions to be selected according to the target field and the plurality of slot attributes to obtain a plurality of second candidate intentions.
Optionally, in a fourth implementation manner of the first aspect of the present invention, the screening the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the true intention of the complaint user includes: fusing the plurality of first candidate intentions and the plurality of second candidate intentions to obtain a plurality of candidate intentions to be evaluated; calling a plurality of preset identification models to score each candidate intention to be evaluated respectively to obtain a scoring result corresponding to each candidate intention to be evaluated, wherein the scoring result comprises a plurality of target scores and different target scores corresponding to different preset identification models; acquiring a weight value of each preset identification model; calculating according to the weight value of each preset identification model and the grading result corresponding to each candidate intention to be evaluated, and determining the comprehensive score of each candidate intention to be evaluated to obtain a plurality of comprehensive scores; and sequencing the multiple comprehensive scores, and determining the candidate intention to be evaluated with the highest comprehensive score as the real intention of the complaint user.
Optionally, in a fifth implementation manner of the first aspect of the present invention, when the service system receives a complaint work order initiated by a user, the method for evaluating and processing the complaint work order further includes: the method comprises the steps of obtaining a user speech technology in the express logistics field, and obtaining related corpora in the user speech technology; performing data cleaning on the linguistic data related to the user speech technology to obtain the cleaned linguistic data; and generating a plurality of classification models based on a bidirectional long-short term memory algorithm and the cleaned linguistic data, and selecting the classification model with the highest performance score to determine as a natural language understanding engine.
Optionally, in a sixth implementation manner of the first aspect of the present invention, when a complaint work order initiated by a user is received, reading order information corresponding to the complaint work order, where the order information includes a client number of the complaint user and a processing result of the complaint order, and before the method for evaluating and processing the complaint work order further includes: and carrying out communication connection between the robot system and the service system based on the session initiation protocol.
A second aspect of the present invention provides an evaluation processing apparatus for a complaint work order, including: the service system comprises a reading module, a processing module and a processing module, wherein the reading module is used for reading order information corresponding to a complaint work order when the service system receives the complaint work order initiated by a user, and the order information comprises a contact way of the complaint user and a processing result of the complaint order; a synthesis module, configured to synthesize the processing result into an initial voice based on a deep neural network, where the initial voice is used to indicate the processing result, and the initial voice includes a plurality of preset problems; the conversation module is used for initiating a voice conversation to a target terminal of the complaint user according to the contact information of the complaint user and sending the initial voice to the target terminal; the first acquisition module is used for acquiring reply voice fed back by the target terminal within a preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text; the recognition module is used for carrying out intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user; and the matching broadcasting module is used for calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through a robot system.
Optionally, in a first implementation manner of the second aspect of the present invention, the identification module includes: the first identification unit is used for identifying the intentions of the reply text through a natural language understanding engine to obtain a plurality of first candidate intentions; the second recognition unit is used for carrying out semantic recognition on the reply text based on a semantic slot dialogue model to obtain a plurality of second candidate intents; and the screening unit is used for screening the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the real intentions of the complaint users.
Optionally, in a second implementation manner of the second aspect of the present invention, the first identifying unit is specifically configured to: inputting the reply text into a natural language understanding engine, and analyzing the reply text to obtain a plurality of candidate words; extracting high-frequency phrases from the candidate words to obtain a plurality of high-frequency phrases, wherein the high-frequency phrases are the candidate words with the occurrence frequency exceeding a threshold value; judging whether the high-frequency phrases are matched with a preset phrase library or not; if the high-frequency phrases are matched with a preset phrase library, directly matching a plurality of preset intentions, and determining the preset intentions as a plurality of first candidate intentions; and if the high-frequency phrases are not matched with a preset phrase library, respectively calling a plurality of preset identification models to identify the reply text to obtain a plurality of first candidate intents.
Optionally, in a third implementation manner of the second aspect of the present invention, the second identifying unit is specifically configured to: identifying the field of the reply text to obtain a target field; performing intention recognition on the reply text through a preset recognition algorithm to obtain a plurality of intentions to be selected; filling slots in the reply text through a semantic slot dialogue model to obtain a plurality of slot attributes; and screening the plurality of intentions to be selected according to the target field and the plurality of slot attributes to obtain a plurality of second candidate intentions.
Optionally, in a fourth implementation manner of the second aspect of the present invention, the screening unit is specifically configured to: fusing the plurality of first candidate intentions and the plurality of second candidate intentions to obtain a plurality of candidate intentions to be evaluated; calling a plurality of preset identification models to score each candidate intention to be evaluated respectively to obtain a scoring result corresponding to each candidate intention to be evaluated, wherein the scoring result comprises a plurality of target scores and different target scores corresponding to different preset identification models; acquiring a weight value of each preset identification model; calculating according to the weight value of each preset identification model and the grading result corresponding to each candidate intention to be evaluated, and determining the comprehensive score of each candidate intention to be evaluated to obtain a plurality of comprehensive scores; and sequencing the multiple comprehensive scores, and determining the candidate intention to be evaluated with the highest comprehensive score as the real intention of the complaint user.
Optionally, in a fifth implementation manner of the second aspect of the present invention, the apparatus for evaluating a complaint work order further includes: the second acquisition module is used for acquiring the user dialect in the express logistics field and acquiring the related linguistic data in the user dialect; the data cleaning module is used for cleaning the data of the linguistic data related to the user speech technology to obtain the cleaned linguistic data; and the generation determining module is used for generating a plurality of classification models based on the bidirectional long-short term memory algorithm and the cleaned linguistic data, and selecting the classification model with the highest performance score to determine as the natural language understanding engine.
Optionally, in a sixth implementation manner of the second aspect of the present invention, the apparatus for evaluating a complaint work order further includes: and the connection module is used for carrying out communication connection on the robot system and the service system based on the session initial protocol.
A third aspect of the present invention provides an evaluation processing apparatus for a complaint work order, including: a memory having instructions stored therein and at least one processor, the memory and the at least one processor interconnected by a line; the at least one processor calls the instructions in the memory to enable the evaluation processing equipment of the complaint work order to execute the evaluation processing method of the complaint work order.
A fourth aspect of the present invention provides a computer-readable storage medium having stored therein instructions, which when run on a computer, cause the computer to execute the above-described evaluation processing method of a complaint work order.
In the technical scheme provided by the invention, when a service system receives a complaint work order initiated by a user, the service system reads order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order; synthesizing the processing result into initial voice based on the deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems; initiating a voice conversation to a target terminal of the complaint user according to the contact way of the complaint user, and sending initial voice to the target terminal; acquiring reply voice fed back by a target terminal within a preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text; performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user; and calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through the robot system. In the embodiment of the invention, the call is initiated to the user through the synthesized intelligent voice, the evaluation of the processing result of the initiated complaint work order by the user is rapidly obtained, and the voice broadcasting is carried out according to the evaluation, so that the evaluation processing efficiency of the complaint work order is improved, and the finishing rate of the complaint work order is improved.
Drawings
FIG. 1 is a schematic diagram of an embodiment of an evaluation processing method for complaint work orders in the embodiment of the invention;
FIG. 2 is a schematic diagram of another embodiment of the method for evaluating and processing complaint work orders according to the embodiment of the invention;
FIG. 3 is a schematic diagram of an embodiment of an evaluation processing apparatus for complaint work orders according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of another embodiment of the evaluation processing device for complaint work orders in the embodiment of the invention;
FIG. 5 is a schematic diagram of an embodiment of an evaluation processing apparatus for complaint work orders in the embodiment of the present invention.
Detailed Description
The embodiment of the invention provides an evaluation processing method, device, equipment and storage medium for a complaint work order, which are used for initiating a call to a user through synthesized intelligent voice, rapidly acquiring the evaluation of the processing result of the initiated complaint work order by the user, improving the evaluation processing efficiency of the complaint work order and improving the finishing rate of the complaint work order.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It will be appreciated that the data so used may be interchanged under appropriate circumstances such that the embodiments described herein may be practiced otherwise than as specifically illustrated or described herein. Furthermore, the terms "comprises," "comprising," or "having," and any variations thereof, are intended to cover non-exclusive inclusions, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
For convenience of understanding, a specific flow of an embodiment of the present invention is described below, and referring to fig. 1, an embodiment of the method for evaluating and processing a complaint work order in an embodiment of the present invention includes:
101. when the service system receives a complaint work order initiated by a user, reading order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order.
When the service system receives a complaint work order initiated by a user, the server reads order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order. The contact information of the complaint user can be a mobile phone number registered or bound by the complaint user or a client number bound by the complaint user, as long as a target terminal corresponding to the contact information of the complaint user can receive a voice session initiated by the robot system, and the specific point is not limited herein.
The processing result of the complaint order comprises a plurality of states of finished, processing, waiting to be processed and the like, and the estimated processing time length or the estimated finishing date and the like can be provided by combining the states of the complaint order. It should be noted that the processing result of the complaint order may include the current processing state, complaint processing opinions, complaint processing measures, and the like.
It should be understood that the execution subject of the present invention may be an evaluation processing device of a complaint work order, and may also be a terminal or a server, which is not limited herein. The embodiment of the present invention is described by taking a server as an execution subject.
102. And synthesizing the processing result into initial voice based on the deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems.
And the server synthesizes the processing result into initial voice based on the deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems. The processing results of a plurality of preset questions and complaint orders are converted into voice from characters through a preset deep neural network algorithm, and then initial voice is obtained.
103. And initiating a voice conversation to a target terminal of the complaint user according to the contact way of the complaint user, and sending the initial voice to the target terminal.
The server initiates a voice conversation to a target terminal of the complaint user according to the contact information of the complaint user, and sends the initial voice to the target terminal. If the contact way of the complaint user is a mobile phone number registered or bound by the complaint user, the server calls the mobile phone number and sends the pre-generated initial voice to a mobile phone terminal of the complaint user; if the contact information of the complaint user is the client number bound by the complaint user, the server initiates a voice call to the client number and sends the pre-generated initial voice to the client of the complaint user.
For example, several questions that are preset may be included in the initial speech, such as question 1: "do you like Zhao lady/mr. honoring, i am a rhyme express company to serve a minor rhyme for customer service, about the problem of food express feedbacks you give back 1 month 1 day 2020, we have already dealt with feedback from the network, do you want and you check it, ask you for your convenience now? ", or problem 2: "you feed back the single number mantissa: 3037 (last four variables) shows a question of sign-in but you have not received. Asking you to receive the express now? ", or problem 3: asking for which side you feel dissatisfied? Such as processing results, timeliness, service attitudes, etc., do you describe in trouble with describing them exactly once? ".
104. And acquiring the reply voice fed back by the target terminal within the preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text.
The server obtains the reply voice fed back by the target terminal within the preset duration range, and the reply voice is converted by calling a preset voice recognition algorithm to obtain a reply text.
For example, if the complaint user answers question 1 in step 103, the resulting reply speech may be: "busy", "inconvenient", "convenient" or "change time"; the reply voice for question 2 may be: "none", "there", or "received"; the reply voice for question 3 may be: "slow delivery," poor attitude, "or" satisfied.
It should be noted that the server converts the speech into words through an Automatic Speech Recognition (ASR) algorithm to obtain the reply text, wherein the ASR algorithm takes the speech as a research object and allows the machine to automatically recognize and understand the speech dictated by human through speech signal processing and pattern recognition. Speech recognition is a technique that allows a machine to convert speech signals into corresponding text or commands through a recognition and understanding process. The speech recognition is a very extensive cross discipline, and has a very close relationship with the disciplines of acoustics, phonetics, linguistics, information theory, pattern recognition theory, neurobiology and the like, and details are not repeated here.
105. And performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user.
The server performs intention recognition on the reply text through a natural language understanding engine to obtain a plurality of first candidate intentions; the server carries out semantic recognition on the reply text based on the semantic slot conversation model to obtain a plurality of second candidate intents; and the server screens the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the real intentions of the complaint users.
106. And calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through the robot system.
And the server calls a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and the target answer is broadcasted to the client of the complaint user through the robot system. For example, we can get the matching target answer based on the actual intention of the complaint user. For example, if the actual intent of the complaint user is that no express delivery was received, the answer dialog may be:
step 1: "do you good, XXX women/mr. i am a minor rhyme for customer service of a rhyme express company, about a question about a certain express item fed back by date X of year X, the feedback of a network of us has been processed, and do you want to ask you for convenience with you check it? "
Substep 1 a: recognizing "busy/inconvenient/time change" then reply: "good, get a lot of apology bother you, i am to you later, congratulate you for pleasure, goodbye. "
Substep 1 b: if the 'convenient/possible' is identified, the step 2 is entered;
step 2: "you are good, about you feed back the single number mantissa: a question that you have not received but is shown by the last four variables. Asking you to receive the express now? "
Substep 2 a: reply with the recognition of "none": "is good, sorry, ask for you to ask after you complain that do we have a network of people to contact you? "
Substep 2 b: recognition of "present/absent" replies: "good, please do not want to be urgent first, i contact the website immediately for you to solve, we will deal with you until you are satisfied. If you have any problems, we can also be contacted at any time 95546. "
Substep 2 b-1: recognition of "any" replies: "good, wish you live pleasantly, and goodbye. "
Substep 2 b-2: identify "yes/received/what" replies: "ask you for your satisfaction with the after-sales service of our customer service personnel? "
Substep 2 b-2-1: if the 'satisfaction' is recognized, the step 3 is carried out;
substep 2 b-2-2: recognition of "dissatisfaction" replies: "good, ask you to which side you feel dissatisfied? Such as processing results, timeliness, service attitudes, etc., do you describe in trouble with describing them exactly once? "
Substep 2 b-2-2-1: recognizing 'delivery slow/attitude difference', and continuing to enter the step 3;
and step 3: and (3) recovering: "en, thank you for your feedback, ask for your next express mail, how do you want we deliver you in? You can tell me through a button, 1, call before; 2. delivering goods to a door; 3. directly send to your nearby collection point, press 0 "for the listening request.
Substep 3 a: identify "any", then reply: "good, wish you live pleasantly, see again".
It should be noted that the answering call completes the broadcast through the target terminal corresponding to the complaint user of the robot system.
In the embodiment of the invention, the call is initiated to the user through the synthesized intelligent voice, the evaluation of the processing result of the initiated complaint work order by the user is rapidly obtained, and the voice broadcasting is carried out according to the evaluation, so that the evaluation processing efficiency of the complaint work order is improved, and the finishing rate of the complaint work order is improved.
Referring to fig. 2, another embodiment of the method for evaluating and processing a complaint work order according to the embodiment of the present invention includes:
201. when the service system receives a complaint work order initiated by a user, reading order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order.
When the service system receives a complaint work order initiated by a user, the server reads order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order. The contact information of the complaint user can be a mobile phone number registered or bound by the complaint user or a client number bound by the complaint user, as long as a target terminal corresponding to the contact information of the complaint user can receive a voice session initiated by the robot system, and the specific point is not limited herein.
The processing result of the complaint order comprises a plurality of states of finished, processing, waiting to be processed and the like, and the estimated processing time length or the estimated finishing date and the like can be provided by combining the states of the complaint order. It should be noted that the processing result of the complaint order may include the current processing state, complaint processing opinions, complaint processing measures, and the like.
It should be understood that the execution subject of the present invention may be an evaluation processing device of a complaint work order, and may also be a terminal or a server, which is not limited herein. The embodiment of the present invention is described by taking a server as an execution subject.
Optionally, step 201 further includes:
the method comprises the steps that a server obtains a user speech technology in the express logistics field and obtains related corpora in the user speech technology;
for example, the server first collects customer statements in the field of mass delivery logistics and posts label categories such as "satisfied", "not connected", "complaint", and the like.
The server performs data cleaning on the linguistic data related to the user speech technology to obtain the cleaned linguistic data;
specifically, data cleaning is performed, which mainly includes operations of high-correlation data deduplication, low variance data cleaning, data balance distribution adjustment, discarding of excessively long and short corpora, high-similarity corpus cleaning and the like. The specific treatment process is as follows:
and (3) high-correlation data deduplication: for a given data set, firstly selecting a text and the rest sample sets to carry out pairwise similarity estimation, rejecting samples with very high similarity (eg.0.95) with the text data, and carrying out cyclic processing. Calculating the similarity of every two by adopting a cosine similarity formula, wherein the specific formula is as follows:
wherein, XiAnd YiRepresenting two vectors respectively. The cosine value between the included angles of the two vectors in a vector space is used as the measure of the difference between the two individuals, the cosine value is close to 1, the included angle tends to 0, the more similar the two vectors are, the cosine value is close to 0, and the included angle tends to 90 degrees, the more dissimilar the two vectors are.
And (3) low variance data elimination: and performing down-sampling extraction on a given data set after hierarchical clustering, so as to avoid unbalanced data distribution.
Eliminating abnormal data of which the text data length deviates from the distribution of the sample set: and (4) counting the statistical characteristics of the length distribution of the corpus under the production environment, and eliminating samples (overlong and overlong texts) which deviate from the statistical distribution from one end to the other end.
And the server generates a plurality of classification models based on the bidirectional long-short term memory algorithm and the cleaned linguistic data, and selects the classification model with the highest performance score to determine the classification model as the natural language understanding engine.
For example, combining representations of words into a representation of a sentence, an additive approach may be used, i.e., summing the representations of all words, or an equal approach may be taken, but these approaches do not take into account the order of the words before and after the sentence.
The use of L STM model can better capture longer distance dependencies because L STM learns which information to remember and which information to forget through the training process.
However, modeling sentences using L STM has the problem that no information from the back to the front can be encoded, in finer grained classification, such as the five classification tasks of strong recognition, weak recognition, neutral, weak dereference, and strong dereference, attention is paid to the interaction between emotional words, degree words, and negative words, as an example, "this restaurant is dirty and has no next door to good", where "not go" is a modification of the degree of "dirty" and thus, Bi-directional semantic dependence can be better captured by Bi L, where the Bi-directional long-short term memory (Bi L STM) algorithm is a combination of forward L and backward L STM, both of which are commonly used for context information in natural language processing tasks.
Optionally, step 201 further includes:
and carrying out communication connection between the robot system and the service system based on the session initiation protocol.
Session Initiation Protocol (SIP) is a multimedia communication protocol established by the Internet Engineering Task Force (IETF). It is a text-based application-layer control protocol for creating, modifying and releasing sessions of one or more participants. The SIP is an IP voice session control protocol originated from the Internet, and has the characteristics of flexibility, easiness in implementation, convenience in expansion and the like.
202. And synthesizing the processing result into initial voice based on the deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems.
And the server synthesizes the processing result into initial voice based on the deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems. The processing results of a plurality of preset questions and complaint orders are converted into voice from characters through a preset deep neural network algorithm, and then initial voice is obtained.
203. And initiating a voice conversation to a target terminal of the complaint user according to the contact way of the complaint user, and sending the initial voice to the target terminal.
The server initiates a voice conversation to a target terminal of the complaint user according to the contact information of the complaint user, and sends the initial voice to the target terminal. If the contact way of the complaint user is a mobile phone number registered or bound by the complaint user, the server calls the mobile phone number and sends the pre-generated initial voice to a mobile phone terminal of the complaint user; if the contact information of the complaint user is the client number bound by the complaint user, the server initiates a voice call to the client number and sends the pre-generated initial voice to the client of the complaint user.
For example, several questions that are preset may be included in the initial speech, such as question 1: "do you like Zhao lady/mr. honoring, i am a rhyme express company to serve a minor rhyme for customer service, about the problem of food express feedbacks you give back 1 month 1 day 2020, we have already dealt with feedback from the network, do you want and you check it, ask you for your convenience now? ", or problem 2: "you feed back the single number mantissa: 3037 (last four variables) shows a question of sign-in but you have not received. Asking you to receive the express now? ", or problem 3: asking for which side you feel dissatisfied? Such as processing results, timeliness, service attitudes, etc., do you describe in trouble with describing them exactly once? ".
204. And acquiring the reply voice fed back by the target terminal within the preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text.
The server obtains the reply voice fed back by the target terminal within the preset duration range, and the reply voice is converted by calling a preset voice recognition algorithm to obtain a reply text.
For example, if the complaint user answers question 1 in step 103, the resulting reply speech may be: "busy", "inconvenient", "convenient" or "change time"; the reply voice for question 2 may be: "none", "there", or "received"; the reply voice for question 3 may be: "slow delivery," poor attitude, "or" satisfied.
It should be noted that the server converts the speech into words through an Automatic Speech Recognition (ASR) algorithm to obtain the reply text, wherein the ASR algorithm takes the speech as a research object and allows the machine to automatically recognize and understand the speech dictated by human through speech signal processing and pattern recognition. Speech recognition is a technique that allows a machine to convert speech signals into corresponding text or commands through a recognition and understanding process. The speech recognition is a very extensive cross discipline, and has a very close relationship with the disciplines of acoustics, phonetics, linguistics, information theory, pattern recognition theory, neurobiology and the like, and details are not repeated here.
205. Performing intention recognition on the reply text through a natural language understanding engine to obtain a plurality of first candidate intents;
specifically, the server inputs the reply text into a natural language understanding engine, and analyzes the reply text to obtain a plurality of candidate words; the server extracts high-frequency phrases from the candidate words to obtain a plurality of high-frequency phrases, wherein the high-frequency phrases are the candidate words with the occurrence frequency exceeding a threshold value; the server judges whether the high-frequency phrases are matched with a preset phrase library or not; if the high-frequency phrases are matched with the preset phrase library, the server directly matches a plurality of preset intentions, and the preset intentions are determined as a plurality of first candidate intentions; if the high-frequency phrases are not matched with the preset phrase library, the server calls the preset recognition models to recognize the reply text respectively to obtain a plurality of first candidate intentions.
It should be noted that there is a short preprocessing stage before entering the specific intent recognition subtasks. The pretreatment links are as follows: high frequency phrases match intent exactly. Because of the spoken language feature in the intelligent dialogue field, most of users can speak briefly and speak in a spoken language, a high-frequency phrase library is formed by analyzing the online conversation record and extracting the high-frequency phrase,
206. performing semantic recognition on the reply text based on a semantic slot dialogue model to obtain a plurality of second candidate intents;
the server identifies the field of the reply text to obtain a target field; the server identifies the intentions of the reply text through a preset identification algorithm to obtain a plurality of intentions to be selected; the server fills the slots in the reply text through a semantic slot dialogue model to obtain a plurality of slot attributes; and the server screens the plurality of intentions to be selected according to the target field and the plurality of slot attributes to obtain a plurality of second candidate intentions.
It is understood that in a certain dialogue field, such as telecommunication, credit card, real estate agency marketing, etc., according to the marketing dialogue demand flow chart of the client, the training set with each intention label as the characteristic attribute is extracted by combining the historical call data. And finding out the characteristic attribute which enables the information entropy of the training data to be minimum according to an ID3 algorithm and a C4.5 algorithm, and using the characteristic attribute as a judgment attribute in the conversation flow chart to construct an optimal decision tree, so that the optimal marketing conversation flow chart is obtained, the marketing efficiency is improved, and the maximum potential valuable user is obtained.
It should be noted that, in the session, the server obtains the information that is critical and valuable to the service in the user's speech to complete the slot drawing and filling actions. Particularly, the method is applied to an intelligent customer service conversation scene, generally, a user calls the system of the user with questions, information and word slot data described in the dialect of the user are required to be extracted to complete a conversation process, and the method is not limited to relying on the intention of the user. The semantic slot dialogue depends on external structured data, a large number of entity libraries and a rule library to assist in extracting word slot information. At present, a semantic slot conversation model is successfully applied to an order placing conversation system in the field of express delivery, and a very good operation effect is achieved through service evaluation data.
207. And screening the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the real intentions of the complaint users.
The server fuses the plurality of first candidate intentions and the plurality of second candidate intentions to obtain a plurality of candidate intentions to be evaluated; the server calls a plurality of preset identification models to score each candidate intention to be evaluated respectively to obtain a scoring result corresponding to each candidate intention to be evaluated, wherein the scoring result comprises a plurality of target scores and different target scores corresponding to different preset identification models; the server acquires the weight value of each preset identification model; the server calculates according to the weight value of each preset identification model and the grading result corresponding to each candidate intention to be evaluated, determines the comprehensive score of each candidate intention to be evaluated, and obtains a plurality of comprehensive scores; and the server ranks the multiple comprehensive scores and determines the candidate intention to be evaluated with the highest comprehensive score as the real intention of the complaint user.
For example, the preset recognition models include a similarity intention model, a rule template matching intention model and a recurrent neural network intention classification model, wherein 1, the similarity intention model: a corpus is constructed by analyzing online dialogs and collecting user-specific consultations, non-high frequency occurrences of the dialogs. The intent results are ordered when applied by computing similarity of word vectors. The application of the method is mainly applied to the application of a model which needs generalized intention recognition but lacks explanatory property. The computational efficiency is lower than that of other schemes, and the corpus cannot be infinitely expanded to increase the generalization.
2. Matching the rule template with the intention model: collecting dialects with regularity, structuralization, parallel arrangement, double and turning intentions, and extracting key characters by a mining tool to establish a regular formula library and a rule library. Thus forming a template library of rules and labels. When applied, when multiple intents are matched, and edit distance is used to compute the ranking of the intents.
3. The cyclic neural network intention classification model is characterized by collecting a large number of user dialogues, carrying out manual or unsupervised learning labeling operation after data cleaning to form a training and testing set, carrying out classification training and evaluation by using a Convolutional Neural Network (CNN) or a bidirectional long short-term memory (Bi L STM) algorithm model to obtain an optimal model.
The execution of the above submodels is completed in parallel computing in different hardware environments, and the computing efficiency is not influenced theoretically. Therefore, the sorting intentions identified by each sub-model need to be comprehensively sorted according to certain weight parameters to obtain the best intentions (namely the true intentions of the complaining users),
for example, the candidate intents to be evaluated include 6 candidate intents, which are c1, c2, c3, c4, c5 and c 6. For example, assuming that the weight value of the similarity intention model is 0.5, the candidate intention ranking result is: c2, c5, c3, c1, c4 and c6, wherein the corresponding score results are as follows: x1, x2, x3, x4, x5, x 6; assuming that the weight value of the rule template matching intention model is 0.3, the candidate intention ranking result is: c6, c2, c3, c5, c4 and c1, wherein the corresponding score results are as follows: y1, y2, y3, y4, y5, y 6; assuming that the weight value of the recurrent neural network intention classification model is 0.2, the candidate intention ranking result is: c1, c2, c3, c4, c5 and c6, wherein the corresponding score results are as follows: z1, z2, z3, z4, z5, z 6. The composite score results were as follows:
the overall score of c1 was: 0.5 × 4+0.3 × y6+0.2 × z 1;
the overall score of c2 was: 0.5 × 1+0.3 × y2+0.2 × z 2;
the overall score of c3 was: 0.5 × 3+0.3 × y3+0.2 × z 3;
the overall score of c4 was: 0.5 × 5+0.3 × y5+0.2 × z 4;
the overall score of c5 was: 0.5 × 2+0.3 × y5+0.2 × z 5;
the overall score of c6 was: 0.5 × 6+0.3 × y1+0.2 × z 6.
And selecting the intention with the highest composite score to determine the intention of the complaint user.
Wherein, a corpus is formed by analyzing online dialogs and collecting user-specific counseling and non-high-frequency appearing dialogs. The intent results are ordered when applied by computing similarity of word vectors.
The application of the method is mainly applied to the application of a model which needs generalized intention recognition but lacks explanatory property. The computational efficiency is lower than that of other schemes, and the corpus cannot be infinitely expanded to increase the generalization.
2. Matching intention of the rule template: collecting dialects with regularity, structuralization, parallel arrangement, double and turning intentions, and extracting key characters by a mining tool to establish a regular formula library and a rule library. Thus forming a template library of rules and labels. When applied, when multiple intents are matched, and edit distance is used to compute the ranking of the intents.
3. And (3) cyclic neural network intention classification, namely collecting a large number of user dialects, cleaning data, performing manual or unsupervised learning labeling operation to form a training and testing set, performing classification training and evaluation by using a CNN or Bi L STM model to obtain an optimal model, and sequencing intention classification results in application.
The application side of the method is mainly characterized by strong generalization, and the method has the characteristic of end-to-end training without much manual intervention.
The execution of the subtasks is completed in parallel in different hardware environments, and the calculation efficiency is not influenced theoretically. Therefore, the sorting intents identified by each sub-module need to be comprehensively sorted according to certain weight parameters to obtain the optimal intents,
208. and calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through the robot system.
And the server calls a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and the target answer is broadcasted to the client of the complaint user through the robot system. For example, we can get the matching target answer based on the actual intention of the complaint user. For example, if the actual intent of the complaint user is that no express delivery was received, the answer dialog may be:
step 1: "do you good, XXX women/mr. i am a minor rhyme for customer service of a rhyme express company, about a question about a certain express item fed back by date X of year X, the feedback of a network of us has been processed, and do you want to ask you for convenience with you check it? "
Substep 1 a: recognizing "busy/inconvenient/time change" then reply: "good, get a lot of apology bother you, i am to you later, congratulate you for pleasure, goodbye. "
Substep 1 b: if the 'convenient/possible' is identified, the step 2 is entered;
step 2: "you are good, about you feed back the single number mantissa: a question that you have not received but is shown by the last four variables. Asking you to receive the express now? "
Substep 2 a: reply with the recognition of "none": "is good, sorry, ask for you to ask after you complain that do we have a network of people to contact you? "
Substep 2 b: recognition of "present/absent" replies: "good, please do not want to be urgent first, i contact the website immediately for you to solve, we will deal with you until you are satisfied. If you have any problems, we can also be contacted at any time 95546. "
Substep 2 b-1: recognition of "any" replies: "good, wish you live pleasantly, and goodbye. "
Substep 2 b-2: identify "yes/received/what" replies: "ask you for your satisfaction with the after-sales service of our customer service personnel? "
Substep 2 b-2-1: if the 'satisfaction' is recognized, the step 3 is carried out;
substep 2 b-2-2: recognition of "dissatisfaction" replies: "good, ask you to which side you feel dissatisfied? Such as processing results, timeliness, service attitudes, etc., do you describe in trouble with describing them exactly once? "
Substep 2 b-2-2-1: recognizing 'delivery slow/attitude difference', and continuing to enter the step 3;
and step 3: and (3) recovering: "en, thank you for your feedback, ask for your next express mail, how do you want we deliver you in? You can tell me through a button, 1, call before; 2. delivering goods to a door; 3. directly send to your nearby collection point, press 0 "for the listening request.
Substep 3 a: identify "any", then reply: "good, wish you live pleasantly, see again".
It should be noted that the answering call completes the broadcast through the target terminal corresponding to the complaint user of the robot system.
In the embodiment of the invention, the call is initiated to the user through the synthesized intelligent voice, the evaluation of the processing result of the initiated complaint work order by the user is rapidly obtained, and the voice broadcasting is carried out according to the evaluation, so that the evaluation processing efficiency of the complaint work order is improved, and the finishing rate of the complaint work order is improved.
With reference to fig. 3, the method for evaluating and processing a complaint work order in the embodiment of the present invention is described above, and an evaluation and processing apparatus for a complaint work order in the embodiment of the present invention is described below, where an embodiment of the evaluation and processing apparatus for a complaint work order in the embodiment of the present invention includes:
the reading module 301 is configured to, when a service system receives a complaint work order initiated by a user, read order information corresponding to the complaint work order, where the order information includes a contact information of the complaint user and a processing result of the complaint order;
a synthesizing module 302, configured to synthesize the processing result into an initial voice based on a deep neural network, where the initial voice is used to indicate the processing result, and the initial voice includes a plurality of preset questions;
a session module 303, configured to initiate a voice session to a target terminal of the complaint user according to the contact information of the complaint user, and send the initial voice to the target terminal;
a first obtaining module 304, configured to obtain a reply voice fed back by the target terminal within a preset duration range, and convert the reply voice by calling a preset voice recognition algorithm to obtain a reply text;
the recognition module 305 is used for performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user;
and the matching broadcasting module 306 is configured to invoke a preset question and answer management engine to match the real intention of the complaint user, obtain a matched target answer, and broadcast the target answer to the client of the complaint user through the robot system.
In the embodiment of the invention, the call is initiated to the user through the synthesized intelligent voice, the evaluation of the processing result of the initiated complaint work order by the user is rapidly obtained, and the voice broadcasting is carried out according to the evaluation, so that the evaluation processing efficiency of the complaint work order is improved, and the finishing rate of the complaint work order is improved.
Referring to fig. 4, another embodiment of the evaluation processing apparatus for complaint work orders according to the embodiment of the present invention includes:
the reading module 301 is configured to, when a service system receives a complaint work order initiated by a user, read order information corresponding to the complaint work order, where the order information includes a contact information of the complaint user and a processing result of the complaint order;
a synthesizing module 302, configured to synthesize the processing result into an initial voice based on a deep neural network, where the initial voice is used to indicate the processing result, and the initial voice includes a plurality of preset questions;
a session module 303, configured to initiate a voice session to a target terminal of the complaint user according to the contact information of the complaint user, and send the initial voice to the target terminal;
a first obtaining module 304, configured to obtain a reply voice fed back by the target terminal within a preset duration range, and convert the reply voice by calling a preset voice recognition algorithm to obtain a reply text;
the recognition module 305 is used for performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user;
and the matching broadcasting module 306 is configured to invoke a preset question and answer management engine to match the real intention of the complaint user, obtain a matched target answer, and broadcast the target answer to the client of the complaint user through the robot system.
Optionally, the identification module 305 includes:
the first identification unit 3051, configured to perform intent identification on the reply text through a natural language understanding engine to obtain a plurality of first candidate intents;
the second identification unit 3052, configured to perform semantic identification on the reply text based on a semantic slot conversation model to obtain a plurality of second candidate intents;
a screening unit 3053, configured to screen the plurality of first candidate intentions and the plurality of second candidate intentions to obtain a real intention of the complaint user.
Optionally, the first identifying unit 3051 is specifically configured to:
inputting the reply text into a natural language understanding engine, and analyzing the reply text to obtain a plurality of candidate words; extracting high-frequency phrases from the candidate words to obtain a plurality of high-frequency phrases, wherein the high-frequency phrases are the candidate words with the occurrence frequency exceeding a threshold value; judging whether the high-frequency phrases are matched with a preset phrase library or not; if the high-frequency phrases are matched with a preset phrase library, directly matching a plurality of preset intentions, and determining the preset intentions as a plurality of first candidate intentions; and if the high-frequency phrases are not matched with a preset phrase library, respectively calling a plurality of preset identification models to identify the reply text to obtain a plurality of first candidate intents.
Optionally, the second identifying unit 3052 is specifically configured to:
identifying the field of the reply text to obtain a target field; performing intention recognition on the reply text through a preset recognition algorithm to obtain a plurality of intentions to be selected; filling slots in the reply text through a semantic slot dialogue model to obtain a plurality of slot attributes; and screening the plurality of intentions to be selected according to the target field and the plurality of slot attributes to obtain a plurality of second candidate intentions.
Optionally, the screening unit 3053 is specifically configured to:
fusing the plurality of first candidate intentions and the plurality of second candidate intentions to obtain a plurality of candidate intentions to be evaluated; calling a plurality of preset identification models to score each candidate intention to be evaluated respectively to obtain a scoring result corresponding to each candidate intention to be evaluated, wherein the scoring result comprises a plurality of target scores and different target scores corresponding to different preset identification models; acquiring a weight value of each preset identification model; calculating according to the weight value of each preset identification model and the grading result corresponding to each candidate intention to be evaluated, and determining the comprehensive score of each candidate intention to be evaluated to obtain a plurality of comprehensive scores; and sequencing the multiple comprehensive scores, and determining the candidate intention to be evaluated with the highest comprehensive score as the real intention of the complaint user.
Optionally, the evaluation processing apparatus for the complaint work order further includes:
a second obtaining module 307, configured to obtain a user speech technology in the field of express logistics, and obtain a corpus related to the user speech technology;
a data cleaning module 308, configured to perform data cleaning on the linguistic data related to the user's speech technology to obtain cleaned linguistic data;
and a generation determining module 309, configured to generate a plurality of classification models based on the bidirectional long-short term memory algorithm and the cleaned corpus, and select the classification model with the highest performance score to determine the classification model as the natural language understanding engine.
Optionally, the evaluation processing apparatus for the complaint work order further includes: :
and a connection module 310, configured to perform communication connection between the robot system and the service system based on the session initiation protocol.
In the embodiment of the invention, the call is initiated to the user through the synthesized intelligent voice, the evaluation of the processing result of the initiated complaint work order by the user is rapidly obtained, and the voice broadcasting is carried out according to the evaluation, so that the evaluation processing efficiency of the complaint work order is improved, and the finishing rate of the complaint work order is improved.
Fig. 3 and 4 describe the evaluation processing device of the complaint work order in the embodiment of the present invention in detail from the perspective of the modular functional entity, and the evaluation processing device of the complaint work order in the embodiment of the present invention is described in detail from the perspective of hardware processing.
Fig. 5 is a schematic structural diagram of an evaluation processing apparatus for a complaint work order, where the evaluation processing apparatus 500 for a complaint work order may have a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 510 (e.g., one or more processors) and a memory 520, and one or more storage media 530 (e.g., one or more mass storage devices) storing an application program 533 or data 532. Memory 520 and storage media 530 may be, among other things, transient or persistent storage. The program stored on the storage medium 530 may include one or more modules (not shown), each of which may include a series of instruction operations in the evaluation processing device 500 for a complaint work order. Still further, the processor 510 may be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the evaluation processing device 500 of the complaint work order.
The evaluation equipment 500 of the complaint work order may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input-output interfaces 560, and/or one or more operating systems 531, such as Windows Server, Mac OS X, Unix, L inux, FreeBSD, etc. it will be understood by those skilled in the art that the evaluation equipment configuration of the complaint work order shown in FIG. 5 does not constitute a limitation of the evaluation equipment of the complaint work order, may include more or fewer components than shown, or may combine some components, or a different arrangement of components.
The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium, and which may also be a volatile computer-readable storage medium, having stored therein instructions that, when executed on a computer, cause the computer to execute the steps of the method for evaluating and processing a complaint work order.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for evaluating a complaint work order, comprising:
when a service system receives a complaint work order initiated by a user, reading order information corresponding to the complaint work order, wherein the order information comprises a contact way of the complaint user and a processing result of the complaint order;
synthesizing the processing result into initial voice based on a deep neural network, wherein the initial voice is used for indicating the processing result and comprises a plurality of preset problems;
initiating a voice conversation to a target terminal of the complaint user according to the contact way of the complaint user, and sending the initial voice to the target terminal;
acquiring reply voice fed back by the target terminal within a preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text;
performing intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user;
and calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through a robot system.
2. The method for evaluating and processing the complaint work order according to claim 1, wherein the intention recognition of the reply text by a natural language understanding engine to obtain the real intention of the complaint user comprises:
performing intention recognition on the reply text through a natural language understanding engine to obtain a plurality of first candidate intentions;
performing semantic recognition on the reply text based on a semantic slot dialogue model to obtain a plurality of second candidate intents;
and screening the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the real intentions of the complaint users.
3. The method for evaluating a complaint work order according to claim 2, wherein the identifying the intent of the reply text by a natural language understanding engine to obtain a plurality of first candidate intentions comprises:
inputting the reply text into a natural language understanding engine, and analyzing the reply text to obtain a plurality of candidate words;
extracting high-frequency phrases from the candidate words to obtain a plurality of high-frequency phrases, wherein the high-frequency phrases are the candidate words with the occurrence frequency exceeding a threshold value;
judging whether the high-frequency phrases are matched with a preset phrase library or not;
if the high-frequency phrases are matched with a preset phrase library, directly matching a plurality of preset intentions, and determining the preset intentions as a plurality of first candidate intentions;
and if the high-frequency phrases are not matched with a preset phrase library, respectively calling a plurality of preset identification models to identify the reply text to obtain a plurality of first candidate intents.
4. The method for evaluating and processing complaint work orders according to claim 2, wherein the semantic recognition of the reply text based on a semantic slot dialogue model to obtain a plurality of second candidate intents comprises:
identifying the field of the reply text to obtain a target field;
performing intention recognition on the reply text through a preset recognition algorithm to obtain a plurality of intentions to be selected;
filling slots in the reply text through a semantic slot dialogue model to obtain a plurality of slot attributes;
and screening the plurality of intentions to be selected according to the target field and the plurality of slot attributes to obtain a plurality of second candidate intentions.
5. The method for evaluating and processing the complaint work order according to claim 2, wherein the screening the plurality of first candidate intentions and the plurality of second candidate intentions to obtain the true intentions of the complaint user comprises:
fusing the plurality of first candidate intentions and the plurality of second candidate intentions to obtain a plurality of candidate intentions to be evaluated;
calling a plurality of preset identification models to score each candidate intention to be evaluated respectively to obtain a scoring result corresponding to each candidate intention to be evaluated, wherein the scoring result comprises a plurality of target scores and different target scores corresponding to different preset identification models;
acquiring a weight value of each preset identification model;
calculating according to the weight value of each preset identification model and the grading result corresponding to each candidate intention to be evaluated, and determining the comprehensive score of each candidate intention to be evaluated to obtain a plurality of comprehensive scores;
and sequencing the multiple comprehensive scores, and determining the candidate intention to be evaluated with the highest comprehensive score as the real intention of the complaint user.
6. The method for evaluating and processing the complaint work order according to any one of claims 1-5, wherein when the service system receives a complaint work order initiated by a user, reading order information corresponding to the complaint work order, wherein the order information includes a contact information of the complaint user and a processing result of the complaint order, the method for evaluating and processing the complaint work order further comprises:
the method comprises the steps of obtaining a user speech technology in the express logistics field, and obtaining related corpora in the user speech technology;
performing data cleaning on the linguistic data related to the user speech technology to obtain the cleaned linguistic data;
and generating a plurality of classification models based on a bidirectional long-short term memory algorithm and the cleaned linguistic data, and selecting the classification model with the highest performance score to determine as a natural language understanding engine.
7. The method for evaluating and processing the complaint work order according to any one of claims 1-5, wherein when the complaint work order initiated by the user is received, reading order information corresponding to the complaint work order, wherein the order information includes a client number of the complaint user and a processing result of the complaint order, the method for evaluating and processing the complaint work order further comprises:
and carrying out communication connection between the robot system and the service system based on the session initiation protocol.
8. An evaluation processing device for a complaint work order, comprising:
the service system comprises a reading module, a processing module and a processing module, wherein the reading module is used for reading order information corresponding to a complaint work order when the service system receives the complaint work order initiated by a user, and the order information comprises a contact way of the complaint user and a processing result of the complaint order;
a synthesis module, configured to synthesize the processing result into an initial voice based on a deep neural network, where the initial voice is used to indicate the processing result, and the initial voice includes a plurality of preset problems;
the conversation module is used for initiating a voice conversation to a target terminal of the complaint user according to the contact information of the complaint user and sending the initial voice to the target terminal;
the first acquisition module is used for acquiring reply voice fed back by the target terminal within a preset duration range, and calling a preset voice recognition algorithm to convert the reply voice to obtain a reply text;
the recognition module is used for carrying out intention recognition on the reply text through a natural language understanding engine to obtain the real intention of the complaint user;
and the matching broadcasting module is used for calling a preset question-answer management engine to match the real intention of the complaint user to obtain a matched target answer, and broadcasting the target answer to the client of the complaint user through a robot system.
9. An evaluation processing apparatus for a complaint work order, characterized by comprising: a memory having instructions stored therein and at least one processor, the memory and the at least one processor interconnected by a line;
the at least one processor invokes the instructions in the memory to cause the evaluation processing equipment of the complaint work order to execute the evaluation processing method of the complaint work order according to any of claims 1-7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the method for evaluation processing of a complaint work order according to any one of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010227729.8A CN111429157A (en) | 2020-03-27 | 2020-03-27 | Method, device and equipment for evaluating and processing complaint work order and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010227729.8A CN111429157A (en) | 2020-03-27 | 2020-03-27 | Method, device and equipment for evaluating and processing complaint work order and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111429157A true CN111429157A (en) | 2020-07-17 |
Family
ID=71548918
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010227729.8A Pending CN111429157A (en) | 2020-03-27 | 2020-03-27 | Method, device and equipment for evaluating and processing complaint work order and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111429157A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112199477A (en) * | 2020-09-04 | 2021-01-08 | 北京邮电大学 | Dialogue management scheme and dialogue management corpus construction method |
CN113239701A (en) * | 2021-05-07 | 2021-08-10 | 京东数字科技控股股份有限公司 | Data analysis processing method and device |
CN114490975A (en) * | 2021-12-31 | 2022-05-13 | 马上消费金融股份有限公司 | User question labeling method and device |
CN116468341A (en) * | 2023-04-03 | 2023-07-21 | 上海乾臻信息科技有限公司 | Processing method, device and system of arbitration worksheet and storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109345262A (en) * | 2017-08-01 | 2019-02-15 | 兰州大学 | It is a kind of intelligently to complain classification and processing system |
CN109587360A (en) * | 2018-11-12 | 2019-04-05 | 平安科技(深圳)有限公司 | Electronic device should talk with art recommended method and computer readable storage medium |
CN109684475A (en) * | 2018-11-21 | 2019-04-26 | 斑马网络技术有限公司 | Processing method, device, equipment and the storage medium of complaint |
CN109727041A (en) * | 2018-07-03 | 2019-05-07 | 平安科技(深圳)有限公司 | Intelligent customer service takes turns answering method, equipment, storage medium and device more |
CN109816399A (en) * | 2019-01-07 | 2019-05-28 | 平安科技(深圳)有限公司 | Complain management method, device, computer equipment and the storage medium of part |
CN109981910A (en) * | 2019-02-22 | 2019-07-05 | 中国联合网络通信集团有限公司 | Business recommended method and apparatus |
CN109993543A (en) * | 2017-12-28 | 2019-07-09 | 中国移动通信集团广东有限公司 | A kind of complaint handling method and system |
CN110704641A (en) * | 2019-10-11 | 2020-01-17 | 零犀(北京)科技有限公司 | Ten-thousand-level intention classification method and device, storage medium and electronic equipment |
CN110765244A (en) * | 2019-09-18 | 2020-02-07 | 平安科技(深圳)有限公司 | Method and device for acquiring answering, computer equipment and storage medium |
-
2020
- 2020-03-27 CN CN202010227729.8A patent/CN111429157A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109345262A (en) * | 2017-08-01 | 2019-02-15 | 兰州大学 | It is a kind of intelligently to complain classification and processing system |
CN109993543A (en) * | 2017-12-28 | 2019-07-09 | 中国移动通信集团广东有限公司 | A kind of complaint handling method and system |
CN109727041A (en) * | 2018-07-03 | 2019-05-07 | 平安科技(深圳)有限公司 | Intelligent customer service takes turns answering method, equipment, storage medium and device more |
CN109587360A (en) * | 2018-11-12 | 2019-04-05 | 平安科技(深圳)有限公司 | Electronic device should talk with art recommended method and computer readable storage medium |
CN109684475A (en) * | 2018-11-21 | 2019-04-26 | 斑马网络技术有限公司 | Processing method, device, equipment and the storage medium of complaint |
CN109816399A (en) * | 2019-01-07 | 2019-05-28 | 平安科技(深圳)有限公司 | Complain management method, device, computer equipment and the storage medium of part |
CN109981910A (en) * | 2019-02-22 | 2019-07-05 | 中国联合网络通信集团有限公司 | Business recommended method and apparatus |
CN110765244A (en) * | 2019-09-18 | 2020-02-07 | 平安科技(深圳)有限公司 | Method and device for acquiring answering, computer equipment and storage medium |
CN110704641A (en) * | 2019-10-11 | 2020-01-17 | 零犀(北京)科技有限公司 | Ten-thousand-level intention classification method and device, storage medium and electronic equipment |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112199477A (en) * | 2020-09-04 | 2021-01-08 | 北京邮电大学 | Dialogue management scheme and dialogue management corpus construction method |
CN112199477B (en) * | 2020-09-04 | 2023-03-14 | 北京邮电大学 | Dialogue management scheme and dialogue management corpus construction method |
CN113239701A (en) * | 2021-05-07 | 2021-08-10 | 京东数字科技控股股份有限公司 | Data analysis processing method and device |
CN114490975A (en) * | 2021-12-31 | 2022-05-13 | 马上消费金融股份有限公司 | User question labeling method and device |
CN114490975B (en) * | 2021-12-31 | 2023-02-07 | 马上消费金融股份有限公司 | User question labeling method and device |
CN116468341A (en) * | 2023-04-03 | 2023-07-21 | 上海乾臻信息科技有限公司 | Processing method, device and system of arbitration worksheet and storage medium |
CN116468341B (en) * | 2023-04-03 | 2024-04-09 | 上海乾臻信息科技有限公司 | Processing method, device and system of arbitration worksheet and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112000791B (en) | Motor fault knowledge extraction system and method | |
CN111429157A (en) | Method, device and equipment for evaluating and processing complaint work order and storage medium | |
CN109151218A (en) | Call voice quality detecting method, device, computer equipment and storage medium | |
CN110046230B (en) | Method for generating recommended speaking collection, and recommended speaking method and device | |
CN104462600A (en) | Method and device for achieving automatic classification of calling reasons | |
CN113360622B (en) | User dialogue information processing method and device and computer equipment | |
CN110266900B (en) | Method and device for identifying customer intention and customer service system | |
CN110610705A (en) | Voice interaction prompter based on artificial intelligence | |
CN112818109B (en) | Intelligent reply method, medium, device and computing equipment for mail | |
CN112000776A (en) | Topic matching method, device and equipment based on voice semantics and storage medium | |
CN111858854A (en) | Question-answer matching method based on historical dialogue information and related device | |
CN112632248A (en) | Question answering method, device, computer equipment and storage medium | |
JP7160778B2 (en) | Evaluation system, evaluation method, and computer program. | |
CN111062211A (en) | Information extraction method and device, electronic equipment and storage medium | |
CN110750626B (en) | Scene-based task-driven multi-turn dialogue method and system | |
CN116304046A (en) | Dialogue data processing method and device, storage medium and electronic equipment | |
CN113505606B (en) | Training information acquisition method and device, electronic equipment and storage medium | |
CN110765242A (en) | Method, device and system for providing customer service information | |
CN115688758A (en) | Statement intention identification method and device and storage medium | |
CN110517672A (en) | User's intension recognizing method, method for executing user command, system and equipment | |
CN114969295A (en) | Dialog interaction data processing method, device and equipment based on artificial intelligence | |
CN111818290B (en) | Online interviewing method and system | |
CN113468306A (en) | Voice conversation method, device, electronic equipment and storage medium | |
CN113297367A (en) | Method for generating user conversation linking language and related equipment | |
TW202133027A (en) | Dialogue system and method for human-machine cooperation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200717 |
|
RJ01 | Rejection of invention patent application after publication |