CN109902957A - A kind of data processing method and device - Google Patents
A kind of data processing method and device Download PDFInfo
- Publication number
- CN109902957A CN109902957A CN201910150618.9A CN201910150618A CN109902957A CN 109902957 A CN109902957 A CN 109902957A CN 201910150618 A CN201910150618 A CN 201910150618A CN 109902957 A CN109902957 A CN 109902957A
- Authority
- CN
- China
- Prior art keywords
- quality inspection
- target
- text
- data
- service
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention discloses a kind of data processing method and device, this method comprises: obtaining target text information from the destination service text data got, and are carried out abnormality detection according to the target quality inspection points in target quality inspection rule to target text information;If there are the target quality inspection points for meeting target quality inspection rule in target text information, first identifier corresponding with target quality inspection points is set in target text;Associated text information associated with target text information is obtained from destination service text data, and associated text information is carried out abnormality detection according to the association quality inspection points in target quality inspection rule;If there are the association quality inspection points met in target quality inspection rule in associated text information, the then setting second identifier corresponding with association quality inspection points in associated text information, and the destination service text data for carrying first identifier, second identifier is determined as target exception text.Using the embodiment of the present invention, the workload of quality inspection personnel can be reduced, improves quality inspection efficiency.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of data processing method and device.
Background technique
In customer service quality inspection field, service voice data with interact being continuously increased for text data, in customer service quality check process
The speed of data processing more stringent requirements are proposed.
In the prior art, quality check process can be by way of manually inspecting by random samples to (the i.e. customer service and use of the service work order of magnanimity
Obtained text data when being interacted between family) it is sampled, it is manually looked into passing through from the service work order that sampling obtains
Wrong mode is found to score in the presence of each of abnormal thin item, so, when sampling samples are more, quality inspection can be greatly increased
The workload of member.
In addition, existing quality check process can also carry out preliminary screening by service work order of the machine to magnanimity, will obtain
Divide lower service work order to pick out and carries out manual review to quality inspection personnel.Due to being carried out by machine, quality inspection is obtained to be
Scoring for each service work order is still required so Quality Inspector is subsequent when carrying out secondary review by manually looking into
Wrong mode finds out each thin item of error one by one from each service work order, cumbersome, and then leads to quality inspection inefficiency.
Summary of the invention
The embodiment of the present invention provides a kind of data processing method and device, can reduce the workload of quality inspection personnel, and mentions
High quality inspection efficiency.
One aspect of the present invention provides a kind of data processing method, comprising:
Destination service text data is obtained, the acquisition target text information from the destination service text data, and according to
Target quality inspection points in target quality inspection rule carry out abnormality detection the target text information;
If there are the target quality inspection points for meeting target quality inspection rule in the target text information, in the target text
It is middle that first identifier corresponding with the target quality inspection points is set;
Associated text information associated with the target text information is obtained from the destination service text data, and
The associated text information is carried out abnormality detection according to the association quality inspection points in the target quality inspection rule;
If there are the association quality inspection points met in the target quality inspection rule in the associated text information, in the pass
Join setting second identifier corresponding with the association quality inspection points in text information, and the first identifier, described second will be carried
The destination service text data of mark is determined as target exception text.
One aspect of the present invention provides a kind of data processing equipment, comprising:
Text information obtains module and obtains from the destination service text data for obtaining destination service text data
Target text information is taken, and abnormal inspection is carried out to the target text information according to the target quality inspection points in target quality inspection rule
It surveys;
First setup module, if for there is the target quality inspection for meeting target quality inspection rule in the target text information
First identifier corresponding with the target quality inspection points is then arranged in point in the target text;
Related information obtains module, for obtaining and the target text information phase from the destination service text data
Associated associated text information, and the associated text information is carried out according to the association quality inspection points in the target quality inspection rule
Abnormality detection;
Second setup module, if for there is the association met in the target quality inspection rule in the associated text information
Second identifier corresponding with the association quality inspection points is then arranged in quality inspection points in the associated text information, and will be described in carrying
First identifier, the second identifier destination service text data be determined as target exception text.
One aspect of the present invention provides a kind of data processing equipment, comprising: processor and memory;
The processor is connected with memory, wherein for storing program code, the processor is used for the memory
Said program code is called, to execute such as the method in the embodiment of the present invention in one side.
On the one hand the embodiment of the present invention provides a kind of computer storage medium, the computer storage medium is stored with meter
Calculation machine program, the computer program include program instruction, and described program is instructed when being executed by a processor, executed such as the present invention
Method in embodiment in one side.
The embodiment of the present invention is by obtaining the target text information in destination service text data, according to target quality inspection rule
Determine in above-mentioned target text information with the presence or absence of target quality inspection points, and if it exists, then in above-mentioned target text information setting with
The corresponding first identifier of target quality inspection points, and then associated text information associated with above-mentioned target text information, root can be obtained
It determines in above-mentioned associated text information according to above-mentioned target quality inspection rule with the presence or absence of association quality inspection points, and if it exists, then in above-mentioned pass
Join in text information setting be associated with the corresponding second identifier of quality inspection points, and then can will carry first identifier and second identifier
Destination service text data is determined as target exception text.It therefore, can be by the target that gets in entire quality check process
Quality inspection rule, rapidly finds out the thin item in the presence of error, and to the thin Xiang Jinhang that these find out from destination service text data
Label to obtain above-mentioned first identifier and second identifier, and then the destination service text data for carrying label can be determined as
Target exception text.In other words, due to carrying first identifier and second identifier in target exception text, to facilitate subsequent
Quality inspection personnel can quickly orient the thin item of error according to these labels, to reduce the workload of quality inspection personnel, and improve matter
Examine efficiency.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of schematic diagram of a scenario of data processing method provided in an embodiment of the present invention;
Fig. 2 is a kind of flow diagram of data processing method provided in an embodiment of the present invention;
Fig. 3 is the flow diagram of another data processing method provided in an embodiment of the present invention;
Fig. 4 is a kind of circuit theory schematic diagram of customer service quality inspection method provided in an embodiment of the present invention;
Fig. 5 is the flow diagram of another data processing method provided in an embodiment of the present invention;
Fig. 6 a, Fig. 6 b are the circuit theory schematic diagrams of another customer service quality inspection method provided in an embodiment of the present invention;
Fig. 7 is a kind of time diagram of customer service quality inspection method provided in an embodiment of the present invention;
Fig. 8 is a kind of structural schematic diagram of data processing equipment provided in an embodiment of the present invention;
Fig. 9 is the structural schematic diagram of another data processing equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to Figure 1, Fig. 1 is a kind of schematic diagram of a scenario of data processing method provided in an embodiment of the present invention.Such as Fig. 1 institute
Show, during carrying out quality inspection to customer service, server 100 can be from each channel (for example, in online customer service, calling
Heart etc.) obtain session content between customer service and user, i.e., the service voice data between customer service and user, server 100 can
The service voice data that will acquire are input in Application on Voiceprint Recognition model, since Application on Voiceprint Recognition model is according to comprising shifting to an earlier date typing
Customer service vocal print and the voice print database of common words art (for example, " be very glad for you service ") of customer service be trained to obtain
, it therefore in embodiments of the present invention, can be by Application on Voiceprint Recognition model that above-mentioned training obtains voice speaker for identification.It changes
Yan Zhi can identify the voice content for belonging to customer service role by Application on Voiceprint Recognition model, i.e., from above-mentioned service voice data
Service voice data can carried out to determine that customer service role is corresponding in the obtained service text data of text conversion
First text data.So as to which remaining voice content to be determined as to the second text of user role in service text data
Data can determine target text information according to above-mentioned first text data and above-mentioned second text data
Wherein, server 100 can carry out speech recognition to the above-mentioned service voice data carried out after role's differentiation,
Above-mentioned service voice data are converted into service text data, the service text data after the conversion may include above-mentioned first
Text data and above-mentioned second text data.It is understood that above-mentioned speech recognition modeling is according in corpus data library
Largely voice data are trained obtained, so, which can be understood as the function for having had text conversion
Energy.It should be appreciated that in embodiments of the present invention, the service voice data can be understood as acquired from multiple services channels
To any one services channels under for characterizing the voice data to conversate between customer service and user.
After obtaining service text data by speech recognition, the intelligence in text data input quality inspection library can will be serviced
Analysis engine, and above-mentioned service text data is carried out abnormality detection by the Checking model in intellectual analysis engine, so as to
To determine target exception text according to abnormality detection result.In other words, above-mentioned quality inspection can be passed through in embodiments of the present invention
Configured quality inspection rule is screened and is marked to above-mentioned service text data in model.In the mistake of above-mentioned abnormality detection
The service text data not malfunctioned can be determined as normal text by Cheng Zhong, for can be direct according to above-mentioned quality inspection rule
The service text data for confirming error, can directly detect, and be determined as confirming abnormal text, and for quality inspection come out it is doubtful go out
Wrong service text data, above-mentioned Checking model can mark the doubtful error item of the service text data of above-mentioned doubtful error
Note, and the service text data of above-mentioned doubtful error is determined as target exception text, it can be by determining target exception text
It is added to abnormal text data to concentrate, and above-mentioned target exception text is sent to quality inspection terminal 200, so as to quality inspection personnel docking
The target exception text received carries out manual review, and in other words, quality inspection terminal 200 can be according to the carrying error item mark received
Remember that the service work order (i.e. target exception text) of information carries out secondary review, so as to abnormal to above-mentioned target in quality inspection personnel
After text carries out manual review, determine that the service work order that there is abnormal error item is sent to from the service work order of doubtful error
Server 100, server 100 can according to the abnormal quality inspection points in the service work order of abnormal error item to above-mentioned quality inspection rule into
Row updates.
Wherein, above-mentioned quality inspection rule refers to can be used for servicing work order to magnanimity and (carry out between i.e. above-mentioned client and user
Obtained service text data during session) carry out quality inspection rule, the quality inspection rule may include by checkpoint by
According to or non-be combined obtained inspection formula.Wherein it is possible to checkpoint is referred to as quality inspection points, and the checkpoint can
Be magnanimity service work order in any one service work order service link in data, for example, the checkpoint can be it is above-mentioned
Data in the corresponding text data of Consumer Role.Above-mentioned Checking model, which refers to, can service work to the magnanimity in input quality inspection library
In the error item in service work order that singly carries out abnormality detection, and can malfunction to confirmation and the service work order of doubtful error
Doubtful error item carries out a kind of model of smart tags, such as: deep neural network generates and fights network etc..Above-mentioned exception is literary
Notebook data collection may include all history service text datas for being targeted abnormal text, i.e., for storing to be gone out by quality inspection
The service work order of the doubtful error come.
Quality inspection terminal 200 is obtaining matter to the doubtful secondary review of service work order (i.e. target exception text) progress gone wrong
After examining result, the corresponding quality inspection rule of the quality inspection result can be sent to server 100.In order to facilitate understanding, can will take
Being engaged in device 100, configured quality inspection rule is as the first quality inspection rule, using the quality inspection rule in secondary review as the second quality inspection
Rule.In other words, the second quality inspection rule can be fed back to server 100, server 100 available by quality inspection terminal 200
Abnormal quality inspection points in two quality inspections rule, and the first quality inspection rule is updated.
Fig. 2 is referred to, Fig. 2 is a kind of flow diagram of data processing method provided in an embodiment of the present invention.Such as Fig. 2 institute
Show, this method may include:
Step S201 obtains destination service text data, and target text letter is obtained from the destination service text data
Breath, and the target text information is carried out abnormality detection according to the target quality inspection points in target quality inspection rule;
Specifically, server can select service text data (the i.e. customer service for needing to carry out quality inspection from quality inspection library
Interaction data between user) it is used as destination service text data, it can be obtained according to the different role in service text data
Target text information is taken, and above-mentioned target text information is analyzed using the intellectual analysis engine in quality inspection library, in judgement
It states whether target text information has abnormal conditions, in other words, utilizes configured target in above-mentioned intellectual analysis engine
Target quality inspection points in quality inspection rule analyze above-mentioned target text information, judge whether above-mentioned target text information hits
The target quality inspection points.It is understood that it is in need carry out quality inspection service text data can be transferred in quality inspection library,
Quality check process of every execution can select a service text data as destination service text data from quality inspection library,
And then target text information is obtained from destination service text data, above-mentioned target text information can be with unlimited particular content and number
Amount can be and belong to the content of text of customer service in destination service text data (i.e. what is said or talked about for customer service, can be what customer service was said
Wherein a word or a few words) be used as target text information, i.e., belong to content of text (the i.e. user of user in target text data
What is said or talked about, can be wherein a word or a few words that user is said) as target text information, it can also be that one section includes
The content of text that customer service and user continuously talk with;Target quality inspection rule refer in multiple quality inspections rule with destination service text data
The quality inspection rule that affiliated service scenarios match, target quality inspection points can be the preposition inspection in the target quality inspection rule
The hit situation that point, i.e. remaining checkpoint in the target quality inspection rule all rely on the preposition checkpoint.For example, being risen in service
In grade scene, occur keyword " upgrading " in destination service text data, then it can will be in the corresponding text of " upgrading " keyword
Hold (for example, 2 word of front and back for " upgrading " keyword occur) and is determined as target text information, the target matter in target quality inspection rule
It is cautious can for " you is helped to upgrade at once " described in customer service or user described in " me please be help to upgrade " etc..
Optionally, in intellectual analysis engine, multiple destination service text datas can be determined simultaneously, to above-mentioned multiple mesh
Mark service text data carries out parallel parsing, that is, quality check process of every execution, while to multiple destination service textual datas
According to carrying out abnormality detection.
Step S202, if there are the target quality inspection points for meeting target quality inspection rule in the target text information, in institute
It states and first identifier corresponding with the target quality inspection points is set in target text;
Specifically, during being carried out abnormality detection according to above-mentioned target quality inspection points to above-mentioned target text information, it can
To include following two situation: if there are the target quality inspection points for meeting target quality inspection rule, i.e. mesh in above-mentioned target text information
Mark text information has hit above-mentioned target quality inspection points, then the target quality inspection points in target text information can be marked, and marks
Note information can be referred to as first identifier corresponding with target quality inspection points, so as to can be according to above-mentioned the when subsequent progress manual review
One mark is directly targeted to error item;If there is no the target quality inspections for meeting target quality inspection rule in above-mentioned target text information
Above-mentioned destination service text data then can be determined as not having by the above-mentioned target quality inspection points of point, i.e. target text information miss
The normal text of error.Wherein, above-mentioned first identifier is used to mark the hit situations of target quality inspection points, can with numerical value or remaining
Character representation.For example, the target quality inspection points in target quality inspection rule are " to help described in customer service in service upgrade scene at once
You upgrade " or user described in the information keys such as " me please be help to upgrade ", if occurring in above-mentioned target text information
Any one of information keys such as " you is helped to upgrade at once ", " me please be helped to upgrade ", then show above-mentioned target text information
Meet the target quality inspection points in above-mentioned target quality inspection rule, and the keyword of appearance is identified in target text information;
If any in the information keys such as " you is helped to upgrade at once ", " me please be help to upgrade " without occurring in above-mentioned target text information
Kind, then above-mentioned destination service text data can be directly confirmed as the normal text not malfunctioned.
Step S203 obtains association text associated with the target text information from the destination service text data
This information, and the associated text information is carried out abnormality detection according to the association quality inspection points in the target quality inspection rule;
Specifically, after above-mentioned target text information hits target quality inspection points, it can be from above-mentioned destination service text data
It is middle to obtain associated text information associated with above-mentioned target text information, and utilize the association matter in above-mentioned target quality inspection rule
It is cautious that associated text information is analyzed, judge whether above-mentioned associated text information hits above-mentioned association quality inspection points.Wherein, on
Stating associated text information can be the preceding a few words and/or rear a few words for referring to target text information in destination service text data,
It may also mean that and refer in target quality inspection rule with text similar in target text information semantic, above-mentioned association quality inspection points, according to
Remaining quality inspection points of the hit situation of the above-mentioned target quality inspection points of Lai Yu, association quality inspection points can be one in feeling the pulse with the finger-tip mark quality inspection rule
A or multiple quality inspection points.For example, equally in service upgrade scene, it can be by 5 word of front and back of target text information as pass
Join text information, the association quality inspection points in target quality inspection rule may include two quality inspection points (quality inspection points 1, quality inspection points 2), such as matter
Cautious 1 can be customer service role corresponding " you has been helped successfully to upgrade " is not detected, quality inspection points 2 can be that customer service angle is not detected
Color is corresponding " please noting that the notification information checked and accepted and upgraded successfully later ".
Step S204, if there are the association quality inspection points met in the target quality inspection rule in the associated text information,
Second identifier corresponding with the association quality inspection points is then set in the associated text information, and first mark will be carried
Know, the destination service text data of the second identifier is determined as target exception text.
Specifically, during being carried out abnormality detection according to above-mentioned association quality inspection points to above-mentioned associated text information, if
There are the association quality inspection points for meeting target quality inspection rule in above-mentioned associated text information, i.e. associated text information has hit above-mentioned pass
The part quality inspection points that connection quality inspection points are included, then can be corresponding to the part quality inspection points of hit in associated text information
Position is marked, and mark information at this time can be referred to as second identifier, can be according to upper when so as to subsequent progress manual review
It states second identifier and is directly targeted to error item, it can be true by the destination service text data for carrying above-mentioned first identifier and second identifier
It is set to clue, which can be understood as the destination service text data of doubtful error, at this point it is possible to by the mesh of the doubtful error
Mark service text data is referred to as the first text or target exception text;Optionally, if above-mentioned associated text information is hit
Above-mentioned destination service text data can be then determined as the error of detection by all quality inspection points that above-mentioned association quality inspection points are included
Text, the error text of the detection are the destination service text without carrying out manual review for passing through the quality inspection system and being confirmed
Notebook data, at this point it is possible to which the destination service text data that this is confirmed is referred to as the second text or confirms abnormal text.Example
Such as, equally in service upgrade scene, the association quality inspection points in target quality inspection rule may include quality inspection points 1: visitor is not detected
It takes role's corresponding " you has been helped successfully to upgrade " and quality inspection points 2: being not detected that customer service role is corresponding " to be please noted that look into later
Receive the notification information upgraded successfully ", if being not detected in above-mentioned associated text information, customer service role is corresponding " to have helped you successfully to rise
Grade " and to be not detected in preset duration customer service role corresponding " please noting that the notification information checked and accepted and upgraded successfully later ", then
The improper rule of upgrading has been hit in determination, and then above-mentioned destination service text information can be determined as to the second of aforementioned confirmation error
Text (confirms abnormal text);Optionally, equally in the service upgrade scene, if being hit in above-mentioned associated text information
Quality inspection points 1 in above-mentioned association quality inspection points: it is corresponding " you has been helped successfully to upgrade " that customer service role is not detected, and when default
Long miss is associated with quality inspection points 2, i.e., " please noting that the notification information checked and accepted and upgraded successfully later " is had received in preset duration,
Then determine that there are part quality inspection points to have hit the improper rule of the upgrading (i.e. above-mentioned target quality inspection rule), so according to the target matter
Inspection rule temporarily can not carry out accurate judgement to the related information text therefore can be in associated text information to above-mentioned pass
Connection quality inspection points 1 are marked, and the destination service text data of carrying association quality inspection points 1 is different as doubtful abnormal target
Chang Wenben.
The embodiment of the present invention is by obtaining the target text information in destination service text data, according to target quality inspection rule
Determine in above-mentioned target text information with the presence or absence of target quality inspection points, and if it exists, then in above-mentioned target text information setting with
The corresponding first identifier of target quality inspection points, and then associated text information associated with above-mentioned target text information, root can be obtained
It determines in above-mentioned associated text information according to above-mentioned target quality inspection rule with the presence or absence of association quality inspection points, and if it exists, then in above-mentioned pass
Join in text information setting be associated with the corresponding second identifier of quality inspection points, and then can will carry first identifier and second identifier
Destination service text data is determined as target exception text.It therefore, can be by the target that gets in entire quality check process
Quality inspection rule, rapidly finds out the thin item in the presence of error, and to the thin Xiang Jinhang that these find out from destination service text data
Label to obtain above-mentioned first identifier and second identifier, and then the destination service text data for carrying label can be determined as
Target exception text.In other words, due to carrying first identifier and second identifier in target exception text, to facilitate subsequent
Quality inspection personnel can quickly orient the thin item of error according to these labels, to reduce the workload of quality inspection personnel, and improve matter
Examine efficiency.
Fig. 3 is referred to, Fig. 3 is the flow diagram of another data processing method provided in an embodiment of the present invention.Such as Fig. 3
Shown, this method may include:
Step S301 obtains at least one service voice data, selects one from least one described service voice data
A service voice data obtain the speech spectral characteristics in the target speech data as target speech data;
Specifically, server can obtain the interaction language between all customer services and user from the customer service of multiple support channels
Sound data, i.e., all service voice data can select a service voice data from the service voice data got
(the primary complete call between customer service and user) is used as target speech data, and can be located in advance to target speech data
Reason, including non-speech data (the i.e. customer service and remaining in user interaction process, in ambient enviroment in removal target speech data
Sound, such as people walk sound) and to target speech data carry out sub-frame processing (frame length be generally 10 to 30 milliseconds it
Between), it is believed that each frame voice data is the smoothly regularity of distribution of the relevant feature parameters of that is, each frame voice data
It is consistent.It, can be from each frame voice data when extracting speech spectral characteristics to the target speech data after sub-frame processing
In extract the feature that can characterize the frame voice data, and then the characteristic sequence of available target speech data, i.e.,
Speech spectral characteristics (for example, mel-frequency cepstrum coefficient, perception linear predictor coefficient etc.).
Step S302 is based on the speech spectral characteristics, the first identification model, carries out vocal print to the target speech data
Identification, obtains the corresponding Application on Voiceprint Recognition result of first identification model;Include the voice frequency in the Application on Voiceprint Recognition result
The matching degree between multiple attribute spectrum signatures in spectrum signature and first identification model;
Specifically, the speech spectral characteristics of above-mentioned acquisition can be inputted into the first identification model (i.e. Application on Voiceprint Recognition model),
Since the Application on Voiceprint Recognition model has passed through sample voice data and attribute spectrum signature corresponding with the sample voice data
Training is completed, that is to say, that has had voice Speaker Identification function.Therefore it is obtained using the identification function of Application on Voiceprint Recognition model
Application on Voiceprint Recognition corresponding with target speech data is as a result, contain above-mentioned speech spectral characteristics and sound in above-mentioned Application on Voiceprint Recognition result
Matching degree in line identification model between multiple attribute spectrum signatures.Since everyone voice is in sound quality, the duration of a sound, loudness of a sound, sound
It is high all different, therefore different speech spectral characteristics can be presented in voiceprint map, Application on Voiceprint Recognition model is from sample
The characteristics of multiclass human hair sound has been arrived in study in voice data (can be understood as between the speech spectral characteristics that different people is spoken not
Together).In other words, above-mentioned Application on Voiceprint Recognition model has learnt to difference of the different people when sending out sound same that is.Example
Such as, the voice of multiple people, respectively speaker 1 are contained in sample voice data, speaker 2 ..., and (n is nature to speaker n
Number), above-mentioned speaker can be indicated with different label informations, if the corresponding label information of speaker 1 is numerical value 1, spoken
The corresponding label information of people 2 is numerical value 2 ..., and the corresponding label information of speaker n is numerical value n, and Application on Voiceprint Recognition model passes through training
It may learn the relationship between the attribute spectrum signature and corresponding label information of each speaker's voice, for the mesh newly inputted
Voice data is marked, can have been learnt by calculating in the speech spectral characteristics in the target speech data and Application on Voiceprint Recognition model
Matching degree between the corresponding attribute spectrum signature of the multiple speakers arrived, to determine speaking in target speech data
People.
Step S303, the attribute that highest matching degree will be had with the speech spectral characteristics according to the Application on Voiceprint Recognition result
Label information associated by spectrum signature, as the first role identified from the target speech data;
Specifically, special according to multiple attribute frequency spectrums in the speech spectral characteristics and Application on Voiceprint Recognition model in target speech data
The corresponding label information of attribute spectrum signature with highest matching degree can be determined as above-mentioned mesh by the matching degree between sign
Mark the first role (i.e. customer service role) in voice data.In quality inspection field, by Application on Voiceprint Recognition model by target voice
Speech spectral characteristics in data are matched with the attribute spectrum signature of all customer service voices of typing, to determine target
Customer service role in voice data, in addition, while the attribute spectrum signature with the customer service voices of typing is matched, also
It can be matched with the common words art of the customer service of typing (such as " being very glad for your service "), customer service angle can be more quickly determined
Color.It should be noted that the voice of Consumer Role and user role is contained only in each voice data in quality inspection field,
When therefore carrying out Application on Voiceprint Recognition to target speech data, it is only necessary to distinguish that for belonging to customer service role in target speech data
Part of speech, so that it may the voice for belonging to user role is determined, also It is not necessary to have to identify in target speech data
Specific contact staff, therefore, when detecting that words art is commonly used in customer service in target speech data, above-mentioned Application on Voiceprint Recognition model can be incited somebody to action
This is commonly used the corresponding speech recognition of words art and comes out, to confirm customer service role;When the common words art of customer service is not detected, can pass through
It is matched with the attribute spectrum signature of all customer service voices of typing, to determine customer service role.
Remaining role in the target speech data in addition to the first role is determined as second jiao by step S304
Color is turned the target speech data for carrying the first role and second role based on control frequency parameter, the second identification model
It is changed to service text data;
It, can will be another in target speech data specifically, after determining the customer service role in target speech data
Role is determined as second role (i.e. user role), can obtain control frequency parameter, and will be upper based on above-mentioned control frequency parameter
State target speech data the second identification model (the namely speech recognition of input for having carried out customer service role and user role differentiation
Model), since the speech recognition modeling has been subjected to through the sample voice data and the sample voice data in corpus data library
Corresponding text information training is completed, that is, has had text conversion function.Therefore it can be incited somebody to action using the speech recognition modeling
Above-mentioned target speech data is converted into service text data.Wherein, it is for oral administration to can be used for controlling the unit time for control frequency parameter
Business voice data is converted into the quantity of service text data, that is, indicates that the service voice data that will acquire are converted in batches
At service text data.
Step S305, when each service voice data at least one described service voice data are by as target language
When sound data, the corresponding service text data of each service voice data is obtained, and from the service textual data being converted to
Select a service text data as destination service text data according to middle;
Specifically, when the above-mentioned all service voice data got by as target speech data when, can use
Above-mentioned Application on Voiceprint Recognition model determines customer service role and user role in each service voice data, and will be after progress role's differentiation
Each service voice data conversion at corresponding service text data, can be selected from all service text datas being converted to
A service text data is selected as destination service text data.It is appreciated that service voice data conversion is literary at service
When notebook data, speech recognition modeling can carry out parallel processing, that is, multiple target speech datas can be inputted language simultaneously
In sound identification model, the corresponding service text data of above-mentioned multiple target speech datas, specific target voice number are obtained
Amount can be determined by the control frequency parameter of above-mentioned acquisition.
Step S306 counts the destination number for all service text datas being converted to;
Specifically, server can count institute when all service voice data are converted into service text data
There are the total quantity (i.e. destination number) of service text data, that is, the talk times between the customer service got and user.
Step S307, if detecting, the destination number is greater than backup amount threshold, by all service textual datas
Back-up processing is carried out more than the service text data of the backup amount threshold in;
Specifically, the backup amount threshold of storage service text data can be obtained, if detecting above-mentioned service text data
Total quantity be greater than above-mentioned backup amount threshold, then can will be more than above-mentioned backup amount threshold in all service text datas
It services text data and carries out back-up processing, i.e., when the total quantity for servicing text data is more than the memory space in storing data library,
The part service text data that can be will exceed is transferred in backup storing data library, carries out back-up processing.For example, counting on
The total quantity of all service text datas is 700,000, and the memory space (i.e. backup amount threshold) in storing data library is 500,000,
The 200000 service text datas that can then will exceed are transferred in backup storing data library.
Step S308 obtains corresponding first text data of the first role from the destination service text data,
And corresponding second text data of the second role;
It is taken specifically, selecting a service text data from all service text datas being converted to as target
It is engaged in after text data, corresponding first text data of first role, i.e. customer service angle can be obtained from destination service text data
The corresponding speech content of color obtains corresponding second text data of second role, the i.e. corresponding speech content of user role.It is optional
, above-mentioned first text data is not limited to whole speech contents of the customer service role in destination service text data, can be visitor
A word described in role is taken, a few words described in customer service role are also possible to, can be determined according to specific scene, here
Without limitation.Similarly, above-mentioned second text data is also not specifically limited, and can be determined according to concrete scene.
Step S309, obtains target quality inspection rule from multiple quality inspections rule, and by first text data and/or institute
The second text data is stated as target text information, and is chosen from multiple quality inspection points that target quality inspection rule is included with most
The quality inspection points of high priority carry out exception to the target text information as target quality inspection points, and based on the target quality inspection points
Detection;
Specifically, can be using above-mentioned first text data and/or the second text data as target text information.Change speech
It, target text data can be the corresponding content of text of customer service role, or the corresponding content of text of user role, also
It can be one section of continuous session content between customer service role and user role.It can be from multiple quality inspections in intellectual analysis engine
The target quality inspection rule to match with the affiliated scene of destination service text data is obtained in rule, and is obtained in target quality inspection rule
Multiple quality inspection points priority, using the quality inspection points with highest priority as target quality inspection points, and the target can be based on
Quality inspection points carry out abnormality detection target text information.Target quality inspection rule refers to for belonging to the destination service text information
Scene and the quality inspection rule being arranged may include multiple quality inspection points in target quality inspection rule, if in the institute of the target quality inspection rule
Have in quality inspection points there are a quality inspection points so that the hit situation of remaining quality inspection points all rely on it is upper after the hits of the quality inspection points
It hereafter, can be with it may be considered that the quality inspection points have highest priority in all quality inspection points that target quality inspection rule is included
Using the quality inspection points as target quality inspection points, also referred to as preposition quality inspection points.For example, (referring to that customer service is more than one in customer service hold line
Fix time and do not respond user) in scene, target quality inspection rule is for detecting to lead to the long period due to customer service self reason
The service text information of user is not responded, and target quality inspection points can be between customer service role and user role in the target quality inspection rule
Dialogue interval is more than certain time (such as 50s), and the hit of remaining quality inspection points in the target quality inspection rule all relies on the target
The hit situation of quality inspection points can be from the hit of the target quality inspection points that is, when target text information hits the target quality inspection points
Hereinafter continue to determine remaining quality inspection points, when the target text information miss target quality inspection points, then can directly determine above-mentioned mesh
Mark service remaining quality inspection points of text data miss.
Step S310, if there are the target quality inspection points for meeting target quality inspection rule in the target text information, in institute
It states and first identifier corresponding with the target quality inspection points is set in target text;
Wherein, the specific implementation of above-mentioned steps S310 may refer in embodiment corresponding to above-mentioned Fig. 2 to step
The description of S202, is not discussed here.
Step S311 using the corresponding target quality inspection points of the first identifier as starting checkpoint, and obtains the target
The corresponding detection range threshold value of quality inspection rule;
Specifically, after above-mentioned target text information hits above-mentioned target quality inspection points, can using above-mentioned target quality inspection points as
Checkpoint is originated, and obtains the detection range threshold value in target quality inspection rule for the setting of target quality inspection points, the detection range threshold
Value may include the first detection range threshold value and/or the second detection range threshold value, wherein the first detection range threshold value can refer to
Starting checkpoint starts the range threshold detected forward, and the second examination scope threshold value can refer to originate checkpoint and start backward
The range threshold of detection.
Step S312, by it is described starting checkpoint and the detection range threshold value between text information be determined as with it is described
The associated associated text information of target text information;
Specifically, when detection range threshold value includes the first detection range threshold value, it can be by before above-mentioned starting checkpoint the
Text information between one detection range threshold value is determined as associated text information associated with above-mentioned target text information;Work as inspection
It, can will be between the second detection range threshold value after above-mentioned starting checkpoint when to survey range threshold include the second detection range threshold value
Text information is determined as associated text information associated with above-mentioned target text information;When detection range threshold value includes the first inspection
It, can be by the text between the first detection range threshold value before above-mentioned starting checkpoint when surveying range threshold and the second detection range threshold value
Text information after this information and above-mentioned starting checkpoint between the second detection range threshold value is determined as associated text information.Example
Such as, starting checkpoint be above-mentioned destination service text data in the 10th word, the first detection range threshold value be 3 words, second
Detection range threshold value is 5 words, then when detection range threshold value includes the first detection range threshold value, associated text information is target clothes
The 7th the-the 10 word of word being engaged in text data, when detection range threshold value includes the second detection range threshold value, associated text letter
Breath is the 10th the-the 15 word of word in destination service text data, detection range threshold value including the first detection range threshold value and
When the second detection range threshold value, associated text information is the 7th the-the 15 word of word in destination service text data.
Step S313 is obtained in target quality inspection rule and is associated with quality inspection points with incidence relation with the target quality inspection points,
And the associated text information is carried out abnormality detection according to the association quality inspection points;
Specifically, after associated text information associated with above-mentioned target text information has been determined, available mesh
There are the quality inspection points that are associated with of incidence relation with above-mentioned target quality inspection points, that is, dependent on target quality inspection points in mark quality inspection rule
Remaining quality inspection points of hit situation, and associated text information is carried out abnormality detection according to above-mentioned association quality inspection points, judge to be associated with
The hit situation of above-mentioned association quality inspection points in text information.For example, in customer service hold field of line scape, association quality inspection points may include
The case where user actively allows customer service to wait is excluded, as excluded " I asks for bank card " etc. described in user, if associated text is believed
" I asks for bank card " similar sentence is not detected in breath, then shows that above-mentioned related text hits the association quality inspection points, it is no
Then, then miss association quality inspection points.
Step S314, if there are the association quality inspection points met in the target quality inspection rule in the associated text information,
Second identifier corresponding with the association quality inspection points is then set in the associated text information, and first mark will be carried
Know, the destination service text data of the second identifier is determined as target exception text.
Wherein, the specific implementation of above-mentioned steps S314 may refer in embodiment corresponding to above-mentioned Fig. 2 to step
The description of S204, is not discussed here.
For example, by taking the quality inspection case under customer service hold field of line scape as an example, in order to accurately check hold
The service work order (as above-mentioned service voice data) of line, for a service work order, concrete implementation mode is as follows:
The voice serviced in work order is divided into two kinds of roles, respectively role 1 and role 2 by sound groove recognition technology in e, and
It is matched by the common words art of customer service vocal print and customer service with typing, to determine the customer service role in service work order,
Turn text techniques by voice and the voice data in above-mentioned service work order is converted into textual data, and to the customer service in text data
Role is marked, thus may determine that another role in service work order is user role;Check customer service angle in service work order
Dialogue interval was more than 50 seconds between color and user role, and latter sentence is the session of customer service speech, is labeled as checkpoint 1=
Ture, if miss is labeled as checkpoint 1=false,;As checkpoint 1=ture, 5 above of checkpoint 1 are examined
Rope excludes the case where customer service of user's initiative waits and such as " I asks for bank card " is recorded as the result of checkpoint 2;Work as inspection
When making an inventory of 1=ture, context 3 of checkpoint 1 are retrieved, excludes there are the feelings shown waiting user in customer service speech
Condition, such as " you input ", " having verification to arrive ", are recorded as the result of checkpoint 3;As checkpoint 1=ture, to checkpoint
3 above of 1 are retrieved, and exclude to allow user to re-enter interactive voice answering (Interactive in customer service speech
Voice Response, IVR) key the case where, such as " please being inputted according to voice prompting ", be recorded as the result of checkpoint 4;Work as inspection
When making an inventory of 1=ture, 3 above of checkpoint 1 are retrieved, excludes to allow user to wait customer service to another in customer service speech
Number puts through the case where verifying, such as " I gives you the verifying of beating over of another number, woulds you please listen a Duan Yinle ", is recorded as checkpoint
5 result;It finally needs to be combined these inspection results, combinatorial formula are as follows: 1and (not (2or3or4or5)), 1-5 refers to respectively
The hit situation of a checkpoint carries out Boolean calculation and obtains a result, such as ture and (not (false or false or
False or false))=ture, then hit the rule of customer service hold line, be considered as detection, detection indicates customer service work order quilt
It is confirmed as wrong.Wherein, checkpoint 1 is target quality inspection points, and 1=ture expression in checkpoint hits target quality inspection points, checks
Point 1=false indicates target miss quality inspection points, and the checkpoint checkpoint 2- 5 is to be associated with quality inspection points, mentioned above above 3
Sentence or it is above 5 be above-mentioned first detection range threshold value, hereafter 3 be the second detection range threshold value, said combination formula
1and (not (2or3or4or5)) is the target quality inspection rule under customer service hold field of line scape.When the service work order is hit target
When quality inspection rule, it can be confirmed that the service work order is the service work order of error;When the service work order is the case where checkpoint 1 is hit
Under, the miss target quality inspection rule it is different can be determined as target using the service work order as the service work order of doubtful error
Chang Wenben.
Show please also refer to the frame structure that Fig. 4, Fig. 4 are a kind of customer service quality inspection methods provided in an embodiment of the present invention
It is intended to.As shown, the quality inspection conversation analysis service 43 in intellectual analysis engine can be from calling in customer service quality check process
Service voice data are obtained in the heart 41 can also be timed the service voice data in call center 41 standby at the same time
Part, all service voice data are stored in data backup server 42, so as between subsequent query client and user
Primary voice data.Quality inspection conversation analysis service 43 can be based on after getting service voice data in call center 41
Control 431 parameter of frequency carries out speech recognition 45 to the service voice data got, by service voice data conversion at service
Text data can carry out Application on Voiceprint Recognition, that is, service language during carrying out speech recognition 45 to service text data
Sound data carry out Role judgement 434, and specific 434 process of Role judgement and 45 process of speech recognition may refer to above-mentioned Fig. 3 institute
Step S301- step S305 in corresponding embodiment, is not discussed here.Service voice data are being converted to textual data
According to rear, the service text data store after conversion can be turned in text primary server 46 in voice, when service text data
Total quantity is too big, and then when under causing voice to turn text primary server 46 and can not store, the part service text that can be will exceed
Notebook data is stored in voice and turns in text backup server 47, and the available voice of quality inspection conversation analysis service 43 turns the main clothes of text
Business device 46 and voice turn the service text data in text backup server 47, and divide the service text data got
Analysis, wherein algorithm analysis 433 can indicate to obtain the corresponding target quality inspection rule of service text data, including to semantic matches
It is associated with quality inspection points and carries out intelligent recognition, detection abnormal 432 is represented by according to above-mentioned target quality inspection rule to service text data
It carrying out abnormality detection, specific detection process may refer to the step S308- step S314 in embodiment corresponding to above-mentioned Fig. 3,
It is not discussed here.43 pairs of the service of quality inspection conversation analysis service text datas carry out analysis generation quality inspection result and
It is provided with the service text data of identification information in quality check process, can store in data warehouse 44, meanwhile, quality inspection session
Analysis Service 43 can also obtain service text data from data warehouse 44 and be trained, to update quality inspection rule.
It in embodiments of the present invention, can by carrying out Application on Voiceprint Recognition and speech recognition to the service voice data got
To obtain the corresponding service text data of service voice data, and it is corresponding to obtain customer service role from service text data
First text data the second text data corresponding with user role, and then can be according to the first text data and the second text data
It determines target text data, is determined according to target quality inspection rule with the presence or absence of target quality inspection points in above-mentioned target text information, if
In the presence of corresponding with target quality inspection points first identifier being then arranged in above-mentioned target text information, and then can obtain and above-mentioned mesh
Mark the associated associated text information of text information, according to above-mentioned target quality inspection rule determine in above-mentioned associated text information whether
In the presence of association quality inspection points, and if it exists, the then setting second identifier corresponding with association quality inspection points in above-mentioned associated text information, into
And the destination service text data for carrying first identifier and second identifier can be determined as target exception text.Therefore, whole
, can be by the different role in the target quality inspection rule and destination service text data that get in a quality check process, it can be with
The thin item in the presence of error is rapidly found out from destination service text data, and the thin item that these find out is marked, with
To above-mentioned first identifier and second identifier, and then it is extremely literary the destination service text data for carrying label can be determined as target
This.In other words, due to carrying first identifier and second identifier in target exception text, so that facilitating subsequent quality inspection personnel can
Quickly to orient the thin item of error according to these labels, to reduce the workload of quality inspection personnel, and quality inspection efficiency is improved.
Fig. 5 is referred to, Fig. 5 is the flow diagram of another data processing method provided in an embodiment of the present invention.Such as Fig. 5
Shown, this method may include:
Step S501 obtains at least one service voice data, and will be every at least one described service voice data
A service voice data are respectively converted into service text data, and a service is selected from the service text data being converted to
Text data is as destination service text data;
Wherein, the specific implementation of above-mentioned steps S501 may refer in embodiment corresponding to above-mentioned Fig. 3 to step
The description of S301- step S305, is not discussed here.
Step S502 obtains destination service text data, and target text letter is obtained from the destination service text data
Breath, and the target text information is carried out abnormality detection according to the target quality inspection points in target quality inspection rule;
Step S503, if there are the target quality inspection points for meeting target quality inspection rule in the target text information, in institute
It states and first identifier corresponding with the target quality inspection points is set in target text;
Step S504 obtains association text associated with the target text information from the destination service text data
This information, and the associated text information is carried out abnormality detection according to the association quality inspection points in the target quality inspection rule;
Step S505, if there are the association quality inspection points met in the target quality inspection rule in the associated text information,
Second identifier corresponding with the association quality inspection points is then set in the associated text information, and first mark will be carried
Know, the destination service text data of the second identifier is determined as target exception text;
Wherein, the specific implementation of above-mentioned steps S502- step S505 may refer in embodiment corresponding to above-mentioned Fig. 2
Description to step S201- step S204, is not discussed here.
The target exception text is added to abnormal text data set by step S506, and by the abnormal text data
All target exception texts concentrated are sent to quality inspection terminal;
Specifically, above-mentioned target exception text can be added to abnormal textual data after determining target exception text
According to collection, the exception text data set includes all abnormal texts in quality check process, i.e., the service text of all doubtful errors
Data, all target exception texts that abnormal text data is concentrated are sent to quality inspection terminal, that is, are sent to Quality Inspector's station.
Quality inspection terminal can randomly select a part from all target exception texts received after receiving target exception text
Target exception text carries out manual review, and the process of quality inspection is main including the following steps:
It is found and is needed in daily quality inspection using the type of error of quality inspection rule detection (for example, service gift by manual type
Instrument type, operation error type etc.);It further analyzes the observation point in artificial quality inspection, observation point is organized into each inspection
Point, that is, the several emphasis mainly checked in artificial quality inspection are organized into checkpoint;Configure each checkpoint and checkpoint
Context dependency between checkpoint, that is, determine the incidence relation between checkpoint, for example, can be n above
Sentence, hereafter n or context n etc. (n is natural number);According to or and non-sum preferentially calculate symbol design conditions combination check
Point determines quality inspection rule, to reach quality inspection purpose;A target can be selected from all target exception texts got
Abnormal text verifies above-mentioned quality inspection rule hit situation immediately, for example, the configuration page in quality inspection terminal has a quick verifying
Function, after inputting a target exception text, can return rule hit situation and checkpoint hit situation;According to
The final quality inspection of above-mentioned target exception text can be confirmed as a result, namely for the quality inspection in intellectual analysis engine in hit situation
Rule provides the service text data of doubtful error, can be confirmed whether the service text data is error after manual review
Service text data.Above-mentioned target exception text can be carried out the quality inspection result and correspondence after manual review by quality inspection terminal
Quality inspection rule be sent to intellectual analysis engine.
Step S507 obtains quality inspection terminal exception quality inspection points based on determined by all target exception texts, and base
The target quality inspection rule is updated in the abnormal quality inspection points;
Specifically, quality inspection result and corresponding quality inspection rule after receiving target exception text and carrying out manual review
Afterwards, the abnormal quality inspection points in artificial quality inspection rule, i.e. newly-increased inspection can be obtained according to the regular hit situation in quality inspection result
Abnormal quality inspection points are added in target quality inspection rule, to realize the update to target quality inspection rule by point.For example, moving back trip field
Jing Zhong, the target quality inspection points in target quality inspection rule are to occur " moving back trip " keyword, target matter in the corresponding speech content of user
The language that association quality inspection points in cautious can be pacified for exclusion customer service, such as: " thank you all the time to the branch of the game
Hold ", " it is desirable that you can consider further that whether determine move back trip ", when occurring above-mentioned association quality inspection in destination service text data
Point is similar pacify language " may I ask be because why reason, have anything that can help you ", target quality inspection rule nothing
Method judges it, therefore the destination service text data is determined as target exception text, and after carrying out manual review,
Quality inspection goes out the destination service text data and does not malfunction, and belongs to normal text, then can be according to the quality inspection of manual review as a result, by upper
State " may I ask be because why reason, have anything that can help you " checkpoint new as one be added to above-mentioned target
In quality inspection rule.
The target exception text is added to abnormal text data set by step S508, obtains the abnormal text data
Collect corresponding history exception text, and training sample set is obtained based on the history exception text;
Specifically, after above-mentioned target exception text is added to abnormal text data set, available exception textual data
According to all history exception texts of concentration, and all history exception texts that will acquire are as training sample.Wherein, above-mentioned to go through
History exception text refers to the target exception text confirmed in all quality check process before the secondary quality inspection.
Step S509 is trained the corresponding Checking model of the target quality inspection rule based on the training sample set,
And the Checking model is updated;The abnormal quality inspection points carried in updated Checking model are used for the target quality inspection
Rule is updated.
Specifically, based on the history exception text that above-mentioned training sample is concentrated, it can be corresponding to above-mentioned target quality inspection rule
Checking model be trained, each history exception text can be used as a sample data, to above-mentioned Checking model into
Row training, after the completion of training, above-mentioned Checking model can obtain better model parameter.In other words, the matter after the completion of training
Model is examined instead of the corresponding Checking model of above-mentioned target quality inspection rule, completes the renewal process of Checking model.It updates
Later abnormal quality inspection points are carried in Checking model, above-mentioned goal rule can be carried out more according to above-mentioned abnormal quality inspection points
Newly.Wherein, above-mentioned Checking model can be convolutional neural networks, deep neural network, generation confrontation network etc..By above-mentioned target
The service text data (i.e. target exception text) of all doubtful mistakes that quality inspection rule quality inspection comes out is summarized, and is passed through pair
The service text data of these doubtful mistakes carries out deep learning, that is, passes through these doubtful mistakes of Checking model autonomous learning
Service text data in feature, obtain better model parameter, and then optimize above-mentioned target quality inspection rule, improve quality inspection matter
Amount.For example, can be identified by the ball of red, navy blue, light blue three kinds of colors now with a kind of mechanism in a chest
Red ball and navy blue ball, but can not identify ball azury can be right the characteristics of by learning ball azury
Above-mentioned mechanism optimizes, and so that updated mechanism is identified ball azury, navy blue will be such as directed in original mechanism
The identification method of ball is updated to the recognition mechanism of blue ball (including navy blue ball and ball azury).If the mechanism represents mesh
Quality inspection rule is marked, then red ball can indicate to confirm the text data not malfunctioned in service text data, and navy blue ball can
To indicate to confirm in service text data the text data of error, ball azury can indicate in service text data it is doubtful go out
Wrong text data, the characteristics of learning ball azury can be expressed as using Checking model to the text data of doubtful error into
Row study, updated mechanism can indicate the target quality inspection rule after updating.
It should be noted that the quality inspection result by manual review of above-mentioned steps S506- step S507 description is to target
Quality inspection rule be updated and step S508- step S509 description by history exception text carry out deep learning, no
Disconnected optimization aim quality inspection rule is two kinds of mutual independent target quality inspection Policy Updates modes, each other without interruption.Change speech
It, above two target quality inspection Policy Updates mode can be carried out successively, can also be carried out simultaneously.
It is the frame of another customer service quality inspection method provided in an embodiment of the present invention please also refer to Fig. 6 a- Fig. 6 b
Structural schematic diagram.As shown in Figure 6 a, which may include: the acquisition intersection record data from each channel customer service 601
602, by carrying out data prediction 603 to the intersection record data 602 got, pretreatment includes the voice that will acquire
6031 data are converted into text by automatic speech recognition technology 6032 (Automatic Speech Recognition, ASR)
6033 data may refer to the detailed process that voice turns text the step S301- step in embodiment corresponding to above-mentioned Fig. 3
S305 is not discussed here.The text data of conversion is input to quality inspection library 604, in quality inspection library 604, intellectual analysis
Engine 6041 can analyze the text data after conversion, obtain the corresponding quality inspection of each text data as a result, specific
Quality check process may refer to the step S308- step S314 in embodiment corresponding to above-mentioned Fig. 3, be not discussed here.It is right
In the text data for the doubtful mistake that 6041 quality inspection of intellectual analysis engine comes out, can be carried out by Quality Inspector's station 6042 manually
Anti- review, can be updated the quality inspection rule in intellectual analysis engine 6041 according to the quality inspection result after manual review, i.e.,
The quality inspection rule in intellectual analysis engine 6041 is fed, above-mentioned Quality Inspector's station 6042 is the matter in embodiment corresponding to Fig. 3
Terminal is examined, the specific implementation process of manual review and more fresh target quality inspection rule may refer to real corresponding to above-mentioned Fig. 5
The step S506- step S509 in example is applied, is not discussed here.Operating process data 605 are mainly used for controlling interactive note
Record data flow of the data 602 between the quality check process into quality inspection library 604.
Optionally, the intersection record data 602 got from each channel customer service 601 also may include textual data
According to, for example, from online customer service obtain data, if the intersection record data 602 got be text data, need not execute
Data prediction 603 is stated, can be directly inputted in quality inspection library.
As shown in Figure 6 b, which can indicate a specific quality check process, and the quality inspection rule 606 in quality check process can
With using checkpoint, the scope of application 6065 and logical relation 6066 (as and/or/it is non-) combination in conversation analysis model 607
Rule verified, above-mentioned checkpoint is the minimum unit for forming quality inspection rule 606, can be any service work order content
It checks, including but not limited to keyword 6061, service work order attribute 6062, dialogue interval 6063, semantic matches 6064 wait all
Used cross-check information in artificial quality inspection, the rule in conversation analysis model 607 may include multiple rules, for checking
Problem under each scene services work order, for example, rule A 6071 can indicate hold line rule, regular B 6072 can indicate to move back
Trip rule, regular C 6073 can indicate recommended products rule.Conversation analysis model 607 can be based on above-mentioned rule to quality inspection library
Service work order in 608 carries out quality inspection, can obtain clue list (the service work order for the doubtful mistake that i.e. intelligent quality inspection comes out,
It is equal to the abnormal text referred in above-described embodiment), artificial intelligence (Artificial Intelligence, AI) clue 609
In include multiple clues, that is, multiple clue lists can be gone out with quality inspection, such as clue 1 6091, clue 2 6092, clue 3 6093,
It should be noted that rule A 6071, regular B 6072, regular C 6073 and clue 1 6091, clue 2 6092, clue 3
6093 be not one-to-one relationship, for the same rule, multiple clue lists can be gone out with quality inspection, for conversation analysis model
The clue list that 607 quality inspections come out can carry out manual review 610 for transmission to artificial quality inspection bench, according to the quality inspection knot of manual review
Fruit, regular hit situation, checkpoint hit situation can be updated quality inspection rule 606, such as increase checkpoint.It can also lead to
It crosses machine learning general character difference 611 to optimize quality inspection rule 606, i.e., clue list is learnt using machine learning, it is right
Quality inspection rule 606 carries out intelligent optimization.
It is a kind of timing signal of customer service quality inspection method provided in an embodiment of the present invention please also refer to Fig. 7, Fig. 7
Figure.As shown, this method may include:
Step S701, front end send work order content to backstage;
Specifically, after front end gets work order content (i.e. service work order content) from each channel customer service, it can be backward
Platform sends the service work order content got.
Work order information is checked on step S702, backstage;
Specifically, backstage is after the service work order content for receiving foreground transmission, can first detection service work order data class
The service work order is converted to text data, and then check service work order information if service work order is voice data by type;If
Service work order is text data, then can directly check service work order information.The service work order information of verification mainly includes service work
Single remarks, service work order label, the session content etc. in service work order.
Step S703, extracting rule;
Specifically, the quality inspection rule to match with the service work order information can be extracted according to service work order information.For example,
For the service work order information of Products Show class, need to extract Products Show rule;For upgrading the service work order information of class, need
Extract upgrade rule etc..
Step S704, front end is to backstage delivery rules;
Specifically, the quality inspection rule extracted can be passed to backstage by front end, so that backstage can be according to quality inspection rule to clothes
List of working carries out quality inspection.
Step S705, backstage verification check logic;
Specifically, can check each checkpoint in quality inspection rule after the quality inspection rule that platform is sent upon receipt of backstage and patrol
The relationship of collecting, and then determine the hit situation of each checkpoint and the hit situation of quality inspection rule, the miss quality inspection is advised
Service work order then, it may be determined that not malfunction, for the service work order of the quality inspection rule of hit, it may be determined that determine
Mistake can be used as the service work order of doubtful error, that is, quality inspection clue for the rule also in adjustment.
Step S706 sends quality inspection clue to artificial from the background;
Specifically, quality inspection clue can be sent to artificial matter after the result that intelligent quality inspection goes out above-mentioned service work order by backstage
Station is examined, that is, the service work order of doubtful error is sent to artificial quality inspection station, allows quality inspection personnel to carry out manually multiple
Core.
Step S707 manually sends quality inspection result to backstage;
Specifically, the quality inspection result after manual review can be returned to backstage by artificial quality inspection station, it from the background can be to people
Quality inspection result after work review is analyzed.
Step S708, adjustment rule;
Specifically, the adjustment information of quality inspection rule can be obtained according to the quality inspection result after manual review, for example, increasing inspection
Make an inventory of or optimize checkpoint.
Step S709, front end carry out Policy Updates.
Specifically, front end can be updated quality inspection rule according to above-mentioned adjustment information, the improvement of quality inspection rule is obtained
Scheme or new rule.
Wherein, above-mentioned front end and backstage correspond to the server 100 in embodiment corresponding to above-mentioned Fig. 1, manually correspond to
Quality inspection terminal 200 in embodiment corresponding to above-mentioned Fig. 1.
The embodiment of the present invention is by obtaining the target text information in destination service text data, according to target quality inspection rule
Determine in above-mentioned target text information with the presence or absence of target quality inspection points, and if it exists, then in above-mentioned target text information setting with
The corresponding first identifier of target quality inspection points, and then associated text information associated with above-mentioned target text information, root can be obtained
It determines in above-mentioned associated text information according to above-mentioned target quality inspection rule with the presence or absence of association quality inspection points, and if it exists, then in above-mentioned pass
Join in text information setting be associated with the corresponding second identifier of quality inspection points, and then can will carry first identifier and second identifier
Destination service text data is determined as target exception text.It therefore, can be by the target that gets in entire quality check process
Quality inspection rule, rapidly finds out the thin item in the presence of error, and to the thin Xiang Jinhang that these find out from destination service text data
Label to obtain above-mentioned first identifier and second identifier, and then the destination service text data for carrying label can be determined as
Target exception text.In other words, due to carrying first identifier and second identifier in target exception text, to facilitate subsequent
Quality inspection personnel can quickly orient the thin item of error according to these labels, to reduce the workload of quality inspection personnel, and improve matter
Efficiency is examined, and can be by the abnormal quality inspection points in manual review or by carrying out deep learning acquisition to history exception text
Abnormal quality inspection points are updated target quality inspection rule, and quality inspection quality can be improved.
Fig. 8 is referred to, Fig. 8 is a kind of structural schematic diagram of data processing equipment provided in an embodiment of the present invention.Such as Fig. 8 institute
Show, which may include: that text information obtains module 101, the first setup module 102, and related information obtains mould
Block 103, the second setup module 104;
Text information obtains module 101, for obtaining destination service text data, from the destination service text data
Target text information is obtained, and abnormal inspection is carried out to the target text information according to the target quality inspection points in target quality inspection rule
It surveys;
First setup module 102, if for there is the target matter for meeting target quality inspection rule in the target text information
It is cautious, then first identifier corresponding with the target quality inspection points is set in the target text;
Related information obtains module 103, believes for obtaining from the destination service text data with the target text
The associated associated text information of manner of breathing, and according to the association quality inspection points in the target quality inspection rule to the associated text information
It carries out abnormality detection;
Second setup module 104, if meeting in the target quality inspection rule for existing in the associated text information
Quality inspection points are associated with, then second identifier corresponding with the association quality inspection points is set in the associated text information, and will carry
The first identifier, the second identifier destination service text data be determined as target exception text.
Wherein, text information obtains module 101, the first setup module 102, and related information obtains module 103, the second setting
The concrete function implementation of module 104 may refer to the step S201- step S204 in embodiment corresponding to above-mentioned Fig. 2, this
In no longer repeated.
As shown in figure 8, the data processing equipment 1 can also include: text conversion module 105, statistical module 106, backup
Module 107, abnormal text sending module 108, the first update module 109, training sample determining module 110, the second update module
111;
Text conversion module 105, for obtaining at least one service voice data, and will at least one described service voice
Each service voice data in data are respectively converted into service text data, and select from the service text data being converted to
A service text data is selected as destination service text data;
Statistical module 106, for counting the destination number for all service text datas being converted to;
Backup module 107, if for detecting that the destination number is greater than backup amount threshold, by all services
Back-up processing is carried out more than the service text data of the backup amount threshold in text data;
Abnormal text sending module 108, for the target exception text to be added to abnormal text data set, and by institute
It states all target exception texts that abnormal text data is concentrated and is sent to quality inspection terminal;
First update module 109 is abnormal based on determined by all target exception texts for obtaining the quality inspection terminal
Quality inspection points, and the target quality inspection rule is updated based on the abnormal quality inspection points;
Training sample determining module 110 obtains institute for the target exception text to be added to abnormal text data set
The corresponding history exception text of abnormal text data set is stated, and training sample set is obtained based on the history exception text;
Second update module 111, for being based on the training sample set to the corresponding quality inspection mould of the target quality inspection rule
Type is trained, and is updated to the Checking model;The abnormal quality inspection points carried in updated Checking model for pair
The target quality inspection rule is updated.
Wherein, the concrete function implementation of text conversion module 105, statistical module 106, backup module 107 can join
See the step S301- step S307 in embodiment corresponding to above-mentioned Fig. 3, is not discussed here, abnormal text sending module
108, the first update module 109, training sample determining module 110, the concrete function implementation of the second update module 111 can be with
Referring to the step S506- step S509 in embodiment corresponding to above-mentioned Fig. 5, it is not discussed here.
As shown in figure 8, it may include: first acquisition unit 1011, target quality inspection points that above-mentioned text information, which obtains module 101,
Determination unit 1012;
First acquisition unit 1011, for obtaining the first role corresponding from the destination service text data
One text data and corresponding second text data of the second role;
Target quality inspection points determination unit 1012, for obtaining target quality inspection rule from multiple quality inspection rules, and will be described
First text data and/or second text data as target text information, and included from target quality inspection rule it is more
Choosing in a quality inspection points has the quality inspection points of highest priority as target quality inspection points, and based on the target quality inspection points to described
Target text information carries out abnormality detection.
Wherein, the concrete function implementation of first acquisition unit 1011, target quality inspection points determination unit 1012 can join
See the step S308- step S309 in embodiment corresponding to above-mentioned Fig. 3, is not discussed here.
As shown in figure 8, it may include: checkpoint determination unit 1031 that above-mentioned related information, which obtains module 103, first is determined
Unit 1032, abnormality detecting unit 1033;
Checkpoint determination unit 1031, for using the corresponding target quality inspection points of the first identifier as originate checkpoint,
And obtain the corresponding detection range threshold value of the target quality inspection rule;
First determination unit 1032, for by it is described starting checkpoint and the detection range threshold value between text information
It is determined as associated text information associated with the target text information;
Abnormality detecting unit 1033 has incidence relation with the target quality inspection points for obtaining in target quality inspection rule
Quality inspection points are associated with, and the associated text information is carried out abnormality detection according to the association quality inspection points.
Wherein, the concrete function of checkpoint determination unit 1031, the first determination unit 1032, abnormality detecting unit 1033 is real
Existing mode may refer to the step S311- step S313 in embodiment corresponding to above-mentioned Fig. 3, be not discussed here.
As shown in figure 8, above-mentioned text conversion module 105 may include: selecting unit 1051, Application on Voiceprint Recognition unit 1052,
First role determination unit 1053, second role determination unit 1054, text data determination unit 1055;
Selecting unit 1051, for selected from least one described service voice data a service voice data as
Target speech data, and obtain the speech spectral characteristics in the target speech data;
Application on Voiceprint Recognition unit 1052 is based on the speech spectral characteristics, the first identification model, to the target speech data
Application on Voiceprint Recognition is carried out, the corresponding Application on Voiceprint Recognition result of first identification model is obtained;It include institute in the Application on Voiceprint Recognition result
State the matching degree between multiple attribute spectrum signatures in speech spectral characteristics and first identification model;
First role determination unit 1053, for that will be had with the speech spectral characteristics according to the Application on Voiceprint Recognition result
Label information associated by the attribute spectrum signature of highest matching degree, as identified from the target speech data
One role;
Second role determination unit 1054, by the remaining role in the target speech data in addition to the first role
It is determined as second role, the mesh of the first role and second role will be carried based on control frequency parameter, the second identification model
Poster sound data are converted to service text data;
Text data determination unit 1055, for when each service voice number at least one described service voice data
According to by as target speech data when, obtain the corresponding service text data of each service voice data.
Wherein, selecting unit 1051, Application on Voiceprint Recognition unit 1052, first role determination unit 1053, second role determine
The concrete function implementation of unit 1054, text data determination unit 1055 may refer in embodiment corresponding to above-mentioned Fig. 3
Step S301- step S305, is not discussed here
The embodiment of the present invention is by obtaining the target text information in destination service text data, according to target quality inspection rule
Determine in above-mentioned target text information with the presence or absence of target quality inspection points, and if it exists, then in above-mentioned target text information setting with
The corresponding first identifier of target quality inspection points, and then associated text information associated with above-mentioned target text information, root can be obtained
It determines in above-mentioned associated text information according to above-mentioned target quality inspection rule with the presence or absence of association quality inspection points, and if it exists, then in above-mentioned pass
Join in text information setting be associated with the corresponding second identifier of quality inspection points, and then can will carry first identifier and second identifier
Destination service text data is determined as target exception text.It therefore, can be by the target that gets in entire quality check process
Quality inspection rule, rapidly finds out the thin item in the presence of error, and to the thin Xiang Jinhang that these find out from destination service text data
Label to obtain above-mentioned first identifier and second identifier, and then the destination service text data for carrying label can be determined as
Target exception text.In other words, due to carrying first identifier and second identifier in target exception text, to facilitate subsequent
Quality inspection personnel can quickly orient the thin item of error according to these labels, to reduce the workload of quality inspection personnel, and improve matter
Efficiency is examined, and can be by the abnormal quality inspection points in manual review or by carrying out deep learning acquisition to history exception text
Abnormal quality inspection points are updated target quality inspection rule, and quality inspection quality can be improved.
Fig. 9 is referred to, Fig. 9 is the structural schematic diagram of another data processing equipment provided in an embodiment of the present invention.Such as Fig. 9
Shown, which can correspond to the server 100 in embodiment corresponding to above-mentioned Fig. 1, the data processing
Device 1000 may include: processor 1001, network interface 1004 and memory 1005, in addition, above-mentioned data processing equipment
1000 can also include: user interface 1003 and at least one communication bus 1002.Wherein, communication bus 1002 for realizing
Connection communication between these components.Wherein, user interface 1003 may include display screen (Display), keyboard
(Keyboard), optional user interface 1003 can also include standard wireline interface and wireless interface.Network interface 1004 is optional
May include standard wireline interface and wireless interface (such as WI-FI interface).Memory 1004 can be high speed RAM memory,
It is also possible to non-labile memory (non-volatile memory), for example, at least a magnetic disk storage.Memory
1005 optionally can also be that at least one is located remotely from the storage device of aforementioned processor 1001.As shown in figure 9, as one kind
It may include operating system, network communication module, Subscriber Interface Module SIM in the memory 1005 of computer storage medium and set
Standby control application program.
In data processing equipment 1000 as shown in Figure 9, network interface 1004 can provide network communication function;And user
Interface 1003 is mainly used for providing the interface of input for user;And processor 1001 can be used for calling and store in memory 1005
Equipment control application program, with realize:
Destination service text data is obtained, the acquisition target text information from the destination service text data, and according to
Target quality inspection points in target quality inspection rule carry out abnormality detection the target text information;
If there are the target quality inspection points for meeting target quality inspection rule in the target text information, in the target text
It is middle that first identifier corresponding with the target quality inspection points is set;
Associated text information associated with the target text information is obtained from the destination service text data, and
The associated text information is carried out abnormality detection according to the association quality inspection points in the target quality inspection rule;
If there are the association quality inspection points met in the target quality inspection rule in the associated text information, in the pass
Join setting second identifier corresponding with the association quality inspection points in text information, and the first identifier, described second will be carried
The destination service text data of mark is determined as target exception text.
In one embodiment, above-mentioned processor 1001 can also be realized:
Obtain at least one service voice data, and by each service voice at least one described service voice data
Data are respectively converted into service text data, and select a service text data to make from the service text data being converted to
For destination service text data.
In one embodiment, above-mentioned processor 1001 can also be realized:
Count the destination number for all service text datas being converted to;
If detecting, the destination number is greater than backup amount threshold, will be more than institute in all service text datas
The service text data for stating backup amount threshold carries out back-up processing.
In one embodiment, above-mentioned processor 1001 execute it is above-mentioned will be at least one described service voice data
When each service voice data are respectively converted into service text data, following steps are specifically executed:
It selects a service voice data as target speech data from least one described service voice data, and obtains
Take the speech spectral characteristics in the target speech data;
Based on the speech spectral characteristics, the first identification model, Application on Voiceprint Recognition is carried out to the target speech data, is obtained
The corresponding Application on Voiceprint Recognition result of first identification model;It include the speech spectral characteristics and institute in the Application on Voiceprint Recognition result
State the matching degree between multiple attribute spectrum signatures in the first identification model;
The attribute spectrum signature that will there is highest matching degree with the speech spectral characteristics according to the Application on Voiceprint Recognition result
Associated label information, as the first role identified from the target speech data;
Remaining role in the target speech data in addition to the first role is determined as second role, based on control
The target speech data for carrying the first role and second role is converted to service text by frequency parameter processed, the second identification model
Notebook data;
When each service voice data at least one described service voice data by as target speech data when,
Obtain the corresponding service text data of each service voice data.
In one embodiment, above-mentioned processor 1001 execute it is above-mentioned from the destination service text data obtain with
The associated associated text information of target text information, and according to the association quality inspection points in the target quality inspection rule to institute
When stating associated text information and carrying out abnormality detection, following steps are specifically executed:
Using the corresponding target quality inspection points of the first identifier as starting checkpoint, and it is right to obtain the target quality inspection rule
The detection range threshold value answered;
Text information between the starting checkpoint and the detection range threshold value is determined as and the target text
The associated associated text information of information;
It obtains in target quality inspection rule and is associated with quality inspection points with incidence relation with the target quality inspection points, and according to described
Association quality inspection points carry out abnormality detection the associated text information.
In one embodiment, above-mentioned processor 1001 can also be performed:
The target exception text is added to abnormal text data set, and is owned what the abnormal text data was concentrated
Target exception text is sent to quality inspection terminal;
Quality inspection terminal exception quality inspection points based on determined by all target exception texts are obtained, and are based on the exception
Quality inspection points are updated the target quality inspection rule.
In one embodiment, above-mentioned processor 1001 can also be performed:
The target exception text is added to abnormal text data set, obtains that described abnormal text data set is corresponding goes through
History exception text, and training sample set is obtained based on the history exception text;
The corresponding Checking model of the target quality inspection rule is trained based on the training sample set, and to the matter
Inspection model is updated;The abnormal quality inspection points carried in updated Checking model are used to carry out more the target quality inspection rule
Newly.
The embodiment of the present invention is by obtaining the target text information in destination service text data, according to target quality inspection rule
Determine in above-mentioned target text information with the presence or absence of target quality inspection points, and if it exists, then in above-mentioned target text information setting with
The corresponding first identifier of target quality inspection points, and then associated text information associated with above-mentioned target text information, root can be obtained
It determines in above-mentioned associated text information according to above-mentioned target quality inspection rule with the presence or absence of association quality inspection points, and if it exists, then in above-mentioned pass
Join in text information setting be associated with the corresponding second identifier of quality inspection points, and then can will carry first identifier and second identifier
Destination service text data is determined as target exception text.It therefore, can be by the target that gets in entire quality check process
Quality inspection rule, rapidly finds out the thin item in the presence of error, and to the thin Xiang Jinhang that these find out from destination service text data
Label to obtain above-mentioned first identifier and second identifier, and then the destination service text data for carrying label can be determined as
Target exception text.In other words, due to carrying first identifier and second identifier in target exception text, to facilitate subsequent
Quality inspection personnel can quickly orient the thin item of error according to these labels, to reduce the workload of quality inspection personnel, and improve matter
Efficiency is examined, and can be by the abnormal quality inspection points in manual review or by carrying out deep learning acquisition to history exception text
Abnormal quality inspection points are updated target quality inspection rule, and quality inspection quality can be improved.
It should be appreciated that the executable Fig. 2 to Fig. 7 above of data processing equipment 1000 described in the embodiment of the present invention is any
Description in a corresponding embodiment to the data processing method, also can be performed in embodiment corresponding to Fig. 8 above to described
The description of data processing equipment 1, details are not described herein.In addition, being described to using the beneficial effect of same procedure, also no longer carry out
It repeats.
In addition, it need to be noted that: the embodiment of the invention also provides a kind of computer storage medium, and the meter
Computer program performed by the data processing equipment 1 being mentioned above, and the computer journey are stored in calculation machine storage medium
Sequence includes program instruction, and when the processor executes described program instruction, it is right to be able to carry out any one institute of Fig. 2 to Fig. 7 above
The description in embodiment to the data processing method is answered, therefore, will no longer be repeated here.In addition, to phase Tongfang is used
The beneficial effect of method describes, and is also no longer repeated.For not draped over one's shoulders in computer storage medium embodiment according to the present invention
The technical detail of dew please refers to the description of embodiment of the present invention method.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
The above disclosure is only the preferred embodiments of the present invention, cannot limit the right model of the present invention with this certainly
It encloses, therefore equivalent changes made in accordance with the claims of the present invention, is still within the scope of the present invention.
Claims (10)
1. a kind of data processing method characterized by comprising
Destination service text data is obtained, target text information is obtained from the destination service text data, and according to target
Target quality inspection points in quality inspection rule carry out abnormality detection the target text information;
If there are the target quality inspection points for meeting target quality inspection rule in the target text information, set in the target text
Set first identifier corresponding with the target quality inspection points;
The acquisition associated text information associated with the target text information from the destination service text data, and according to
Association quality inspection points in the target quality inspection rule carry out abnormality detection the associated text information;
If there are the association quality inspection points met in the target quality inspection rule in the associated text information, in the association text
Second identifier corresponding with the association quality inspection points is set in this information, and the first identifier, the second identifier will be carried
Destination service text data be determined as target exception text.
2. the method according to claim 1, wherein before the acquisition destination service text data, further includes:
Obtain at least one service voice data, and by each service voice data at least one described service voice data
It is respectively converted into service text data, and selects a service text data as mesh from the service text data being converted to
Mark service text data.
3. according to the method described in claim 2, it is characterized by further comprising:
Count the destination number for all service text datas being converted to;
If detecting, the destination number is greater than backup amount threshold, will be more than described standby in all service text datas
The service text data of part amount threshold carries out back-up processing.
4. according to the method described in claim 2, it is characterized in that, described will be every at least one described service voice data
A service voice data are respectively converted into service text data, comprising:
It selects a service voice data as target speech data from least one described service voice data, and obtains institute
State the speech spectral characteristics in target speech data;
Based on the speech spectral characteristics, the first identification model, Application on Voiceprint Recognition is carried out to the target speech data, is obtained described
The corresponding Application on Voiceprint Recognition result of first identification model;Include the speech spectral characteristics and described the in the Application on Voiceprint Recognition result
The matching degree between multiple attribute spectrum signatures in one identification model;
With the speech spectral characteristics will there is the attribute spectrum signature of highest matching degree to be closed according to the Application on Voiceprint Recognition result
The label information of connection, as the first role identified from the target speech data;
Remaining role in the target speech data in addition to the first role is determined as second role, based on control frequency
The target speech data for carrying the first role and second role is converted to service textual data by rate parameter, the second identification model
According to;
When each service voice data at least one described service voice data by as target speech data when, obtain
The corresponding service text data of each service voice data.
5. according to the method described in claim 4, it is characterized in that, described obtain target from the destination service text data
Text information, and the target text information is carried out abnormality detection according to the target quality inspection points in target quality inspection rule, comprising:
Corresponding first text data of the first role and second jiao described is obtained from the destination service text data
Corresponding second text data of color;
Obtain target quality inspection rule from multiple quality inspections rule, and by first text data and/or second textual data
According to as target text information, and from multiple quality inspection points that target quality inspection rule is included, selection has the matter of highest priority
It is cautious to be used as target quality inspection points, and the target text information is carried out abnormality detection based on the target quality inspection points.
6. the method according to claim 1, wherein acquisition and the institute from the destination service text data
The associated associated text information of target text information is stated, and according to the association quality inspection points in the target quality inspection rule to described
Associated text information carries out abnormality detection, comprising:
Using the corresponding target quality inspection points of the first identifier as starting checkpoint, and it is corresponding to obtain the target quality inspection rule
Detection range threshold value;
Text information between the starting checkpoint and the detection range threshold value is determined as and the target text information
Associated associated text information;
It obtains in target quality inspection rule and is associated with quality inspection points with incidence relation with the target quality inspection points, and according to the association
Quality inspection points carry out abnormality detection the associated text information.
7. the method according to claim 1, wherein further include:
The target exception text is added to abnormal text data set, and all targets that the abnormal text data is concentrated
Abnormal text is sent to quality inspection terminal;
Quality inspection terminal exception quality inspection points based on determined by all target exception texts are obtained, and based on the abnormal quality inspection
Point is updated the target quality inspection rule.
8. the method according to claim 1, wherein further include:
The target exception text is added to abnormal text data set, it is different to obtain the corresponding history of the abnormal text data set
Chang Wenben, and training sample set is obtained based on the history exception text;
The corresponding Checking model of the target quality inspection rule is trained based on the training sample set, and to the quality inspection mould
Type is updated;The abnormal quality inspection points carried in updated Checking model are for being updated the target quality inspection rule.
9. a kind of data processing equipment characterized by comprising
Text information obtains module and obtains mesh from the destination service text data for obtaining destination service text data
Text information is marked, and the target text information is carried out abnormality detection according to the target quality inspection points in target quality inspection rule;
First setup module, if for there are the target quality inspection points for meeting target quality inspection rule in the target text information,
First identifier corresponding with the target quality inspection points is set in the target text;
Related information obtains module, associated with the target text information for obtaining from the destination service text data
Associated text information, and according to the association quality inspection points in the target quality inspection rule associated text information is carried out abnormal
Detection;
Second setup module, if for there is the association quality inspection met in the target quality inspection rule in the associated text information
Second identifier corresponding with the association quality inspection points is then arranged in the associated text information, and will carry described first for point
It identifies, the destination service text data of the second identifier is determined as target exception text.
10. a kind of data processing equipment characterized by comprising processor and memory;
The processor is connected with memory, wherein the memory is for storing program code, and the processor is for calling
Said program code, to execute the method according to claim 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910150618.9A CN109902957B (en) | 2019-02-28 | 2019-02-28 | Data processing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910150618.9A CN109902957B (en) | 2019-02-28 | 2019-02-28 | Data processing method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109902957A true CN109902957A (en) | 2019-06-18 |
CN109902957B CN109902957B (en) | 2022-12-09 |
Family
ID=66945850
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910150618.9A Active CN109902957B (en) | 2019-02-28 | 2019-02-28 | Data processing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109902957B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110334639A (en) * | 2019-06-28 | 2019-10-15 | 北京精英系统科技有限公司 | A kind of device and method for the error detection result filtering analyzing and detecting algorithm |
CN110334241A (en) * | 2019-07-10 | 2019-10-15 | 深圳前海微众银行股份有限公司 | Quality detecting method, device, equipment and the computer readable storage medium of customer service recording |
CN110413635A (en) * | 2019-06-20 | 2019-11-05 | 口碑(上海)信息技术有限公司 | A kind of data processing method and device |
CN110929011A (en) * | 2019-11-28 | 2020-03-27 | 北京思特奇信息技术股份有限公司 | Conversation analysis method, device and equipment |
CN112069161A (en) * | 2020-09-01 | 2020-12-11 | 上海佰贝科技发展股份有限公司 | Data cleaning method and device |
WO2021031505A1 (en) * | 2019-08-22 | 2021-02-25 | 深圳追一科技有限公司 | Method and apparatus for detecting errors during audio annotation, computer device and storage medium |
CN112507008A (en) * | 2020-12-02 | 2021-03-16 | 贵州航天云网科技有限公司 | Data information resource sharing system and method based on dynamic interface service |
CN113032266A (en) * | 2021-03-30 | 2021-06-25 | 北京有竹居网络技术有限公司 | Code checking method, device, equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105187674A (en) * | 2015-08-14 | 2015-12-23 | 上海银天下科技有限公司 | Compliance checking method and device for service recorded sound |
WO2018109419A1 (en) * | 2016-12-15 | 2018-06-21 | Orange | Method for controlling a radio signal emitted by a gateway, and corresponding gateway and computer program |
CN108737667A (en) * | 2018-05-03 | 2018-11-02 | 平安科技(深圳)有限公司 | Voice quality detecting method, device, computer equipment and storage medium |
CN108962282A (en) * | 2018-06-19 | 2018-12-07 | 京北方信息技术股份有限公司 | Speech detection analysis method, apparatus, computer equipment and storage medium |
CN109243465A (en) * | 2018-12-06 | 2019-01-18 | 平安科技(深圳)有限公司 | Voiceprint authentication method, device, computer equipment and storage medium |
-
2019
- 2019-02-28 CN CN201910150618.9A patent/CN109902957B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105187674A (en) * | 2015-08-14 | 2015-12-23 | 上海银天下科技有限公司 | Compliance checking method and device for service recorded sound |
WO2018109419A1 (en) * | 2016-12-15 | 2018-06-21 | Orange | Method for controlling a radio signal emitted by a gateway, and corresponding gateway and computer program |
CN108737667A (en) * | 2018-05-03 | 2018-11-02 | 平安科技(深圳)有限公司 | Voice quality detecting method, device, computer equipment and storage medium |
CN108962282A (en) * | 2018-06-19 | 2018-12-07 | 京北方信息技术股份有限公司 | Speech detection analysis method, apparatus, computer equipment and storage medium |
CN109243465A (en) * | 2018-12-06 | 2019-01-18 | 平安科技(深圳)有限公司 | Voiceprint authentication method, device, computer equipment and storage medium |
Non-Patent Citations (1)
Title |
---|
钟迎红: "XX银行呼叫中心服务质量管理研究", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110413635A (en) * | 2019-06-20 | 2019-11-05 | 口碑(上海)信息技术有限公司 | A kind of data processing method and device |
CN110334639A (en) * | 2019-06-28 | 2019-10-15 | 北京精英系统科技有限公司 | A kind of device and method for the error detection result filtering analyzing and detecting algorithm |
CN110334639B (en) * | 2019-06-28 | 2021-08-10 | 北京精英系统科技有限公司 | Device and method for filtering error detection result of image analysis detection algorithm |
CN110334241A (en) * | 2019-07-10 | 2019-10-15 | 深圳前海微众银行股份有限公司 | Quality detecting method, device, equipment and the computer readable storage medium of customer service recording |
CN110334241B (en) * | 2019-07-10 | 2023-08-25 | 深圳前海微众银行股份有限公司 | Quality inspection method, device and equipment for customer service record and computer readable storage medium |
WO2021031505A1 (en) * | 2019-08-22 | 2021-02-25 | 深圳追一科技有限公司 | Method and apparatus for detecting errors during audio annotation, computer device and storage medium |
CN110929011A (en) * | 2019-11-28 | 2020-03-27 | 北京思特奇信息技术股份有限公司 | Conversation analysis method, device and equipment |
CN112069161A (en) * | 2020-09-01 | 2020-12-11 | 上海佰贝科技发展股份有限公司 | Data cleaning method and device |
CN112069161B (en) * | 2020-09-01 | 2023-11-03 | 上海佰贝科技发展股份有限公司 | Data cleaning method and device |
CN112507008A (en) * | 2020-12-02 | 2021-03-16 | 贵州航天云网科技有限公司 | Data information resource sharing system and method based on dynamic interface service |
CN112507008B (en) * | 2020-12-02 | 2022-06-14 | 贵州航天云网科技有限公司 | Data information resource sharing system and method based on dynamic interface service |
CN113032266A (en) * | 2021-03-30 | 2021-06-25 | 北京有竹居网络技术有限公司 | Code checking method, device, equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109902957B (en) | 2022-12-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109902957A (en) | A kind of data processing method and device | |
CN110266899B (en) | Client intention identification method and customer service system | |
Chen et al. | Codah: An adversarially authored question-answer dataset for common sense | |
US10776580B2 (en) | Method for providing dialogue service with chatbot assisted by human agents | |
US10643632B2 (en) | Automated voice assistant personality selector | |
CN105408891B (en) | System and method for the multilingual communication of multi-user | |
CN108428446A (en) | Audio recognition method and device | |
CN109151218A (en) | Call voice quality detecting method, device, computer equipment and storage medium | |
CN111241357A (en) | Dialogue training method, device, system and storage medium | |
CN113360622B (en) | User dialogue information processing method and device and computer equipment | |
CN111179935B (en) | Voice quality inspection method and device | |
CN109086276B (en) | Data translation method, device, terminal and storage medium | |
CN109003600B (en) | Message processing method and device | |
CN110489747A (en) | A kind of image processing method, device, storage medium and electronic equipment | |
US20230394247A1 (en) | Human-machine collaborative conversation interaction system and method | |
CN111259124A (en) | Dialogue management method, device, system and storage medium | |
CN110428816A (en) | A kind of method and device voice cell bank training and shared | |
CN113591463A (en) | Intention recognition method and device, electronic equipment and storage medium | |
US20220129628A1 (en) | Artificial intelligence system for business processes | |
CN111933107A (en) | Speech recognition method, speech recognition device, storage medium and processor | |
CN116883888A (en) | Bank counter service problem tracing system and method based on multi-mode feature fusion | |
CN116303951A (en) | Dialogue processing method, device, electronic equipment and storage medium | |
CN113505606B (en) | Training information acquisition method and device, electronic equipment and storage medium | |
CN110765242A (en) | Method, device and system for providing customer service information | |
CN116127011A (en) | Intention recognition method, device, electronic equipment and storage medium |
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 |